Neoplasias Malignas de Cabeza y Cuello en Chile: Análisis de Mortalidad Hospitalaria y Duración de Estadía (GRD 2019–2024)
This is a retrospective cohort study based on hospital discharge records from the GRD (Diagnosis-Related Groups) system in Chile, covering the period 2019–2024. The GRD system captures comprehensive data on public hospital discharges including:
Cases of head and neck malignancies (ICD-10 codes C00–C14, C30–C32) are identified by screening all 35 diagnostic fields for these codes, with the cancer site determined by the first (earliest) occurrence position.
All patients with an ICD-10 code between C00–C14 and C30–C32 in any of the 35 diagnostic fields (DIAGNOSTICO1–DIAGNOSTICO35) are included. The primary cancer site is classified based on the first diagnostic field containing a relevant code:
Oral Cavity (C00–C06) - Lip (C00): ICD-10 code C00 - Base of tongue (C01): ICD-10 code C01 - Other tongue (C02): ICD-10 code C02 - Gum (C03): ICD-10 code C03 - Floor of mouth (C04): ICD-10 code C04 - Palate (C05): ICD-10 code C05 - Other mouth (C06): ICD-10 code C06
Salivary Glands (C07–C08) - Parotid gland (C07): ICD-10 code C07 - Other salivary glands (C08): ICD-10 code C08
Pharynx (C09–C14) - Tonsil (C09): ICD-10 code C09 - Oropharynx (C10): ICD-10 code C10 - Nasopharynx (C11): ICD-10 code C11 - Piriform sinus (C12): ICD-10 code C12 - Hypopharynx (C13): ICD-10 code C13 - Other lip, oral, pharynx (C14): ICD-10 code C14
Sinonasal and Larynx (C30–C32) - Nasal cavity and middle ear (C30): ICD-10 code C30 - Accessory sinuses (C31): ICD-10 code C31 - Larynx (C32): ICD-10 code C32
For analysis, sites are grouped into clinical subsites: Oral Cavity (C00–C06), Salivary Glands (C07–C08), Oropharynx (C09–C10), Nasopharynx (C11), Hypopharynx (C12–C13), Other Pharynx (C14), Sinonasal (C30–C31), and Larynx (C32).
Comorbidities are identified from secondary diagnoses (DIAGNOSTICO2 through DIAGNOSTICO35) and classified into 48 detailed clinical groups organized in 10 systemic categories, using the same scheme as the comorbidity network analysis (neoplasias_cabeza_cuello_redes.qmd) to ensure consistency across analyses.
| Systemic Category | Detailed Groups | Representative ICD-10 Codes |
|---|---|---|
| Cardiovascular (CV) — 7 groups | Essential Hypertension; Complicated Hypertension; Coronary Disease; Heart Failure; Arrhythmias; Cerebrovascular Disease; Peripheral Vascular Disease | I10; I11–I15; I20–I25; I50; I44–I49; I60–I69; I70–I77 |
| Metabolic (MET) — 7 groups | Type 1 Diabetes; Type 2 Diabetes; Complicated Diabetes; Obesity; Dyslipidemia; Metabolic Syndrome; Thyroid Disorders | E10; E11; E12–E14; E66; E78; E88; E00–E07 |
| Respiratory (RESP) — 7 groups | COPD; Asthma; Chronic Bronchitis; Emphysema; Pulmonary Fibrosis; Pulmonary Hypertension; Sleep Apnea | J44; J45–J46; J41–J42; J43; J84; I27; G47 |
| Renal (REN) — 4 groups | Acute Renal Failure; Chronic Renal Failure; End Stage Renal; Diabetic Nephropathy | N17; N18; N18.5–N18.6; E10.2–E14.2 |
| Hepatic (HEP) — 4 groups | Steatosis; Alcoholic Cirrhosis; NonAlcoholic Cirrhosis; Viral Hepatitis | K76; K70; K74; B15–B19 |
| Gastrointestinal (GI) — 4 groups | Reflux; Peptic Ulcer; IBD; IBS | K21; K25–K28; K50–K51; K58 |
| Neurological (NEURO) — 4 groups | Dementia; Parkinson; Epilepsy; Stroke Sequelae | F00–F03, G30; G20–G22; G40–G41; I69 |
| Psychiatric (PSI) — 3 groups | Major Depression; Bipolar; Anxiety | F32–F33; F30–F31; F41 |
| Substance Use (SUST) — 3 groups | Alcohol; Tobacco; Opioids | F10; F17, Z72; F11 |
| Infectious (INF) — 4 groups | HIV; Tuberculosis; Hepatitis C; Hepatitis B | B20–B24; A15–A19; B18; B16 |
Each of the 48 detailed groups is coded as a binary indicator (presence = 1, absence = 0). Additionally, 10 aggregated systemic-category indicators (cat_CV, cat_MET, cat_RESP, cat_REN, cat_HEP, cat_GI, cat_NEURO, cat_PSI, cat_SUST, cat_INF) are derived as the disjunction (any group present) within each category. Code matching uses both 3-character ICD-10 prefixes and 4-character extended codes (e.g., N18.5 for End Stage Renal disease).
Continuous variables (age, length of stay) are reported as mean, median, standard deviation, and range. Categorical variables (sex, service category, comorbidities, discharge status) are reported as counts and proportions. Mortality rates and length of stay distributions are stratified by cancer site and temporal trends.
Independence between each categorical covariate and the binary mortality outcome was tested with the Pearson chi-square statistic
\[ \chi^{2} \;=\; \sum_{i=1}^{R}\sum_{j=1}^{C} \frac{(O_{ij} - E_{ij})^{2}}{E_{ij}}, \]
where:
The Wilcoxon rank-sum (Mann–Whitney \(U\)) test compared the length-of-stay distribution between patients with and without each comorbidity, without assuming normality or homoscedasticity.
In-hospital mortality \(Y_{ij} \in \{0,1\}\) (death = 1) is modeled with a mixed-effects logistic regression with a random intercept for the cluster (year of admission \(j\)):
\[ Y_{ij} \,\mid\, u_j \;\sim\; \text{Bernoulli}(\pi_{ij}), \qquad \mathrm{logit}(\pi_{ij}) \;=\; \log\!\left(\frac{\pi_{ij}}{1-\pi_{ij}}\right) \;=\; \beta_0 \,+\, \mathbf{x}_{ij}^{\top}\boldsymbol{\beta} \,+\, u_j, \]
with \(u_j \sim \mathcal{N}(0,\sigma^{2}_{u})\) independent across years. Symbols:
Models were fit by maximum likelihood with Laplace approximation, separately for each of the eight primary cancer subsites. Estimates are reported as odds ratios with 95% Wald confidence intervals.
Length of stay is a non-negative count with overdispersion relative to Poisson. We use a negative-binomial mixed model with a log link and a random intercept for year:
\[ Y_{ij} \,\mid\, u_j \;\sim\; \mathrm{NB}\!\left(\mu_{ij}, \theta\right), \qquad \log(\mu_{ij}) \;=\; \beta_0 \,+\, \mathbf{x}_{ij}^{\top}\boldsymbol{\beta} \,+\, u_j, \]
with \(u_j \sim \mathcal{N}(0,\sigma^{2}_{u})\) and the NB\((\mu,\theta)\) variance
\[ \mathrm{Var}(Y_{ij}\mid u_j) \;=\; \mu_{ij} \,+\, \mu_{ij}^{2}/\theta. \]
Symbols:
Coefficients are reported as incidence rate ratios \(\mathrm{IRR}_k = e^{\beta_k}\) (multiplicative effect on expected stay) with 95% CIs.
Discrimination of the mortality model is summarized by the area under the ROC curve,
\[ \mathrm{AUC} \;=\; \int_{0}^{1} \mathrm{TPR}\!\left(\mathrm{FPR}^{-1}(u)\right)\,du \;=\; \Pr\!\big(\hat{p}_{i} > \hat{p}_{j} \,\big|\, y_i = 1,\, y_j = 0\big), \]
where for every classification threshold \(c \in [0,1]\) on the predicted probability \(\hat{p}\):
Calibration of the logistic model was assessed with the Hosmer–Lemeshow statistic, which compares observed and expected events across \(G\) deciles of predicted risk:
\[ \mathrm{HL} \;=\; \sum_{g=1}^{G} \frac{\big(O_g - n_g\bar{p}_g\big)^{2}}{n_g\,\bar{p}_g\,(1 - \bar{p}_g)}, \]
where:
Under \(H_0\) (model well-calibrated), \(\mathrm{HL} \;\dot\sim\; \chi^{2}_{G-2}\); \(p > 0.05\) supports adequate calibration.
\[ R^{2}_{\mathrm{McF}} \;=\; 1 \,-\, \frac{\ell(\hat{\boldsymbol{\beta}})}{\ell(\mathbf{0})}, \]
where:
Both log-likelihoods are negative, so \(R^{2}_{\mathrm{McF}} \in [0,1]\); values of 0.2–0.4 are typically considered indicative of good fit in logistic models.
For a binary classifier at a fixed threshold, the Matthews correlation coefficient is
\[ \mathrm{MCC} \;=\; \frac{TP\cdot TN \;-\; FP\cdot FN}{\sqrt{(TP+FP)(TP+FN)(TN+FP)(TN+FN)}}, \]
where \(TP, TN, FP, FN\) are the cells of the confusion matrix (true/false positives/negatives). \(\mathrm{MCC} \in [-1, 1]\): \(+1\) perfect agreement, \(0\) no better than chance, \(-1\) perfect disagreement. Unlike accuracy, MCC stays meaningful with class imbalance.
The Brier score is the mean squared error between predicted probabilities and observed binary outcomes,
\[ \mathrm{BS} \;=\; \frac{1}{N}\sum_{i=1}^{N}\big(y_i - \hat{p}_i\big)^{2}, \]
where:
\(\mathrm{BS} = 0\) is perfect; \(\mathrm{BS} = 0.25\) is the score of a non-informative model that predicts \(\hat{p}_i \equiv 0.5\) for everyone. It rewards both calibration and discrimination.
Predictive accuracy of the continuous length-of-stay model is summarized by
\[ \mathrm{RMSE} \;=\; \sqrt{\frac{1}{N}\sum_{i=1}^{N}\big(\hat{y}_i - y_i\big)^{2}}, \qquad \mathrm{MAE} \;=\; \frac{1}{N}\sum_{i=1}^{N}\big|\hat{y}_i - y_i\big|, \]
where \(y_i\) is the observed length of stay (days), \(\hat{y}_i\) its prediction, and \(N\) the validation sample size. RMSE penalizes large errors more strongly because of the square; MAE is in the same units as the outcome and is more robust to outliers.
For threshold selection on the ROC curve we use Youden’s \(J\):
\[ J(c) \;=\; \mathrm{TPR}(c) \,+\, \mathrm{TNR}(c) \,-\, 1 \;=\; \mathrm{Sensitivity}(c) \,+\, \mathrm{Specificity}(c) \,-\, 1, \]
where \(c\) is the probability cut-off. \(J \in [0,1]\); the optimal cut-off is \(c^{*} = \arg\max_c J(c)\), the point that maximizes the vertical distance between the ROC curve and the chance diagonal.
import pandas as pd
import numpy as np
import os
import warnings
from datetime import datetime
from pathlib import Path
import pickle
from IPython.display import display
warnings.filterwarnings('ignore')
# Configuration
DATA_DIR = "../data"
OUTPUT_DIR = "../data/output_files"
CACHE_DIR = "../data"
# Create output directory if needed
os.makedirs(OUTPUT_DIR, exist_ok=True)
os.makedirs(CACHE_DIR, exist_ok=True)# Function to load GRD data
def load_grd_data(data_dir, years=[2019, 2020, 2021, 2022, 2023, 2024]):
"""Load GRD data from multiple CSV files (pipe-separated, UTF-8)"""
dfs = []
for year in years:
filepath = os.path.join(data_dir, f"GRD_PUBLICO_{year}.csv")
if not os.path.exists(filepath):
continue
try:
df = pd.read_csv(filepath, sep="|", dtype=str, low_memory=False, encoding='utf-8')
df['year'] = str(year)
dfs.append(df)
except Exception as e:
pass
if dfs:
return pd.concat(dfs, ignore_index=True)
else:
return pd.DataFrame()
# Load all GRD data
df_grd = load_grd_data(DATA_DIR)
if len(df_grd) == 0:
display(HTML("<p><strong>Warning:</strong> No GRD data found.</p>"))# Load catalogs
catalog_file = os.path.join(DATA_DIR, "códigos grd.xlsx")
df_grd_codes = None
df_hospitals = None
try:
df_grd_codes = pd.read_excel(catalog_file, sheet_name="IR - GRD")
except Exception as e:
pass
try:
df_hospitals = pd.read_excel(catalog_file, sheet_name="Hospitales")
except Exception as e:
pass# Function to parse dates flexibly
def parse_date_flexible(date_str):
"""Parse dates that may be in DD-MM-YYYY or YYYY-MM-DD format"""
if pd.isna(date_str) or date_str == '':
return pd.NaT
if isinstance(date_str, (int, float)):
return pd.NaT
date_str = str(date_str).strip()
# Try different formats
for fmt in ["%d-%m-%Y", "%Y-%m-%d", "%d/%m/%Y", "%Y/%m/%d"]:
try:
return pd.to_datetime(date_str, format=fmt)
except:
continue
return pd.NaT
# Apply date normalization
date_cols = ['FECHA_NACIMIENTO', 'FECHA_INGRESO', 'FECHAALTA']
for col in date_cols:
if col in df_grd.columns:
df_grd[col] = df_grd[col].apply(parse_date_flexible)
valid_count = df_grd[col].notna().sum()# Calculate derived variables
# Age in years
df_grd['edad'] = (df_grd['FECHA_INGRESO'] - df_grd['FECHA_NACIMIENTO']).dt.days / 365.25
# Length of stay in days
df_grd['dias_estancia'] = (df_grd['FECHAALTA'] - df_grd['FECHA_INGRESO']).dt.days
# Mortality status (discharge type)
df_grd['egresar_fallecido'] = (df_grd['TIPOALTA'] == 'FALLECIDO').astype(int)
df_grd['estado_alta'] = df_grd['TIPOALTA'].apply(
lambda x: "Deceased" if x == "FALLECIDO" else "Discharged"
)# Filter valid records
# Remove unknown sex
df_clean = df_grd[df_grd['SEXO'] != 'DESCONOCIDO'].copy()
# Standardize sex values to English
sex_map = {'FEMENINO': 'Female', 'MASCULINO': 'Male', 'MUJER': 'Female', 'HOMBRE': 'Male',
'Female': 'Female', 'Male': 'Male', 'F': 'Female', 'M': 'Male'}
df_clean['SEXO'] = df_clean['SEXO'].map(sex_map).fillna(df_clean['SEXO'])
# Remove invalid ages
df_clean = df_clean[(df_clean['edad'].notna()) & (df_clean['edad'] > 0) & (df_clean['edad'] < 150)].copy()
# Remove invalid length of stay
df_clean = df_clean[(df_clean['dias_estancia'].notna()) & (df_clean['dias_estancia'] >= 0)].copy()# Function to identify head and neck cancer sites from diagnostic fields
def determine_cancer_site(row):
"""
Scans all DIAGNOSTICO1-35 fields and returns the first C00-C14, C30-C32 code found.
Returns: (cancer_site_name, cancer_code, diagnostic_field_position)
"""
cancer_sites = {
'C00': 'Lip',
'C01': 'Base of tongue',
'C02': 'Other tongue',
'C03': 'Gum',
'C04': 'Floor of mouth',
'C05': 'Palate',
'C06': 'Other mouth',
'C07': 'Parotid gland',
'C08': 'Other salivary glands',
'C09': 'Tonsil',
'C10': 'Oropharynx',
'C11': 'Nasopharynx',
'C12': 'Piriform sinus',
'C13': 'Hypopharynx',
'C14': 'Other pharynx',
'C30': 'Nasal cavity',
'C31': 'Accessory sinuses',
'C32': 'Larynx'
}
for pos in range(1, 36):
col_name = f'DIAGNOSTICO{pos}'
if col_name not in row.index:
continue
diag = row[col_name]
if pd.isna(diag) or diag == '':
continue
diag_str = str(diag).strip().upper()
if len(diag_str) < 3:
continue
icd_code = diag_str[:3]
if icd_code in cancer_sites:
return cancer_sites[icd_code], icd_code, pos
return None, None, None
# Filter for head and neck neoplasms
cancer_results = df_clean.apply(determine_cancer_site, axis=1)
df_clean['cancer_site'] = cancer_results.apply(lambda x: x[0])
df_clean['cancer_code'] = cancer_results.apply(lambda x: x[1])
df_clean['diagnostic_position'] = cancer_results.apply(lambda x: x[2])
# Keep only records with cancer diagnosis
df_head_neck = df_clean[df_clean['cancer_site'].notna()].copy()
vc_site = df_head_neck['cancer_site'].value_counts().sort_values(ascending=False).reset_index()
vc_site.columns = ['Cancer Site', 'N']
vc_site['%'] = (vc_site['N'] / vc_site['N'].sum() * 100).round(1)# Create clinical subsite grouping
def classify_head_neck_subsite(site):
"""Group head and neck sites into clinical subsites"""
oral_cavity = ['Lip', 'Base of tongue', 'Other tongue', 'Gum', 'Floor of mouth', 'Palate', 'Other mouth']
salivary = ['Parotid gland', 'Other salivary glands']
oropharynx_group = ['Tonsil', 'Oropharynx']
hypopharynx_group = ['Piriform sinus', 'Hypopharynx']
sinonasal = ['Nasal cavity', 'Accessory sinuses']
if site in oral_cavity:
return 'Oral Cavity'
elif site in salivary:
return 'Salivary Glands'
elif site in oropharynx_group:
return 'Oropharynx'
elif site == 'Nasopharynx':
return 'Nasopharynx'
elif site in hypopharynx_group:
return 'Hypopharynx'
elif site == 'Other pharynx':
return 'Other Pharynx'
elif site in sinonasal:
return 'Sinonasal'
elif site == 'Larynx':
return 'Larynx'
else:
return site
df_head_neck['cancer_site_group'] = df_head_neck['cancer_site'].apply(classify_head_neck_subsite)
vc = df_head_neck['cancer_site_group'].value_counts().reset_index()
vc.columns = ['Cancer Site', 'N']
vc['%'] = (vc['N'] / vc['N'].sum() * 100).round(1)# Classify service category
def classify_service_category(servicio):
"""Classify admission service into categories"""
if pd.isna(servicio) or servicio == '':
return 'Other'
servicio = str(servicio).upper()
if 'CUIDADOS INTENSIVOS' in servicio or 'CUIDADOS' in servicio:
return 'ICU'
elif 'TRATAMIENTO INTERMEDIO' in servicio or 'UTI' in servicio:
return 'ITU'
elif any(x in servicio for x in ['AGUDOS', 'CIRUGÍA', 'CIRUGÍA', 'RECUPERACIÓN', 'MEDICO-QUIRURGICO']):
return 'MS'
else:
return 'Other'
df_head_neck['service_category'] = df_head_neck['SERVICIOINGRESO'].apply(classify_service_category)
sc = df_head_neck['service_category'].value_counts().reset_index()
sc.columns = ['Service Category', 'N']
sc['%'] = (sc['N'] / sc['N'].sum() * 100).round(1)# Detect comorbidities from secondary diagnoses
# Uses the same comprehensive 48-group / 10-category scheme as the network analysis (redes)
comorbidity_groups = {
# CARDIOVASCULAR (7 groups)
"CV Essential Hypertension": ["I10"],
"CV Complicated Hypertension": ["I11", "I12", "I13", "I15"],
"CV Coronary Disease": ["I20", "I21", "I22", "I23", "I24", "I25"],
"CV Heart Failure": ["I50"],
"CV Arrhythmias": ["I44", "I45", "I46", "I47", "I48", "I49"],
"CV Cerebrovascular Disease": ["I60", "I61", "I62", "I63", "I64", "I65", "I66", "I67", "I68", "I69"],
"CV Peripheral Vascular Disease": ["I70", "I71", "I73", "I74", "I77"],
# METABOLIC (7 groups)
"MET Type1 Diabetes": ["E10"],
"MET Type2 Diabetes": ["E11"],
"MET Complicated Diabetes": ["E12", "E13", "E14"],
"MET Obesity": ["E66"],
"MET Dyslipidemia": ["E78"],
"MET Metabolic Syndrome": ["E88"],
"MET Thyroid Disorders": ["E00", "E01", "E02", "E03", "E04", "E05", "E06", "E07"],
# RESPIRATORY (7 groups)
"RESP COPD": ["J44"],
"RESP Asthma": ["J45", "J46"],
"RESP Chronic Bronchitis": ["J41", "J42"],
"RESP Emphysema": ["J43"],
"RESP Pulmonary Fibrosis": ["J84"],
"RESP Pulmonary Hypertension": ["I27"],
"RESP Sleep Apnea": ["G47"],
# RENAL (4 groups)
"REN Acute Renal Failure": ["N17"],
"REN Chronic Renal Failure": ["N18"],
"REN End Stage Renal": ["N185", "N186"],
"REN Diabetic Nephropathy": ["E102", "E112", "E122", "E132", "E142"],
# HEPATIC (4 groups)
"HEP Steatosis": ["K76"],
"HEP Alcoholic Cirrhosis": ["K70"],
"HEP NonAlcoholic Cirrhosis": ["K74"],
"HEP Viral Hepatitis": ["B15", "B16", "B17", "B18", "B19"],
# GASTROINTESTINAL (4 groups)
"GI Reflux": ["K21"],
"GI Peptic Ulcer": ["K25", "K26", "K27", "K28"],
"GI IBD": ["K50", "K51"],
"GI IBS": ["K58"],
# NEUROLOGICAL (4 groups)
"NEURO Dementia": ["F00", "F01", "F02", "F03", "G30"],
"NEURO Parkinson": ["G20", "G21", "G22"],
"NEURO Epilepsy": ["G40", "G41"],
"NEURO Stroke Sequelae": ["I69"],
# PSYCHIATRIC (3 groups)
"PSI Major Depression": ["F32", "F33"],
"PSI Bipolar": ["F30", "F31"],
"PSI Anxiety": ["F41"],
# SUBSTANCE (3 groups)
"SUST Alcohol": ["F10"],
"SUST Tobacco": ["F17", "Z72"],
"SUST Opioids": ["F11"],
# INFECTIOUS (4 groups)
"INF HIV": ["B20", "B21", "B22", "B23", "B24"],
"INF Tuberculosis": ["A15", "A16", "A17", "A18", "A19"],
"INF Hepatitis C": ["B18"],
"INF Hepatitis B": ["B16"],
}
# Backwards-compatible alias used downstream
life_groups = comorbidity_groups
def get_icd_code(diag):
"""Extract normalized ICD code, returning both 3-char and 4-char (no dot) forms."""
if pd.isna(diag) or diag == '':
return None
s = str(diag).strip().upper().replace('.', '')
return s
def check_comorbidity_groups(row):
"""Detect comorbidities for a record. Matches both 3-char and 4-char codes
so groups like REN End Stage Renal (N185, N186) are correctly captured."""
diag_cols = [f'DIAGNOSTICO{i}' for i in range(2, 36)]
codes_full = set()
codes_3 = set()
for col in diag_cols:
if col in row.index:
icd = get_icd_code(row[col])
if icd:
codes_full.add(icd)
codes_3.add(icd[:3])
comorbidities = {}
for group_name, icd_codes in comorbidity_groups.items():
match = False
for code in icd_codes:
if len(code) == 3:
if code in codes_3:
match = True
break
else:
if any(c.startswith(code) for c in codes_full):
match = True
break
comorbidities[group_name] = 1 if match else 0
return comorbidities
# Reset index to ensure clean alignment after the boolean filter above
df_head_neck = df_head_neck.reset_index(drop=True)
# Apply comorbidity detection
comorbidity_data = df_head_neck.apply(check_comorbidity_groups, axis=1)
comorbidity_df = pd.DataFrame(comorbidity_data.tolist(), index=df_head_neck.index)
# Add to main dataframe (index-aligned)
for col in comorbidity_df.columns:
df_head_neck[col] = comorbidity_df[col].astype(int)
comorbidity_cols = list(comorbidity_groups.keys())
all_comorbidity_cols = list(comorbidity_groups.keys())
# Aggregated systemic category indicators (CV, MET, RESP, REN, HEP, GI, NEURO, PSI, SUST, INF)
systemic_categories = ['CV', 'MET', 'RESP', 'REN', 'HEP', 'GI', 'NEURO', 'PSI', 'SUST', 'INF']
for cat in systemic_categories:
cat_cols = [c for c in comorbidity_cols if c.startswith(cat + ' ')]
df_head_neck[f'cat_{cat}'] = (df_head_neck[cat_cols].sum(axis=1) > 0).astype(int)
# Create total comorbidities and any comorbidity columns
df_head_neck['total_comorbidities'] = df_head_neck[comorbidity_cols].sum(axis=1)
df_head_neck['any_comorbidity'] = (df_head_neck['total_comorbidities'] > 0).astype(int)
# Prepare clustering/random effect variables for mixed models
# COD_HOSPITAL as cluster (random intercept), year as fixed effect
if 'COD_HOSPITAL' in df_head_neck.columns:
df_head_neck['hospital_id'] = df_head_neck['COD_HOSPITAL'].astype(str)
else:
# Fallback: use first available hospital-like column
for hcol in ['ESTABLECIMIENTO', 'HOSPITAL', 'cod_hospital']:
if hcol in df_head_neck.columns:
df_head_neck['hospital_id'] = df_head_neck[hcol].astype(str)
break
else:
df_head_neck['hospital_id'] = '1' # single cluster fallback
df_head_neck['year_factor'] = df_head_neck['year'].astype(str)from IPython.display import display, HTML
# Helper function for clean table display
def _safe_format(x):
"""Format a value: NaN → '—', Inf → 'Inf', large abs → scientific, else 2 decimals"""
if isinstance(x, float):
if np.isnan(x):
return '—'
if np.isinf(x):
return 'Inf' if x > 0 else '-Inf'
if abs(x) > 1e8:
return f'{x:.2e}'
return f'{x:,.2f}'
return str(x)
def sanitize_df(df):
"""Replace NaN with '—' and Inf with 'Inf' across entire DataFrame"""
df_out = df.copy()
for col in df_out.columns:
if df_out[col].dtype in ['float64', 'float32']:
df_out[col] = df_out[col].apply(_safe_format)
else:
df_out[col] = df_out[col].fillna('—')
# Also handle index if it contains NaN
if df_out.index.dtype in ['float64', 'float32']:
df_out.index = [_safe_format(v) for v in df_out.index]
return df_out
def show_table(df, caption="", index=False, float_format=None):
"""Display a DataFrame as a properly formatted HTML table with NaN/Inf handling"""
df_clean = sanitize_df(df)
html = f'<h4>{caption}</h4>' if caption else ''
html += df_clean.to_html(index=index, classes='table table-striped table-hover table-sm',
escape=False)
display(HTML(html))
# Process severity and mortality risk as factors
severity_cols = ['IR_29301_SEVERIDAD', 'IR_29301_MORTALIDAD']
for col in severity_cols:
if col in df_head_neck.columns:
df_head_neck[col] = pd.to_numeric(df_head_neck[col], errors='coerce')
df_head_neck[col] = df_head_neck[col].fillna(df_head_neck[col].median())
df_head_neck[col] = df_head_neck[col].astype('category')
# Create main sites dataset
main_sites = ['Larynx', 'Oropharynx', 'Oral Cavity', 'Nasopharynx']
df_main_sites = df_head_neck[df_head_neck['cancer_site_group'].isin(main_sites)].copy()
# Save processed dataframe to cache
output_cache_path = os.path.join(CACHE_DIR, "df_head_neck_processed.pkl")
os.makedirs(os.path.dirname(output_cache_path), exist_ok=True)
with open(output_cache_path, 'wb') as f:
pickle.dump(df_head_neck, f)
# General summary
info_df = pd.DataFrame({
'Metric': ['Total GRD records', 'Head and neck malignancy cases (C00-C14, C30-C32)',
'Period', 'Mortality rate', 'Median length of stay'],
'Value': [f'{len(df_head_neck):,}', f'{len(df_head_neck):,}',
f'{df_head_neck["year"].min()}-{df_head_neck["year"].max()}',
f'{df_head_neck["egresar_fallecido"].mean()*100:.1f}%',
f'{df_head_neck["dias_estancia"].median():.0f} days']
})
show_table(info_df, "Table 5: Overall Summary Statistics", index=False)| Metric | Value |
|---|---|
| Total GRD records | 9,411 |
| Head and neck malignancy cases (C00-C14, C30-C32) | 9,411 |
| Period | 2019-2024 |
| Mortality rate | 6.4% |
| Median length of stay | 5 days |
# Define formatted display names for all variables
DISPLAY_NAMES = {
'edad': 'Age',
'SEXO_coded': 'Sex (Female)',
'dias_estancia': 'Length of Stay',
'egresar_fallecido': 'In-Hospital Mortality',
'cancer_site_group': 'Cancer Site',
'service_category': 'Service Category',
'service_category_coded': 'Service Category',
'year': 'Year',
'year_factor': 'Year',
'IR_29301_SEVERIDAD': 'Severity',
'IR_29301_MORTALIDAD': 'Mortality Risk',
# Detailed comorbidity groups (48 groups, 10 systemic categories)
**{name: name for name in comorbidity_cols},
# Aggregated systemic categories
'cat_CV': 'Cardiovascular (any)',
'cat_MET': 'Metabolic (any)',
'cat_RESP': 'Respiratory (any)',
'cat_REN': 'Renal (any)',
'cat_HEP': 'Hepatic (any)',
'cat_GI': 'Gastrointestinal (any)',
'cat_NEURO': 'Neurological (any)',
'cat_PSI': 'Psychiatric (any)',
'cat_SUST': 'Substance Use (any)',
'cat_INF': 'Infectious (any)',
'total_comorbidities': 'Total Comorbidities',
'any_comorbidity': 'Any Comorbidity',
'service_ICU': 'Service ICU',
'service_MS': 'Service MS',
'service_ITU': 'Service ITU',
'service_Other': 'Service Other',
'severity_2': 'Severity Medium',
'severity_3': 'Severity High',
'mort_risk_2': 'Mortality Risk Medium',
'mort_risk_3': 'Mortality Risk High',
'const': 'Intercept',
'Intercept': 'Intercept',
'sex_female': 'Sex Female',
}
def format_name(name):
"""Convert variable names to formatted display names"""
if name in DISPLAY_NAMES:
return DISPLAY_NAMES[name]
if name.startswith('year_'):
return f'Year {name.split("_")[1]}'
if name.startswith('severity_'):
level = name.split('_')[1]
labels = {'2': 'Medium', '3': 'High'}
return f'Severity {labels.get(level, level)}'
if name.startswith('mort_risk_'):
level = name.split('_')[2] if len(name.split('_')) > 2 else name.split('_')[1]
labels = {'2': 'Medium', '3': 'High'}
return f'Mortality Risk {labels.get(level, level)}'
if name.startswith('service_'):
svc = name.replace('service_', '')
return f'Service {svc}'
if name.startswith('C('):
return name.replace('C(', '').replace(')', '').replace('[T.', ': ').replace(']', '')
return name.replace('_', ' ').title()
# Create mapping for comorbidity display names
comorb_names = {col: format_name(col) for col in comorbidity_cols}import matplotlib.pyplot as plt
import seaborn as sns
plt.rcParams['figure.facecolor'] = 'white'
sns.set_style("whitegrid")
# Create age groups
df_head_neck['age_group'] = pd.cut(df_head_neck['edad'],
bins=[0, 18, 35, 50, 65, 80, 150],
labels=['<18', '18-34', '35-49', '50-64', '65-79', '80+'])
# Table 1: Patient Distribution by Cancer Site
table1_data = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
site_df = df_head_neck[df_head_neck['cancer_site_group'] == site]
n_total = len(site_df)
n_discharged = (site_df['egresar_fallecido'] == 0).sum()
n_deceased = site_df['egresar_fallecido'].sum()
mortality_pct = (n_deceased / n_total * 100) if n_total > 0 else 0
mean_age = site_df['edad'].mean()
age_range = f"{site_df['edad'].min():.0f}-{site_df['edad'].max():.0f}"
female_pct = (site_df['SEXO'] == 'Female').sum() / n_total * 100
mean_los = site_df['dias_estancia'].mean()
median_los = site_df['dias_estancia'].median()
sd_los = site_df['dias_estancia'].std()
table1_data.append({
'Site': site,
'N': n_total,
'Discharged': n_discharged,
'Deceased': n_deceased,
'Mortality %': mortality_pct,
'Mean Age': mean_age,
'Age Range': age_range,
'Female %': female_pct,
'Mean LOS': mean_los,
'Median LOS': median_los,
'SD LOS': sd_los
})
table1_df = pd.DataFrame(table1_data)
show_table(table1_df, "Table 1: Patient Distribution by Head and Neck Cancer Site", index=False)| Site | N | Discharged | Deceased | Mortality % | Mean Age | Age Range | Female % | Mean LOS | Median LOS | SD LOS |
|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 400 | 367 | 33 | 8.25 | 68.60 | 9-97 | 23.50 | 13.71 | 7.00 | 19.00 |
| Larynx | 2625 | 2427 | 198 | 7.54 | 67.63 | 6-98 | 15.05 | 14.49 | 7.00 | 21.86 |
| Nasopharynx | 368 | 345 | 23 | 6.25 | 50.19 | 3-92 | 31.79 | 8.97 | 4.00 | 18.19 |
| Oral Cavity | 3001 | 2825 | 176 | 5.86 | 64.32 | 0-99 | 39.09 | 9.61 | 4.00 | 16.41 |
| Oropharynx | 951 | 896 | 55 | 5.78 | 62.16 | 8-96 | 26.60 | 8.39 | 4.00 | 13.66 |
| Other Pharynx | 281 | 251 | 30 | 10.68 | 57.49 | 0-96 | 32.38 | 10.80 | 5.00 | 15.99 |
| Salivary Glands | 898 | 852 | 46 | 5.12 | 60.54 | 2-97 | 52.00 | 6.95 | 4.00 | 11.65 |
| Sinonasal | 887 | 846 | 41 | 4.62 | 55.47 | 1-102 | 41.83 | 6.68 | 3.00 | 11.31 |
# Table 2: Trends by Year and Site
table2_data = []
for year in sorted(df_head_neck['year'].unique()):
for site in sorted(df_head_neck['cancer_site_group'].unique()):
subset = df_head_neck[(df_head_neck['year'] == year) & (df_head_neck['cancer_site_group'] == site)]
if len(subset) > 0:
n = len(subset)
mortality_pct = (subset['egresar_fallecido'].sum() / n * 100) if n > 0 else 0
mean_age = subset['edad'].mean()
mean_los = subset['dias_estancia'].mean()
any_comorb_pct = subset['any_comorbidity'].mean() * 100 if 'any_comorbidity' in subset.columns else 0
table2_data.append({
'Year': year,
'Site': site,
'N': n,
'Mortality %': mortality_pct,
'Mean Age': mean_age,
'Mean LOS': mean_los,
'Any Comorbidity %': any_comorb_pct
})
table2_df = pd.DataFrame(table2_data)
show_table(table2_df, "Table 2: Trends by Year and Site", index=False)| Year | Site | N | Mortality % | Mean Age | Mean LOS | Any Comorbidity % |
|---|---|---|---|---|---|---|
| 2019 | Hypopharynx | 63 | 3.17 | 69.85 | 11.84 | 65.08 |
| 2019 | Larynx | 450 | 8.22 | 67.61 | 10.87 | 68.00 |
| 2019 | Nasopharynx | 61 | 1.64 | 51.20 | 9.13 | 54.10 |
| 2019 | Oral Cavity | 571 | 4.38 | 62.50 | 7.34 | 58.49 |
| 2019 | Oropharynx | 183 | 6.56 | 62.18 | 6.81 | 47.54 |
| 2019 | Other Pharynx | 50 | 10.00 | 65.85 | 11.80 | 60.00 |
| 2019 | Salivary Glands | 173 | 2.89 | 59.66 | 6.52 | 60.69 |
| 2019 | Sinonasal | 165 | 3.03 | 53.91 | 5.14 | 49.70 |
| 2020 | Hypopharynx | 55 | 9.09 | 68.24 | 13.29 | 67.27 |
| 2020 | Larynx | 343 | 7.87 | 67.47 | 15.12 | 72.89 |
| 2020 | Nasopharynx | 48 | 6.25 | 58.31 | 9.65 | 43.75 |
| 2020 | Oral Cavity | 387 | 4.39 | 64.82 | 11.89 | 70.80 |
| 2020 | Oropharynx | 108 | 7.41 | 61.00 | 8.35 | 63.89 |
| 2020 | Other Pharynx | 35 | 5.71 | 56.63 | 14.40 | 60.00 |
| 2020 | Salivary Glands | 126 | 7.14 | 57.89 | 7.60 | 57.14 |
| 2020 | Sinonasal | 132 | 6.82 | 52.98 | 6.28 | 40.15 |
| 2021 | Hypopharynx | 60 | 13.33 | 65.86 | 15.50 | 55.00 |
| 2021 | Larynx | 408 | 9.07 | 66.90 | 15.57 | 76.72 |
| 2021 | Nasopharynx | 47 | 6.38 | 43.42 | 6.02 | 55.32 |
| 2021 | Oral Cavity | 418 | 7.89 | 65.41 | 9.92 | 71.05 |
| 2021 | Oropharynx | 126 | 4.76 | 60.39 | 6.49 | 57.94 |
| 2021 | Other Pharynx | 54 | 14.81 | 55.85 | 13.00 | 53.70 |
| 2021 | Salivary Glands | 135 | 5.93 | 56.48 | 8.59 | 56.30 |
| 2021 | Sinonasal | 122 | 3.28 | 55.92 | 6.88 | 57.38 |
| 2022 | Hypopharynx | 71 | 9.86 | 66.15 | 15.85 | 64.79 |
| 2022 | Larynx | 407 | 6.39 | 66.92 | 17.27 | 77.15 |
| 2022 | Nasopharynx | 45 | 6.67 | 59.07 | 9.93 | 62.22 |
| 2022 | Oral Cavity | 488 | 6.35 | 63.20 | 10.19 | 69.26 |
| 2022 | Oropharynx | 160 | 6.25 | 63.57 | 8.95 | 60.62 |
| 2022 | Other Pharynx | 49 | 6.12 | 50.27 | 7.00 | 73.47 |
| 2022 | Salivary Glands | 155 | 5.16 | 63.86 | 6.53 | 70.32 |
| 2022 | Sinonasal | 130 | 7.69 | 55.26 | 8.46 | 59.23 |
| 2023 | Hypopharynx | 77 | 6.49 | 70.62 | 14.64 | 74.03 |
| 2023 | Larynx | 504 | 7.34 | 68.16 | 13.00 | 79.96 |
| 2023 | Nasopharynx | 50 | 6.00 | 52.95 | 5.92 | 64.00 |
| 2023 | Oral Cavity | 505 | 5.74 | 65.01 | 9.69 | 71.68 |
| 2023 | Oropharynx | 164 | 7.32 | 63.42 | 10.55 | 68.29 |
| 2023 | Other Pharynx | 38 | 15.79 | 61.21 | 6.50 | 65.79 |
| 2023 | Salivary Glands | 125 | 3.20 | 60.55 | 5.90 | 62.40 |
| 2023 | Sinonasal | 160 | 3.75 | 55.75 | 6.66 | 65.00 |
| 2024 | Hypopharynx | 74 | 8.11 | 70.29 | 11.18 | 72.97 |
| 2024 | Larynx | 513 | 6.63 | 68.36 | 15.66 | 82.85 |
| 2024 | Nasopharynx | 117 | 8.55 | 44.44 | 10.72 | 57.26 |
| 2024 | Oral Cavity | 632 | 6.49 | 65.27 | 9.55 | 74.53 |
| 2024 | Oropharynx | 210 | 3.33 | 61.73 | 8.83 | 67.62 |
| 2024 | Other Pharynx | 55 | 10.91 | 55.94 | 11.82 | 65.45 |
| 2024 | Salivary Glands | 184 | 6.52 | 63.34 | 6.78 | 72.28 |
| 2024 | Sinonasal | 178 | 3.93 | 58.34 | 6.99 | 56.18 |
# Table 3: Comorbidity Prevalence - All Patients
table3_data = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
site_df = df_head_neck[df_head_neck['cancer_site_group'] == site]
n = len(site_df)
row = {'Site': site}
for col in comorbidity_cols:
if col in site_df.columns:
pct = (site_df[col].sum() / n * 100) if n > 0 else 0
row[format_name(col)] = pct
table3_data.append(row)
table3_df = pd.DataFrame(table3_data)
show_table(table3_df, "Table 3: Comorbidity Prevalence - All Patients (%)", index=False)| Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 36.75 | 0.25 | 4.00 | 1.50 | 3.75 | 0.75 | 1.50 | 0.25 | 14.25 | 0.25 | 1.50 | 9.50 | 0.25 | 16.00 | 6.25 | 1.25 | 0.00 | 2.00 | 1.00 | 0.00 | 0.25 | 1.50 | 4.50 | 1.75 | 0.00 | 1.00 | 1.00 | 0.00 | 0.75 | 0.00 | 0.75 | 0.00 | 0.00 | 1.25 | 0.50 | 1.75 | 0.50 | 2.00 | 0.00 | 1.00 | 6.50 | 16.75 | 0.00 | 1.50 | 0.50 | 0.50 | 0.00 |
| Larynx | 45.79 | 0.50 | 5.79 | 3.50 | 6.93 | 2.93 | 3.09 | 0.34 | 19.81 | 0.23 | 4.99 | 8.42 | 0.46 | 11.16 | 13.79 | 3.62 | 0.04 | 1.75 | 1.03 | 0.46 | 0.99 | 6.10 | 5.18 | 1.10 | 0.30 | 0.91 | 1.30 | 0.08 | 0.15 | 0.57 | 1.14 | 0.00 | 0.04 | 1.33 | 0.72 | 1.56 | 2.10 | 2.32 | 0.27 | 0.88 | 4.53 | 24.27 | 0.04 | 0.65 | 0.27 | 0.11 | 0.04 |
| Nasopharynx | 28.26 | 0.00 | 2.17 | 1.63 | 2.99 | 1.63 | 4.08 | 0.00 | 12.50 | 0.27 | 7.07 | 8.97 | 0.54 | 9.51 | 0.82 | 3.53 | 0.00 | 0.54 | 1.09 | 0.00 | 0.27 | 5.43 | 3.26 | 0.54 | 0.27 | 0.27 | 0.27 | 0.00 | 0.00 | 0.27 | 0.27 | 0.27 | 0.00 | 0.00 | 0.27 | 1.09 | 0.82 | 0.54 | 0.00 | 3.80 | 4.08 | 7.88 | 0.00 | 1.36 | 0.00 | 0.00 | 0.00 |
| Oral Cavity | 43.85 | 0.50 | 3.80 | 2.17 | 5.63 | 2.43 | 2.10 | 0.37 | 18.66 | 0.43 | 5.20 | 11.43 | 0.60 | 12.26 | 5.13 | 1.77 | 0.13 | 0.73 | 0.73 | 0.27 | 0.30 | 4.53 | 4.57 | 1.03 | 0.23 | 1.37 | 0.63 | 0.30 | 0.17 | 0.43 | 0.53 | 0.13 | 0.07 | 1.73 | 0.67 | 1.20 | 1.90 | 2.97 | 0.23 | 0.70 | 4.77 | 14.83 | 0.00 | 0.73 | 0.07 | 0.00 | 0.17 |
| Oropharynx | 37.01 | 0.11 | 2.42 | 1.05 | 3.15 | 2.31 | 2.73 | 0.11 | 13.67 | 0.32 | 4.21 | 7.05 | 0.11 | 8.73 | 3.68 | 1.79 | 0.11 | 1.05 | 0.74 | 0.21 | 0.32 | 3.05 | 2.52 | 0.63 | 0.21 | 1.47 | 0.84 | 0.00 | 0.00 | 0.21 | 0.32 | 0.11 | 0.00 | 0.42 | 0.63 | 2.21 | 2.00 | 2.31 | 0.63 | 1.16 | 4.84 | 13.25 | 0.11 | 2.10 | 0.21 | 0.00 | 0.00 |
| Other Pharynx | 33.45 | 0.36 | 2.14 | 2.49 | 4.98 | 3.56 | 3.91 | 0.71 | 15.30 | 0.00 | 2.85 | 6.05 | 0.00 | 9.25 | 4.98 | 1.07 | 0.00 | 1.42 | 1.07 | 0.00 | 3.20 | 3.56 | 4.27 | 1.42 | 0.00 | 1.07 | 0.36 | 0.00 | 0.00 | 1.07 | 1.78 | 0.36 | 0.36 | 0.71 | 0.00 | 3.56 | 1.78 | 2.14 | 0.36 | 0.71 | 1.78 | 11.03 | 0.00 | 1.42 | 0.36 | 0.00 | 0.00 |
| Salivary Glands | 37.53 | 0.89 | 3.01 | 2.23 | 4.68 | 1.67 | 0.89 | 0.33 | 20.04 | 0.22 | 8.02 | 10.47 | 0.67 | 11.36 | 4.12 | 3.79 | 0.00 | 0.22 | 0.78 | 0.22 | 0.78 | 4.23 | 4.23 | 0.78 | 0.11 | 0.78 | 0.22 | 0.11 | 0.00 | 0.22 | 0.45 | 0.22 | 0.00 | 0.78 | 1.22 | 1.78 | 1.34 | 3.12 | 0.45 | 1.00 | 2.00 | 9.24 | 0.00 | 0.56 | 0.56 | 0.00 | 0.00 |
| Sinonasal | 33.03 | 0.56 | 1.24 | 1.35 | 1.69 | 2.14 | 0.79 | 0.45 | 16.01 | 0.23 | 6.43 | 7.33 | 0.23 | 8.46 | 1.80 | 2.59 | 0.00 | 0.00 | 0.56 | 0.00 | 0.68 | 1.69 | 2.03 | 0.79 | 0.23 | 0.90 | 0.23 | 0.00 | 0.00 | 0.34 | 0.34 | 0.00 | 0.00 | 0.90 | 0.68 | 1.58 | 1.13 | 2.03 | 0.11 | 1.24 | 3.27 | 9.47 | 0.00 | 0.45 | 0.00 | 0.00 | 0.00 |
# Table 4: Comorbidity Prevalence - Discharged
table4_data = []
discharged = df_head_neck[df_head_neck['egresar_fallecido'] == 0]
for site in sorted(discharged['cancer_site_group'].unique()):
site_df = discharged[discharged['cancer_site_group'] == site]
n = len(site_df)
row = {'Site': site}
for col in comorbidity_cols:
if col in site_df.columns:
pct = (site_df[col].sum() / n * 100) if n > 0 else 0
row[format_name(col)] = pct
table4_data.append(row)
table4_df = pd.DataFrame(table4_data)
show_table(table4_df, "Table 4: Comorbidity Prevalence - Discharged (%)", index=False)| Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 36.78 | 0.27 | 3.54 | 1.09 | 2.72 | 0.54 | 1.63 | 0.27 | 14.44 | 0.27 | 1.63 | 9.54 | 0.27 | 15.53 | 5.72 | 1.09 | 0.00 | 1.91 | 1.09 | 0.00 | 0.27 | 1.36 | 4.90 | 1.91 | 0.00 | 1.09 | 0.82 | 0.00 | 0.82 | 0.00 | 0.54 | 0.00 | 0.00 | 1.36 | 0.27 | 1.63 | 0.54 | 1.91 | 0.00 | 0.82 | 7.08 | 17.17 | 0.00 | 1.63 | 0.27 | 0.54 | 0.00 |
| Larynx | 45.78 | 0.45 | 5.64 | 3.09 | 5.48 | 2.76 | 2.68 | 0.37 | 19.45 | 0.25 | 5.23 | 8.61 | 0.37 | 11.08 | 13.39 | 3.67 | 0.04 | 1.48 | 0.91 | 0.37 | 0.99 | 4.70 | 4.61 | 0.95 | 0.29 | 0.82 | 1.03 | 0.00 | 0.16 | 0.62 | 0.91 | 0.00 | 0.00 | 1.28 | 0.74 | 1.44 | 2.02 | 2.22 | 0.29 | 0.95 | 4.41 | 24.64 | 0.04 | 0.66 | 0.25 | 0.12 | 0.04 |
| Nasopharynx | 27.83 | 0.00 | 2.03 | 1.74 | 2.61 | 1.16 | 3.48 | 0.00 | 12.46 | 0.29 | 7.54 | 9.57 | 0.29 | 9.28 | 0.87 | 3.77 | 0.00 | 0.29 | 1.16 | 0.00 | 0.29 | 4.64 | 3.48 | 0.58 | 0.29 | 0.29 | 0.29 | 0.00 | 0.00 | 0.29 | 0.29 | 0.29 | 0.00 | 0.00 | 0.00 | 1.16 | 0.87 | 0.58 | 0.00 | 4.06 | 4.35 | 7.25 | 0.00 | 1.45 | 0.00 | 0.00 | 0.00 |
| Oral Cavity | 43.61 | 0.46 | 3.68 | 2.02 | 4.71 | 2.27 | 1.77 | 0.35 | 18.62 | 0.46 | 5.35 | 11.54 | 0.50 | 11.96 | 5.03 | 1.63 | 0.14 | 0.71 | 0.57 | 0.21 | 0.28 | 3.82 | 4.28 | 0.92 | 0.25 | 1.20 | 0.57 | 0.28 | 0.18 | 0.46 | 0.50 | 0.11 | 0.07 | 1.63 | 0.67 | 1.17 | 1.77 | 2.94 | 0.21 | 0.71 | 4.71 | 14.73 | 0.00 | 0.71 | 0.07 | 0.00 | 0.18 |
| Oropharynx | 36.83 | 0.11 | 2.46 | 0.78 | 2.46 | 2.12 | 2.68 | 0.11 | 13.39 | 0.33 | 4.24 | 7.14 | 0.00 | 9.04 | 3.24 | 1.79 | 0.11 | 1.12 | 0.56 | 0.22 | 0.33 | 2.23 | 2.23 | 0.56 | 0.22 | 1.12 | 0.78 | 0.00 | 0.00 | 0.22 | 0.33 | 0.11 | 0.00 | 0.45 | 0.67 | 2.23 | 1.90 | 2.12 | 0.45 | 1.23 | 4.91 | 13.39 | 0.11 | 2.01 | 0.11 | 0.00 | 0.00 |
| Other Pharynx | 31.87 | 0.00 | 2.39 | 1.99 | 3.98 | 3.19 | 3.59 | 0.40 | 15.54 | 0.00 | 2.79 | 6.77 | 0.00 | 9.96 | 4.38 | 1.20 | 0.00 | 1.20 | 1.20 | 0.00 | 3.59 | 1.99 | 3.98 | 1.20 | 0.00 | 1.20 | 0.00 | 0.00 | 0.00 | 1.20 | 1.59 | 0.40 | 0.40 | 0.80 | 0.00 | 3.19 | 1.59 | 1.59 | 0.00 | 0.80 | 1.59 | 10.36 | 0.00 | 1.59 | 0.40 | 0.00 | 0.00 |
| Salivary Glands | 37.79 | 0.82 | 2.70 | 1.88 | 3.76 | 1.41 | 0.94 | 0.35 | 20.31 | 0.12 | 8.33 | 10.68 | 0.70 | 11.38 | 4.23 | 3.99 | 0.00 | 0.12 | 0.82 | 0.23 | 0.82 | 3.29 | 3.99 | 0.70 | 0.12 | 0.70 | 0.12 | 0.12 | 0.00 | 0.23 | 0.47 | 0.23 | 0.00 | 0.70 | 1.29 | 1.88 | 1.29 | 3.17 | 0.47 | 1.06 | 1.88 | 9.27 | 0.00 | 0.59 | 0.59 | 0.00 | 0.00 |
| Sinonasal | 32.86 | 0.47 | 1.30 | 1.30 | 1.65 | 1.77 | 0.83 | 0.35 | 15.72 | 0.24 | 6.26 | 7.33 | 0.12 | 8.51 | 1.77 | 2.72 | 0.00 | 0.00 | 0.59 | 0.00 | 0.71 | 1.54 | 2.01 | 0.83 | 0.24 | 0.83 | 0.24 | 0.00 | 0.00 | 0.35 | 0.35 | 0.00 | 0.00 | 0.95 | 0.71 | 1.54 | 1.06 | 2.01 | 0.12 | 1.30 | 3.31 | 9.34 | 0.00 | 0.47 | 0.00 | 0.00 | 0.00 |
# Table 5: Comorbidity Prevalence - Deceased
table5_data = []
deceased = df_head_neck[df_head_neck['egresar_fallecido'] == 1]
for site in sorted(deceased['cancer_site_group'].unique()):
site_df = deceased[deceased['cancer_site_group'] == site]
n = len(site_df)
row = {'Site': site}
for col in comorbidity_cols:
if col in site_df.columns:
pct = (site_df[col].sum() / n * 100) if n > 0 else 0
row[format_name(col)] = pct
table5_data.append(row)
table5_df = pd.DataFrame(table5_data)
show_table(table5_df, "Table 5: Comorbidity Prevalence - Deceased (%)", index=False)| Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 36.36 | 0.00 | 9.09 | 6.06 | 15.15 | 3.03 | 0.00 | 0.00 | 12.12 | 0.00 | 0.00 | 9.09 | 0.00 | 21.21 | 12.12 | 3.03 | 0.00 | 3.03 | 0.00 | 0.00 | 0.00 | 3.03 | 0.00 | 0.00 | 0.00 | 0.00 | 3.03 | 0.00 | 0.00 | 0.00 | 3.03 | 0.00 | 0.00 | 0.00 | 3.03 | 3.03 | 0.00 | 3.03 | 0.00 | 3.03 | 0.00 | 12.12 | 0.00 | 0.00 | 3.03 | 0.00 | 0.00 |
| Larynx | 45.96 | 1.01 | 7.58 | 8.59 | 24.75 | 5.05 | 8.08 | 0.00 | 24.24 | 0.00 | 2.02 | 6.06 | 1.52 | 12.12 | 18.69 | 3.03 | 0.00 | 5.05 | 2.53 | 1.52 | 1.01 | 23.23 | 12.12 | 3.03 | 0.51 | 2.02 | 4.55 | 1.01 | 0.00 | 0.00 | 4.04 | 0.00 | 0.51 | 2.02 | 0.51 | 3.03 | 3.03 | 3.54 | 0.00 | 0.00 | 6.06 | 19.70 | 0.00 | 0.51 | 0.51 | 0.00 | 0.00 |
| Nasopharynx | 34.78 | 0.00 | 4.35 | 0.00 | 8.70 | 8.70 | 13.04 | 0.00 | 13.04 | 0.00 | 0.00 | 0.00 | 4.35 | 13.04 | 0.00 | 0.00 | 0.00 | 4.35 | 0.00 | 0.00 | 0.00 | 17.39 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 4.35 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 17.39 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
| Oral Cavity | 47.73 | 1.14 | 5.68 | 4.55 | 20.45 | 5.11 | 7.39 | 0.57 | 19.32 | 0.00 | 2.84 | 9.66 | 2.27 | 17.05 | 6.82 | 3.98 | 0.00 | 1.14 | 3.41 | 1.14 | 0.57 | 15.91 | 9.09 | 2.84 | 0.00 | 3.98 | 1.70 | 0.57 | 0.00 | 0.00 | 1.14 | 0.57 | 0.00 | 3.41 | 0.57 | 1.70 | 3.98 | 3.41 | 0.57 | 0.57 | 5.68 | 16.48 | 0.00 | 1.14 | 0.00 | 0.00 | 0.00 |
| Oropharynx | 40.00 | 0.00 | 1.82 | 5.45 | 14.55 | 5.45 | 3.64 | 0.00 | 18.18 | 0.00 | 3.64 | 5.45 | 1.82 | 3.64 | 10.91 | 1.82 | 0.00 | 0.00 | 3.64 | 0.00 | 0.00 | 16.36 | 7.27 | 1.82 | 0.00 | 7.27 | 1.82 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 1.82 | 3.64 | 5.45 | 3.64 | 0.00 | 3.64 | 10.91 | 0.00 | 3.64 | 1.82 | 0.00 | 0.00 |
| Other Pharynx | 46.67 | 3.33 | 0.00 | 6.67 | 13.33 | 6.67 | 6.67 | 3.33 | 13.33 | 0.00 | 3.33 | 0.00 | 0.00 | 3.33 | 10.00 | 0.00 | 0.00 | 3.33 | 0.00 | 0.00 | 0.00 | 16.67 | 6.67 | 3.33 | 0.00 | 0.00 | 3.33 | 0.00 | 0.00 | 0.00 | 3.33 | 0.00 | 0.00 | 0.00 | 0.00 | 6.67 | 3.33 | 6.67 | 3.33 | 0.00 | 3.33 | 16.67 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
| Salivary Glands | 32.61 | 2.17 | 8.70 | 8.70 | 21.74 | 6.52 | 0.00 | 0.00 | 15.22 | 2.17 | 2.17 | 6.52 | 0.00 | 10.87 | 2.17 | 0.00 | 0.00 | 2.17 | 0.00 | 0.00 | 0.00 | 21.74 | 8.70 | 2.17 | 0.00 | 2.17 | 2.17 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 2.17 | 0.00 | 0.00 | 2.17 | 2.17 | 0.00 | 0.00 | 4.35 | 8.70 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
| Sinonasal | 36.59 | 2.44 | 0.00 | 2.44 | 2.44 | 9.76 | 0.00 | 2.44 | 21.95 | 0.00 | 9.76 | 7.32 | 2.44 | 7.32 | 2.44 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 4.88 | 2.44 | 0.00 | 0.00 | 2.44 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 2.44 | 2.44 | 2.44 | 0.00 | 0.00 | 2.44 | 12.20 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
# Table 6: Summary by Discharge Status
table6_data = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
for status in [0, 1]:
status_label = 'Discharged' if status == 0 else 'Deceased'
subset = df_head_neck[(df_head_neck['cancer_site_group'] == site) & (df_head_neck['egresar_fallecido'] == status)]
if len(subset) > 0:
table6_data.append({
'Site': site,
'Status': status_label,
'N': len(subset),
'Mean Age': subset['edad'].mean(),
'Mean LOS': subset['dias_estancia'].mean(),
'Median LOS': subset['dias_estancia'].median()
})
table6_df = pd.DataFrame(table6_data)
show_table(table6_df, "Table 6: Summary by Discharge Status", index=False)| Site | Status | N | Mean Age | Mean LOS | Median LOS |
|---|---|---|---|---|---|
| Hypopharynx | Discharged | 367 | 68.30 | 12.88 | 7.00 |
| Hypopharynx | Deceased | 33 | 71.94 | 23.00 | 13.00 |
| Larynx | Discharged | 2427 | 67.30 | 13.67 | 7.00 |
| Larynx | Deceased | 198 | 71.64 | 24.63 | 12.50 |
| Nasopharynx | Discharged | 345 | 49.49 | 7.73 | 4.00 |
| Nasopharynx | Deceased | 23 | 60.67 | 27.48 | 13.00 |
| Oral Cavity | Discharged | 2825 | 64.06 | 9.22 | 4.00 |
| Oral Cavity | Deceased | 176 | 68.54 | 15.80 | 6.00 |
| Oropharynx | Discharged | 896 | 61.89 | 7.87 | 4.00 |
| Oropharynx | Deceased | 55 | 66.57 | 16.96 | 9.00 |
| Other Pharynx | Discharged | 251 | 55.71 | 10.04 | 5.00 |
| Other Pharynx | Deceased | 30 | 72.43 | 17.23 | 5.50 |
| Salivary Glands | Discharged | 852 | 60.34 | 6.41 | 3.00 |
| Salivary Glands | Deceased | 46 | 64.11 | 16.93 | 9.00 |
| Sinonasal | Discharged | 846 | 55.32 | 6.27 | 3.00 |
| Sinonasal | Deceased | 41 | 58.38 | 15.20 | 11.00 |
# Table 7: LOS by Site and Discharge Status
table7_data = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
for status in [0, 1]:
status_label = 'Discharged' if status == 0 else 'Deceased'
subset = df_head_neck[(df_head_neck['cancer_site_group'] == site) & (df_head_neck['egresar_fallecido'] == status)]
if len(subset) > 0:
los_vals = subset['dias_estancia']
q1 = los_vals.quantile(0.25)
q3 = los_vals.quantile(0.75)
iqr = q3 - q1
table7_data.append({
'Site': site,
'Status': status_label,
'N': len(subset),
'Mean LOS': los_vals.mean(),
'Median LOS': los_vals.median(),
'IQR': iqr
})
table7_df = pd.DataFrame(table7_data)
show_table(table7_df, "Table 7: Length of Stay by Site and Discharge Status", index=False)| Site | Status | N | Mean LOS | Median LOS | IQR |
|---|---|---|---|---|---|
| Hypopharynx | Discharged | 367 | 12.88 | 7.00 | 14.00 |
| Hypopharynx | Deceased | 33 | 23.00 | 13.00 | 27.00 |
| Larynx | Discharged | 2427 | 13.67 | 7.00 | 16.00 |
| Larynx | Deceased | 198 | 24.63 | 12.50 | 25.50 |
| Nasopharynx | Discharged | 345 | 7.73 | 4.00 | 6.00 |
| Nasopharynx | Deceased | 23 | 27.48 | 13.00 | 14.50 |
| Oral Cavity | Discharged | 2825 | 9.22 | 4.00 | 9.00 |
| Oral Cavity | Deceased | 176 | 15.80 | 6.00 | 12.25 |
| Oropharynx | Discharged | 896 | 7.87 | 4.00 | 8.00 |
| Oropharynx | Deceased | 55 | 16.96 | 9.00 | 18.00 |
| Other Pharynx | Discharged | 251 | 10.04 | 5.00 | 9.50 |
| Other Pharynx | Deceased | 30 | 17.23 | 5.50 | 23.00 |
| Salivary Glands | Discharged | 852 | 6.41 | 3.00 | 6.00 |
| Salivary Glands | Deceased | 46 | 16.93 | 9.00 | 20.25 |
| Sinonasal | Discharged | 846 | 6.27 | 3.00 | 6.00 |
| Sinonasal | Deceased | 41 | 15.20 | 11.00 | 19.00 |
# Table 8: Yearly Trends Aggregate
table8_data = []
for year in sorted(df_head_neck['year'].unique()):
year_df = df_head_neck[df_head_neck['year'] == year]
n_total = len(year_df)
mortality_pct = (year_df['egresar_fallecido'].sum() / n_total * 100) if n_total > 0 else 0
mean_los = year_df['dias_estancia'].mean()
mean_age = year_df['edad'].mean()
row = {
'Year': year,
'Total': n_total,
'Mortality %': mortality_pct,
'Mean Age': mean_age,
'Mean LOS': mean_los
}
# Add per-site counts
for site in sorted(df_head_neck['cancer_site_group'].unique()):
site_count = len(year_df[year_df['cancer_site_group'] == site])
row[f'{site} N'] = site_count
table8_data.append(row)
table8_df = pd.DataFrame(table8_data)
show_table(table8_df, "Table 8: Yearly Trends (Aggregate)", index=False)| Year | Total | Mortality % | Mean Age | Mean LOS | Hypopharynx N | Larynx N | Nasopharynx N | Oral Cavity N | Oropharynx N | Other Pharynx N | Salivary Glands N | Sinonasal N |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2019 | 1716 | 5.36 | 62.66 | 8.27 | 63 | 450 | 61 | 571 | 183 | 50 | 173 | 165 |
| 2020 | 1234 | 6.48 | 62.91 | 11.49 | 55 | 343 | 48 | 387 | 108 | 35 | 126 | 132 |
| 2021 | 1370 | 7.81 | 62.56 | 11.12 | 60 | 408 | 47 | 418 | 126 | 54 | 135 | 122 |
| 2022 | 1505 | 6.51 | 63.22 | 11.60 | 71 | 407 | 45 | 488 | 160 | 49 | 155 | 130 |
| 2023 | 1623 | 6.28 | 64.38 | 10.26 | 77 | 504 | 50 | 505 | 164 | 38 | 125 | 160 |
| 2024 | 1963 | 6.27 | 63.57 | 10.77 | 74 | 513 | 117 | 632 | 210 | 55 | 184 | 178 |
# Create visualization
fig, axes = plt.subplots(2, 3, figsize=(16, 10))
# 1. Age distribution
axes[0, 0].hist(df_head_neck['edad'], bins=50, color='steelblue', edgecolor='black', alpha=0.7)
axes[0, 0].axvline(df_head_neck['edad'].mean(), color='red', linestyle='--', linewidth=2,
label=f'Mean: {df_head_neck["edad"].mean():.1f}')
axes[0, 0].set_xlabel('Age (years)')
axes[0, 0].set_ylabel('Frequency')
axes[0, 0].set_title('Age Distribution')
axes[0, 0].legend()
# 2. Length of stay distribution
axes[0, 1].hist(df_head_neck['dias_estancia'], bins=50, color='forestgreen', edgecolor='black', alpha=0.7)
axes[0, 1].axvline(df_head_neck['dias_estancia'].mean(), color='red', linestyle='--', linewidth=2,
label=f'Mean: {df_head_neck["dias_estancia"].mean():.1f}')
axes[0, 1].set_xlabel('Length of stay (days)')
axes[0, 1].set_ylabel('Frequency')
axes[0, 1].set_title('Length of Stay Distribution')
axes[0, 1].legend()
# 3. Mortality by sex
mortality_by_sex = df_head_neck.groupby('SEXO')['egresar_fallecido'].mean() * 100
axes[0, 2].bar(mortality_by_sex.index, mortality_by_sex.values, color=['#FF6B6B', '#4ECDC4'])
axes[0, 2].set_ylabel('Mortality rate (%)')
axes[0, 2].set_title('Mortality by Sex')
for i, v in enumerate(mortality_by_sex.values):
axes[0, 2].text(i, v + 0.5, f'{v:.1f}%', ha='center', fontweight='bold')
# 4. Records by year
year_counts = df_head_neck['year'].value_counts().sort_index()
axes[1, 0].plot(year_counts.index, year_counts.values, marker='o', linewidth=2, markersize=8, color='darkviolet')
axes[1, 0].fill_between(year_counts.index, year_counts.values, alpha=0.3, color='darkviolet')
axes[1, 0].set_xlabel('Year')
axes[1, 0].set_ylabel('Number of records')
axes[1, 0].set_title('Records by Year')
axes[1, 0].grid(True, alpha=0.3)
# 5. Cancer site distribution
site_counts = df_head_neck['cancer_site_group'].value_counts()
axes[1, 1].barh(site_counts.index, site_counts.values, color='coral')
axes[1, 1].set_xlabel('Number of cases')
axes[1, 1].set_title('Cases by Cancer Site')
for i, v in enumerate(site_counts.values):
axes[1, 1].text(v + 50, i, f'{v:,.0f}', va='center')
# 6. Service category
service_counts = df_head_neck['service_category'].value_counts()
colors_pie = ['#FF6B6B', '#4ECDC4', '#45B7D1', '#FFA07A']
axes[1, 2].pie(service_counts.values, labels=service_counts.index, autopct='%1.1f%%', startangle=90)
axes[1, 2].set_title('Service Category Distribution')
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '01_descriptive_stats.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-descriptive-stats.png", dpi=300, bbox_inches="tight")
plt.show()
fig, axes = plt.subplots(1, 2, figsize=(14, 6))
# Violin plot
ax = axes[0]
for i, site in enumerate(df_head_neck['cancer_site_group'].unique()):
data = df_head_neck[df_head_neck['cancer_site_group'] == site]['edad']
positions = [i]
ax.violinplot([data], positions=positions, widths=0.7, showmeans=True, showmedians=True)
ax.set_xticks(range(len(df_head_neck['cancer_site_group'].unique())))
ax.set_xticklabels(df_head_neck['cancer_site_group'].unique(), rotation=30, ha='right', fontsize=9)
ax.set_ylabel('Age (years)')
ax.set_title('Age Distribution by Cancer Site (Violin Plot)')
ax.grid(True, alpha=0.3, axis='y')
# Box plot
ax = axes[1]
data_by_site = [df_head_neck[df_head_neck['cancer_site_group'] == site]['edad'].values
for site in df_head_neck['cancer_site_group'].unique()]
bp = ax.boxplot(data_by_site, labels=df_head_neck['cancer_site_group'].unique(), patch_artist=True)
for patch in bp['boxes']:
patch.set_facecolor('lightblue')
ax.set_ylabel('Age (years)')
ax.set_title('Age Distribution by Cancer Site (Box Plot)')
plt.setp(ax.get_xticklabels(), rotation=30, ha='right', fontsize=9)
ax.grid(True, alpha=0.3, axis='y')
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '02_age_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-age-by-site.png", dpi=300, bbox_inches="tight")
plt.show()
# Table: Comorbidity prevalence - ALL patients
comorb_all = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row = {'Cancer Site': site}
for col in comorbidity_cols:
prev = (df_site[col].mean() * 100)
row[format_name(col)] = f'{prev:.1f}%'
comorb_all.append(row)
comorb_all_df = pd.DataFrame(comorb_all)
show_table(comorb_all_df, "Table 13: Comorbidity Prevalence - All Patients (%)", index=False)| Cancer Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 36.8% | 0.2% | 4.0% | 1.5% | 3.8% | 0.8% | 1.5% | 0.2% | 14.2% | 0.2% | 1.5% | 9.5% | 0.2% | 16.0% | 6.2% | 1.2% | 0.0% | 2.0% | 1.0% | 0.0% | 0.2% | 1.5% | 4.5% | 1.8% | 0.0% | 1.0% | 1.0% | 0.0% | 0.8% | 0.0% | 0.8% | 0.0% | 0.0% | 1.2% | 0.5% | 1.8% | 0.5% | 2.0% | 0.0% | 1.0% | 6.5% | 16.8% | 0.0% | 1.5% | 0.5% | 0.5% | 0.0% |
| Larynx | 45.8% | 0.5% | 5.8% | 3.5% | 6.9% | 2.9% | 3.1% | 0.3% | 19.8% | 0.2% | 5.0% | 8.4% | 0.5% | 11.2% | 13.8% | 3.6% | 0.0% | 1.8% | 1.0% | 0.5% | 1.0% | 6.1% | 5.2% | 1.1% | 0.3% | 0.9% | 1.3% | 0.1% | 0.2% | 0.6% | 1.1% | 0.0% | 0.0% | 1.3% | 0.7% | 1.6% | 2.1% | 2.3% | 0.3% | 0.9% | 4.5% | 24.3% | 0.0% | 0.6% | 0.3% | 0.1% | 0.0% |
| Nasopharynx | 28.3% | 0.0% | 2.2% | 1.6% | 3.0% | 1.6% | 4.1% | 0.0% | 12.5% | 0.3% | 7.1% | 9.0% | 0.5% | 9.5% | 0.8% | 3.5% | 0.0% | 0.5% | 1.1% | 0.0% | 0.3% | 5.4% | 3.3% | 0.5% | 0.3% | 0.3% | 0.3% | 0.0% | 0.0% | 0.3% | 0.3% | 0.3% | 0.0% | 0.0% | 0.3% | 1.1% | 0.8% | 0.5% | 0.0% | 3.8% | 4.1% | 7.9% | 0.0% | 1.4% | 0.0% | 0.0% | 0.0% |
| Oral Cavity | 43.9% | 0.5% | 3.8% | 2.2% | 5.6% | 2.4% | 2.1% | 0.4% | 18.7% | 0.4% | 5.2% | 11.4% | 0.6% | 12.3% | 5.1% | 1.8% | 0.1% | 0.7% | 0.7% | 0.3% | 0.3% | 4.5% | 4.6% | 1.0% | 0.2% | 1.4% | 0.6% | 0.3% | 0.2% | 0.4% | 0.5% | 0.1% | 0.1% | 1.7% | 0.7% | 1.2% | 1.9% | 3.0% | 0.2% | 0.7% | 4.8% | 14.8% | 0.0% | 0.7% | 0.1% | 0.0% | 0.2% |
| Oropharynx | 37.0% | 0.1% | 2.4% | 1.1% | 3.2% | 2.3% | 2.7% | 0.1% | 13.7% | 0.3% | 4.2% | 7.0% | 0.1% | 8.7% | 3.7% | 1.8% | 0.1% | 1.1% | 0.7% | 0.2% | 0.3% | 3.0% | 2.5% | 0.6% | 0.2% | 1.5% | 0.8% | 0.0% | 0.0% | 0.2% | 0.3% | 0.1% | 0.0% | 0.4% | 0.6% | 2.2% | 2.0% | 2.3% | 0.6% | 1.2% | 4.8% | 13.2% | 0.1% | 2.1% | 0.2% | 0.0% | 0.0% |
| Other Pharynx | 33.5% | 0.4% | 2.1% | 2.5% | 5.0% | 3.6% | 3.9% | 0.7% | 15.3% | 0.0% | 2.8% | 6.0% | 0.0% | 9.3% | 5.0% | 1.1% | 0.0% | 1.4% | 1.1% | 0.0% | 3.2% | 3.6% | 4.3% | 1.4% | 0.0% | 1.1% | 0.4% | 0.0% | 0.0% | 1.1% | 1.8% | 0.4% | 0.4% | 0.7% | 0.0% | 3.6% | 1.8% | 2.1% | 0.4% | 0.7% | 1.8% | 11.0% | 0.0% | 1.4% | 0.4% | 0.0% | 0.0% |
| Salivary Glands | 37.5% | 0.9% | 3.0% | 2.2% | 4.7% | 1.7% | 0.9% | 0.3% | 20.0% | 0.2% | 8.0% | 10.5% | 0.7% | 11.4% | 4.1% | 3.8% | 0.0% | 0.2% | 0.8% | 0.2% | 0.8% | 4.2% | 4.2% | 0.8% | 0.1% | 0.8% | 0.2% | 0.1% | 0.0% | 0.2% | 0.4% | 0.2% | 0.0% | 0.8% | 1.2% | 1.8% | 1.3% | 3.1% | 0.4% | 1.0% | 2.0% | 9.2% | 0.0% | 0.6% | 0.6% | 0.0% | 0.0% |
| Sinonasal | 33.0% | 0.6% | 1.2% | 1.4% | 1.7% | 2.1% | 0.8% | 0.5% | 16.0% | 0.2% | 6.4% | 7.3% | 0.2% | 8.5% | 1.8% | 2.6% | 0.0% | 0.0% | 0.6% | 0.0% | 0.7% | 1.7% | 2.0% | 0.8% | 0.2% | 0.9% | 0.2% | 0.0% | 0.0% | 0.3% | 0.3% | 0.0% | 0.0% | 0.9% | 0.7% | 1.6% | 1.1% | 2.0% | 0.1% | 1.2% | 3.3% | 9.5% | 0.0% | 0.5% | 0.0% | 0.0% | 0.0% |
# Table: Comorbidity prevalence - DISCHARGED patients
comorb_discharged = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[(df_head_neck['cancer_site_group'] == site) & (df_head_neck['egresar_fallecido'] == 0)]
row = {'Cancer Site': site}
for col in comorbidity_cols:
prev = (df_site[col].mean() * 100) if len(df_site) > 0 else 0
row[format_name(col)] = f'{prev:.1f}%'
comorb_discharged.append(row)
comorb_discharged_df = pd.DataFrame(comorb_discharged)
show_table(comorb_discharged_df, "Table 14: Comorbidity Prevalence - Discharged Patients (%)", index=False)| Cancer Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 36.8% | 0.3% | 3.5% | 1.1% | 2.7% | 0.5% | 1.6% | 0.3% | 14.4% | 0.3% | 1.6% | 9.5% | 0.3% | 15.5% | 5.7% | 1.1% | 0.0% | 1.9% | 1.1% | 0.0% | 0.3% | 1.4% | 4.9% | 1.9% | 0.0% | 1.1% | 0.8% | 0.0% | 0.8% | 0.0% | 0.5% | 0.0% | 0.0% | 1.4% | 0.3% | 1.6% | 0.5% | 1.9% | 0.0% | 0.8% | 7.1% | 17.2% | 0.0% | 1.6% | 0.3% | 0.5% | 0.0% |
| Larynx | 45.8% | 0.5% | 5.6% | 3.1% | 5.5% | 2.8% | 2.7% | 0.4% | 19.4% | 0.2% | 5.2% | 8.6% | 0.4% | 11.1% | 13.4% | 3.7% | 0.0% | 1.5% | 0.9% | 0.4% | 1.0% | 4.7% | 4.6% | 0.9% | 0.3% | 0.8% | 1.0% | 0.0% | 0.2% | 0.6% | 0.9% | 0.0% | 0.0% | 1.3% | 0.7% | 1.4% | 2.0% | 2.2% | 0.3% | 0.9% | 4.4% | 24.6% | 0.0% | 0.7% | 0.2% | 0.1% | 0.0% |
| Nasopharynx | 27.8% | 0.0% | 2.0% | 1.7% | 2.6% | 1.2% | 3.5% | 0.0% | 12.5% | 0.3% | 7.5% | 9.6% | 0.3% | 9.3% | 0.9% | 3.8% | 0.0% | 0.3% | 1.2% | 0.0% | 0.3% | 4.6% | 3.5% | 0.6% | 0.3% | 0.3% | 0.3% | 0.0% | 0.0% | 0.3% | 0.3% | 0.3% | 0.0% | 0.0% | 0.0% | 1.2% | 0.9% | 0.6% | 0.0% | 4.1% | 4.3% | 7.2% | 0.0% | 1.4% | 0.0% | 0.0% | 0.0% |
| Oral Cavity | 43.6% | 0.5% | 3.7% | 2.0% | 4.7% | 2.3% | 1.8% | 0.4% | 18.6% | 0.5% | 5.3% | 11.5% | 0.5% | 12.0% | 5.0% | 1.6% | 0.1% | 0.7% | 0.6% | 0.2% | 0.3% | 3.8% | 4.3% | 0.9% | 0.2% | 1.2% | 0.6% | 0.3% | 0.2% | 0.5% | 0.5% | 0.1% | 0.1% | 1.6% | 0.7% | 1.2% | 1.8% | 2.9% | 0.2% | 0.7% | 4.7% | 14.7% | 0.0% | 0.7% | 0.1% | 0.0% | 0.2% |
| Oropharynx | 36.8% | 0.1% | 2.5% | 0.8% | 2.5% | 2.1% | 2.7% | 0.1% | 13.4% | 0.3% | 4.2% | 7.1% | 0.0% | 9.0% | 3.2% | 1.8% | 0.1% | 1.1% | 0.6% | 0.2% | 0.3% | 2.2% | 2.2% | 0.6% | 0.2% | 1.1% | 0.8% | 0.0% | 0.0% | 0.2% | 0.3% | 0.1% | 0.0% | 0.4% | 0.7% | 2.2% | 1.9% | 2.1% | 0.4% | 1.2% | 4.9% | 13.4% | 0.1% | 2.0% | 0.1% | 0.0% | 0.0% |
| Other Pharynx | 31.9% | 0.0% | 2.4% | 2.0% | 4.0% | 3.2% | 3.6% | 0.4% | 15.5% | 0.0% | 2.8% | 6.8% | 0.0% | 10.0% | 4.4% | 1.2% | 0.0% | 1.2% | 1.2% | 0.0% | 3.6% | 2.0% | 4.0% | 1.2% | 0.0% | 1.2% | 0.0% | 0.0% | 0.0% | 1.2% | 1.6% | 0.4% | 0.4% | 0.8% | 0.0% | 3.2% | 1.6% | 1.6% | 0.0% | 0.8% | 1.6% | 10.4% | 0.0% | 1.6% | 0.4% | 0.0% | 0.0% |
| Salivary Glands | 37.8% | 0.8% | 2.7% | 1.9% | 3.8% | 1.4% | 0.9% | 0.4% | 20.3% | 0.1% | 8.3% | 10.7% | 0.7% | 11.4% | 4.2% | 4.0% | 0.0% | 0.1% | 0.8% | 0.2% | 0.8% | 3.3% | 4.0% | 0.7% | 0.1% | 0.7% | 0.1% | 0.1% | 0.0% | 0.2% | 0.5% | 0.2% | 0.0% | 0.7% | 1.3% | 1.9% | 1.3% | 3.2% | 0.5% | 1.1% | 1.9% | 9.3% | 0.0% | 0.6% | 0.6% | 0.0% | 0.0% |
| Sinonasal | 32.9% | 0.5% | 1.3% | 1.3% | 1.7% | 1.8% | 0.8% | 0.4% | 15.7% | 0.2% | 6.3% | 7.3% | 0.1% | 8.5% | 1.8% | 2.7% | 0.0% | 0.0% | 0.6% | 0.0% | 0.7% | 1.5% | 2.0% | 0.8% | 0.2% | 0.8% | 0.2% | 0.0% | 0.0% | 0.4% | 0.4% | 0.0% | 0.0% | 0.9% | 0.7% | 1.5% | 1.1% | 2.0% | 0.1% | 1.3% | 3.3% | 9.3% | 0.0% | 0.5% | 0.0% | 0.0% | 0.0% |
# Table: Comorbidity prevalence - DECEASED patients
comorb_deceased = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[(df_head_neck['cancer_site_group'] == site) & (df_head_neck['egresar_fallecido'] == 1)]
row = {'Cancer Site': site}
for col in comorbidity_cols:
prev = (df_site[col].mean() * 100) if len(df_site) > 0 else 0
row[format_name(col)] = f'{prev:.1f}%'
comorb_deceased.append(row)
comorb_deceased_df = pd.DataFrame(comorb_deceased)
show_table(comorb_deceased_df, "Table 15: Comorbidity Prevalence - Deceased Patients (%)", index=False)| Cancer Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 36.4% | 0.0% | 9.1% | 6.1% | 15.2% | 3.0% | 0.0% | 0.0% | 12.1% | 0.0% | 0.0% | 9.1% | 0.0% | 21.2% | 12.1% | 3.0% | 0.0% | 3.0% | 0.0% | 0.0% | 0.0% | 3.0% | 0.0% | 0.0% | 0.0% | 0.0% | 3.0% | 0.0% | 0.0% | 0.0% | 3.0% | 0.0% | 0.0% | 0.0% | 3.0% | 3.0% | 0.0% | 3.0% | 0.0% | 3.0% | 0.0% | 12.1% | 0.0% | 0.0% | 3.0% | 0.0% | 0.0% |
| Larynx | 46.0% | 1.0% | 7.6% | 8.6% | 24.7% | 5.1% | 8.1% | 0.0% | 24.2% | 0.0% | 2.0% | 6.1% | 1.5% | 12.1% | 18.7% | 3.0% | 0.0% | 5.1% | 2.5% | 1.5% | 1.0% | 23.2% | 12.1% | 3.0% | 0.5% | 2.0% | 4.5% | 1.0% | 0.0% | 0.0% | 4.0% | 0.0% | 0.5% | 2.0% | 0.5% | 3.0% | 3.0% | 3.5% | 0.0% | 0.0% | 6.1% | 19.7% | 0.0% | 0.5% | 0.5% | 0.0% | 0.0% |
| Nasopharynx | 34.8% | 0.0% | 4.3% | 0.0% | 8.7% | 8.7% | 13.0% | 0.0% | 13.0% | 0.0% | 0.0% | 0.0% | 4.3% | 13.0% | 0.0% | 0.0% | 0.0% | 4.3% | 0.0% | 0.0% | 0.0% | 17.4% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 4.3% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 17.4% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Oral Cavity | 47.7% | 1.1% | 5.7% | 4.5% | 20.5% | 5.1% | 7.4% | 0.6% | 19.3% | 0.0% | 2.8% | 9.7% | 2.3% | 17.0% | 6.8% | 4.0% | 0.0% | 1.1% | 3.4% | 1.1% | 0.6% | 15.9% | 9.1% | 2.8% | 0.0% | 4.0% | 1.7% | 0.6% | 0.0% | 0.0% | 1.1% | 0.6% | 0.0% | 3.4% | 0.6% | 1.7% | 4.0% | 3.4% | 0.6% | 0.6% | 5.7% | 16.5% | 0.0% | 1.1% | 0.0% | 0.0% | 0.0% |
| Oropharynx | 40.0% | 0.0% | 1.8% | 5.5% | 14.5% | 5.5% | 3.6% | 0.0% | 18.2% | 0.0% | 3.6% | 5.5% | 1.8% | 3.6% | 10.9% | 1.8% | 0.0% | 0.0% | 3.6% | 0.0% | 0.0% | 16.4% | 7.3% | 1.8% | 0.0% | 7.3% | 1.8% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 1.8% | 3.6% | 5.5% | 3.6% | 0.0% | 3.6% | 10.9% | 0.0% | 3.6% | 1.8% | 0.0% | 0.0% |
| Other Pharynx | 46.7% | 3.3% | 0.0% | 6.7% | 13.3% | 6.7% | 6.7% | 3.3% | 13.3% | 0.0% | 3.3% | 0.0% | 0.0% | 3.3% | 10.0% | 0.0% | 0.0% | 3.3% | 0.0% | 0.0% | 0.0% | 16.7% | 6.7% | 3.3% | 0.0% | 0.0% | 3.3% | 0.0% | 0.0% | 0.0% | 3.3% | 0.0% | 0.0% | 0.0% | 0.0% | 6.7% | 3.3% | 6.7% | 3.3% | 0.0% | 3.3% | 16.7% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Salivary Glands | 32.6% | 2.2% | 8.7% | 8.7% | 21.7% | 6.5% | 0.0% | 0.0% | 15.2% | 2.2% | 2.2% | 6.5% | 0.0% | 10.9% | 2.2% | 0.0% | 0.0% | 2.2% | 0.0% | 0.0% | 0.0% | 21.7% | 8.7% | 2.2% | 0.0% | 2.2% | 2.2% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 2.2% | 0.0% | 0.0% | 2.2% | 2.2% | 0.0% | 0.0% | 4.3% | 8.7% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Sinonasal | 36.6% | 2.4% | 0.0% | 2.4% | 2.4% | 9.8% | 0.0% | 2.4% | 22.0% | 0.0% | 9.8% | 7.3% | 2.4% | 7.3% | 2.4% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 4.9% | 2.4% | 0.0% | 0.0% | 2.4% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 2.4% | 2.4% | 2.4% | 0.0% | 0.0% | 2.4% | 12.2% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
# Table: Difference (Deceased - Discharged) with chi-square
from scipy.stats import chi2_contingency
comorb_diff = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row = {'Cancer Site': site}
for col in comorbidity_cols:
df_discharged = df_site[df_site['egresar_fallecido'] == 0]
df_deceased = df_site[df_site['egresar_fallecido'] == 1]
prev_discharged = (df_discharged[col].mean() * 100) if len(df_discharged) > 0 else 0
prev_deceased = (df_deceased[col].mean() * 100) if len(df_deceased) > 0 else 0
diff = prev_deceased - prev_discharged
# Chi-square test
contingency_table = pd.crosstab(df_site['egresar_fallecido'], df_site[col])
try:
chi2, pval, dof, expected = chi2_contingency(contingency_table)
sig_stars = ''
if pval < 0.001:
sig_stars = '***'
elif pval < 0.01:
sig_stars = '**'
elif pval < 0.05:
sig_stars = '*'
except:
sig_stars = ''
row[format_name(col)] = f'{diff:+.1f}%{sig_stars}'
comorb_diff.append(row)
comorb_diff_df = pd.DataFrame(comorb_diff)
show_table(comorb_diff_df, "Table 16: Difference in Comorbidity Prevalence (Deceased - Discharged, %)\n* p<0.05, ** p<0.01, *** p<0.001", index=False)| Cancer Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | -0.4% | -0.3% | +5.5% | +5.0% | +12.4%** | +2.5% | -1.6% | -0.3% | -2.3% | -0.3% | -1.6% | -0.4% | -0.3% | +5.7% | +6.4% | +1.9% | +0.0% | +1.1% | -1.1% | +0.0% | -0.3% | +1.7% | -4.9% | -1.9% | +0.0% | -1.1% | +2.2% | +0.0% | -0.8% | +0.0% | +2.5% | +0.0% | +0.0% | -1.4% | +2.8% | +1.4% | -0.5% | +1.1% | +0.0% | +2.2% | -7.1% | -5.0% | +0.0% | -1.6% | +2.8% | -0.5% | +0.0% |
| Larynx | +0.2% | +0.6% | +1.9% | +5.5%*** | +19.3%*** | +2.3% | +5.4%*** | -0.4% | +4.8% | -0.2% | -3.2% | -2.6% | +1.1% | +1.0% | +5.3%* | -0.6% | -0.0% | +3.6%*** | +1.6% | +1.1% | +0.0% | +18.5%*** | +7.5%*** | +2.1%* | +0.2% | +1.2% | +3.5%*** | +1.0%*** | -0.2% | -0.6% | +3.1%*** | +0.0% | +0.5% | +0.7% | -0.2% | +1.6% | +1.0% | +1.3% | -0.3% | -0.9% | +1.7% | -4.9% | -0.0% | -0.2% | +0.3% | -0.1% | -0.0% |
| Nasopharynx | +7.0% | +0.0% | +2.3% | -1.7% | +6.1% | +7.5% | +9.6% | +0.0% | +0.6% | -0.3% | -7.5% | -9.6% | +4.1% | +3.8% | -0.9% | -3.8% | +0.0% | +4.1% | -1.2% | +0.0% | -0.3% | +12.8%* | -3.5% | -0.6% | -0.3% | -0.3% | -0.3% | +0.0% | +0.0% | -0.3% | -0.3% | -0.3% | +0.0% | +0.0% | +4.3% | -1.2% | -0.9% | -0.6% | +0.0% | -4.1% | -4.3% | +10.1% | +0.0% | -1.4% | +0.0% | +0.0% | +0.0% |
| Oral Cavity | +4.1% | +0.7% | +2.0% | +2.5%* | +15.7%*** | +2.8%* | +5.6%*** | +0.2% | +0.7% | -0.5% | -2.5% | -1.9% | +1.8%* | +5.1% | +1.8% | +2.3%* | -0.1% | +0.4% | +2.8%*** | +0.9% | +0.3% | +12.1%*** | +4.8%** | +1.9%* | -0.2% | +2.8%** | +1.1% | +0.3% | -0.2% | -0.5% | +0.6% | +0.5% | -0.1% | +1.8% | -0.1% | +0.5% | +2.2% | +0.5% | +0.4% | -0.1% | +1.0% | +1.8% | +0.0% | +0.4% | -0.1% | +0.0% | -0.2% |
| Oropharynx | +3.2% | -0.1% | -0.6% | +4.7%** | +12.1%*** | +3.3% | +1.0% | -0.1% | +4.8% | -0.3% | -0.6% | -1.7% | +1.8% | -5.4% | +7.7%* | +0.0% | -0.1% | -1.1% | +3.1% | -0.2% | -0.3% | +14.1%*** | +5.0% | +1.3% | -0.2% | +6.2%** | +1.0% | +0.0% | +0.0% | -0.2% | -0.3% | -0.1% | +0.0% | -0.4% | -0.7% | -0.4% | +1.7% | +3.3% | +3.2%* | -1.2% | -1.3% | -2.5% | -0.1% | +1.6% | +1.7% | +0.0% | +0.0% |
| Other Pharynx | +14.8% | +3.3% | -2.4% | +4.7% | +9.3% | +3.5% | +3.1% | +2.9% | -2.2% | +0.0% | +0.5% | -6.8% | +0.0% | -6.6% | +5.6% | -1.2% | +0.0% | +2.1% | -1.2% | +0.0% | -3.6% | +14.7%*** | +2.7% | +2.1% | +0.0% | -1.2% | +3.3% | +0.0% | +0.0% | -1.2% | +1.7% | -0.4% | -0.4% | -0.8% | +0.0% | +3.5% | +1.7% | +5.1% | +3.3% | -0.8% | +1.7% | +6.3% | +0.0% | -1.6% | -0.4% | +0.0% | +0.0% |
| Salivary Glands | -5.2% | +1.4% | +6.0% | +6.8%* | +18.0%*** | +5.1%* | -0.9% | -0.4% | -5.1% | +2.1% | -6.2% | -4.2% | -0.7% | -0.5% | -2.1% | -4.0% | +0.0% | +2.1% | -0.8% | -0.2% | -0.8% | +18.5%*** | +4.7% | +1.5% | -0.1% | +1.5% | +2.1% | -0.1% | +0.0% | -0.2% | -0.5% | -0.2% | +0.0% | +1.5% | -1.3% | -1.9% | +0.9% | -1.0% | -0.5% | -1.1% | +2.5% | -0.6% | +0.0% | -0.6% | -0.6% | +0.0% | +0.0% |
| Sinonasal | +3.7% | +2.0% | -1.3% | +1.1% | +0.8% | +8.0%** | -0.8% | +2.1% | +6.2% | -0.2% | +3.5% | -0.0% | +2.3% | -1.2% | +0.7% | -2.7% | +0.0% | +0.0% | -0.6% | +0.0% | -0.7% | +3.3% | +0.4% | -0.8% | -0.2% | +1.6% | -0.2% | +0.0% | +0.0% | -0.4% | -0.4% | +0.0% | +0.0% | -0.9% | -0.7% | +0.9% | +1.4% | +0.4% | -0.1% | -1.3% | -0.9% | +2.9% | +0.0% | -0.5% | +0.0% | +0.0% | +0.0% |
# Prepare data for heatmap
heatmap_data_all = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row = []
for col in comorbidity_cols:
prev = (df_site[col].mean() * 100)
row.append(prev)
heatmap_data_all.append(row)
heatmap_df_all = pd.DataFrame(heatmap_data_all,
index=sorted(df_head_neck['cancer_site_group'].unique()),
columns=[format_name(col) for col in comorbidity_cols])
fig, ax = plt.subplots(figsize=(22, 8))
sns.heatmap(heatmap_df_all, annot=True, fmt='.1f', cmap='viridis',
cbar_kws={'label': 'Prevalence (%)'}, ax=ax,
annot_kws={'rotation': 90, 'fontsize': 8})
ax.set_title('Comorbidity Prevalence by Cancer Site (All Patients)')
ax.set_xlabel('Comorbidity')
ax.set_ylabel('Cancer Site')
plt.setp(ax.get_xticklabels(), rotation=45, ha='right', rotation_mode='anchor', fontsize=8)
plt.setp(ax.get_yticklabels(), rotation=0)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '03_comorb_heatmap_all.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-comorb-heatmap-all.png", dpi=300, bbox_inches="tight")
plt.show()
# Prepare data for heatmap
heatmap_data_deceased = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[(df_head_neck['cancer_site_group'] == site) & (df_head_neck['egresar_fallecido'] == 1)]
row = []
for col in comorbidity_cols:
prev = (df_site[col].mean() * 100) if len(df_site) > 0 else 0
row.append(prev)
heatmap_data_deceased.append(row)
heatmap_df_deceased = pd.DataFrame(heatmap_data_deceased,
index=sorted(df_head_neck['cancer_site_group'].unique()),
columns=[format_name(col) for col in comorbidity_cols])
fig, ax = plt.subplots(figsize=(22, 8))
sns.heatmap(heatmap_df_deceased, annot=True, fmt='.1f', cmap='viridis',
cbar_kws={'label': 'Prevalence (%)'}, ax=ax,
annot_kws={'rotation': 90, 'fontsize': 8})
ax.set_title('Comorbidity Prevalence by Cancer Site (Deceased Patients)')
ax.set_xlabel('Comorbidity')
ax.set_ylabel('Cancer Site')
plt.setp(ax.get_xticklabels(), rotation=45, ha='right', rotation_mode='anchor', fontsize=8)
plt.setp(ax.get_yticklabels(), rotation=0)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '04_comorb_heatmap_deceased.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-comorb-heatmap-deceased.png", dpi=300, bbox_inches="tight")
plt.show()
# Table: LOS by cancer site and discharge status
los_summary = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
for status in ['Discharged', 'Deceased']:
status_code = 0 if status == 'Discharged' else 1
df_status = df_site[df_site['egresar_fallecido'] == status_code]
if len(df_status) > 0:
los_summary.append({
'Cancer Site': site,
'Status': status,
'N': len(df_status),
'Mean': df_status['dias_estancia'].mean(),
'Median': df_status['dias_estancia'].median(),
'SD': df_status['dias_estancia'].std(),
'Min': df_status['dias_estancia'].min(),
'Max': df_status['dias_estancia'].max()
})
los_summary_df = pd.DataFrame(los_summary)
show_table(los_summary_df, "Table 17: Length of Stay by Cancer Site and Discharge Status", index=False)| Cancer Site | Status | N | Mean | Median | SD | Min | Max |
|---|---|---|---|---|---|---|---|
| Hypopharynx | Discharged | 367 | 12.88 | 7.00 | 17.16 | 0.00 | 171.00 |
| Hypopharynx | Deceased | 33 | 23.00 | 13.00 | 32.18 | 0.00 | 164.00 |
| Larynx | Discharged | 2427 | 13.67 | 7.00 | 20.08 | 0.00 | 317.00 |
| Larynx | Deceased | 198 | 24.63 | 12.50 | 35.82 | 0.00 | 250.00 |
| Nasopharynx | Discharged | 345 | 7.73 | 4.00 | 14.79 | 0.00 | 155.00 |
| Nasopharynx | Deceased | 23 | 27.48 | 13.00 | 41.43 | 1.00 | 138.00 |
| Oral Cavity | Discharged | 2825 | 9.22 | 4.00 | 14.98 | 0.00 | 216.00 |
| Oral Cavity | Deceased | 176 | 15.80 | 6.00 | 30.88 | 0.00 | 271.00 |
| Oropharynx | Discharged | 896 | 7.87 | 4.00 | 12.48 | 0.00 | 105.00 |
| Oropharynx | Deceased | 55 | 16.96 | 9.00 | 24.96 | 0.00 | 141.00 |
| Other Pharynx | Discharged | 251 | 10.04 | 5.00 | 14.70 | 0.00 | 96.00 |
| Other Pharynx | Deceased | 30 | 17.23 | 5.50 | 23.66 | 0.00 | 106.00 |
| Salivary Glands | Discharged | 852 | 6.41 | 3.00 | 10.75 | 0.00 | 136.00 |
| Salivary Glands | Deceased | 46 | 16.93 | 9.00 | 20.34 | 0.00 | 100.00 |
| Sinonasal | Discharged | 846 | 6.27 | 3.00 | 10.99 | 0.00 | 151.00 |
| Sinonasal | Deceased | 41 | 15.20 | 11.00 | 14.19 | 0.00 | 51.00 |
# Table: Mean LOS by cancer site and comorbidity presence
los_by_comorb = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row = {'Cancer Site': site}
for col in comorbidity_cols:
df_with = df_site[df_site[col] == 1]
los_with = df_with['dias_estancia'].mean() if len(df_with) > 0 else 0
row[format_name(col)] = f'{los_with:.1f}'
los_by_comorb.append(row)
los_by_comorb_df = pd.DataFrame(los_by_comorb)
show_table(los_by_comorb_df, "Table 18: Mean Length of Stay When Comorbidity Present (days)", index=False)| Cancer Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 15.1 | 8.0 | 19.1 | 18.3 | 15.3 | 8.0 | 45.0 | 4.0 | 14.6 | 1.0 | 10.2 | 14.2 | 30.0 | 19.0 | 19.6 | 17.4 | 0.0 | 41.1 | 32.2 | 0.0 | 108.0 | 20.7 | 21.8 | 32.3 | 0.0 | 12.2 | 18.0 | 0.0 | 14.0 | 0.0 | 22.3 | 0.0 | 0.0 | 15.0 | 18.5 | 25.4 | 5.5 | 33.2 | 0.0 | 13.8 | 24.3 | 19.4 | 0.0 | 12.7 | 48.0 | 5.5 | 0.0 |
| Larynx | 15.1 | 23.3 | 18.2 | 18.4 | 24.7 | 25.3 | 23.6 | 9.9 | 14.2 | 17.0 | 16.2 | 12.3 | 51.0 | 13.7 | 19.7 | 13.6 | 5.0 | 34.6 | 11.1 | 17.4 | 12.3 | 24.1 | 19.9 | 24.6 | 16.6 | 16.5 | 21.0 | 11.0 | 7.8 | 23.0 | 37.3 | 0.0 | 2.0 | 15.5 | 10.8 | 18.9 | 19.4 | 25.3 | 19.4 | 25.6 | 21.8 | 17.9 | 14.0 | 16.9 | 18.9 | 5.7 | 14.0 |
| Nasopharynx | 11.1 | 0.0 | 4.0 | 8.8 | 9.7 | 36.7 | 19.8 | 0.0 | 15.4 | 7.0 | 7.2 | 8.8 | 10.5 | 6.7 | 23.0 | 4.2 | 0.0 | 8.5 | 4.0 | 0.0 | 0.0 | 17.0 | 7.8 | 7.0 | 1.0 | 2.0 | 12.0 | 0.0 | 0.0 | 3.0 | 1.0 | 7.0 | 0.0 | 0.0 | 6.0 | 10.5 | 18.7 | 78.5 | 0.0 | 4.9 | 4.1 | 17.0 | 0.0 | 4.8 | 0.0 | 0.0 | 0.0 |
| Oral Cavity | 10.0 | 12.7 | 11.7 | 12.2 | 15.9 | 10.9 | 17.9 | 11.3 | 10.4 | 6.0 | 12.7 | 9.1 | 16.9 | 10.4 | 12.8 | 10.7 | 24.5 | 20.5 | 21.4 | 8.4 | 38.1 | 17.6 | 13.9 | 14.9 | 9.3 | 11.5 | 13.2 | 16.3 | 6.2 | 4.8 | 13.6 | 6.0 | 8.0 | 9.7 | 12.6 | 12.2 | 9.6 | 16.2 | 21.3 | 25.3 | 13.2 | 12.7 | 0.0 | 17.5 | 4.5 | 0.0 | 6.2 |
| Oropharynx | 9.7 | 12.0 | 14.8 | 18.3 | 14.7 | 13.2 | 19.3 | 7.0 | 9.0 | 7.3 | 8.6 | 6.9 | 27.0 | 8.5 | 20.9 | 17.3 | 12.0 | 15.2 | 21.7 | 5.0 | 11.0 | 18.7 | 8.2 | 10.2 | 6.0 | 25.7 | 28.8 | 0.0 | 0.0 | 12.0 | 26.3 | 1.0 | 0.0 | 37.0 | 13.3 | 12.3 | 12.9 | 21.6 | 21.5 | 15.6 | 12.5 | 11.8 | 55.0 | 5.2 | 44.5 | 0.0 | 0.0 |
| Other Pharynx | 11.5 | 26.0 | 16.7 | 18.1 | 28.7 | 17.7 | 24.8 | 19.0 | 13.1 | 0.0 | 8.8 | 7.6 | 0.0 | 12.3 | 14.6 | 2.7 | 0.0 | 24.2 | 6.0 | 0.0 | 5.8 | 13.7 | 18.5 | 10.2 | 0.0 | 6.7 | 4.0 | 0.0 | 0.0 | 5.7 | 10.6 | 2.0 | 7.0 | 26.0 | 0.0 | 17.2 | 13.4 | 15.2 | 6.0 | 36.0 | 22.2 | 16.5 | 0.0 | 3.5 | 12.0 | 0.0 | 0.0 |
| Salivary Glands | 7.8 | 12.8 | 13.5 | 12.6 | 17.2 | 13.5 | 11.2 | 7.0 | 7.3 | 13.5 | 6.0 | 6.3 | 3.7 | 6.4 | 11.1 | 7.1 | 0.0 | 18.5 | 14.0 | 20.0 | 6.9 | 15.5 | 10.8 | 6.4 | 6.0 | 13.1 | 19.5 | 8.0 | 0.0 | 5.0 | 19.0 | 2.0 | 0.0 | 10.6 | 12.8 | 4.9 | 12.0 | 10.2 | 2.0 | 2.3 | 7.8 | 9.7 | 0.0 | 10.6 | 4.8 | 0.0 | 0.0 |
| Sinonasal | 8.2 | 17.6 | 6.9 | 15.1 | 18.3 | 16.2 | 16.4 | 14.5 | 7.7 | 0.0 | 7.8 | 6.1 | 16.0 | 6.7 | 6.8 | 6.8 | 0.0 | 0.0 | 8.6 | 0.0 | 7.2 | 38.1 | 12.0 | 11.3 | 2.0 | 10.1 | 10.5 | 0.0 | 0.0 | 10.3 | 1.3 | 0.0 | 0.0 | 7.1 | 3.8 | 9.3 | 9.1 | 7.2 | 61.0 | 6.6 | 9.8 | 6.9 | 0.0 | 3.5 | 0.0 | 0.0 | 0.0 |
# Table: LOS impact (difference WITH - WITHOUT comorbidity)
los_impact = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row = {'Cancer Site': site}
for col in comorbidity_cols:
df_with = df_site[df_site[col] == 1]
df_without = df_site[df_site[col] == 0]
los_with = df_with['dias_estancia'].mean() if len(df_with) > 0 else 0
los_without = df_without['dias_estancia'].mean() if len(df_without) > 0 else 0
difference = los_with - los_without
row[format_name(col)] = f'{difference:+.1f}'
los_impact.append(row)
los_impact_df = pd.DataFrame(los_impact)
show_table(los_impact_df, "Table 19: Length of Stay Impact by Comorbidity (Days Difference: WITH - WITHOUT)", index=False)| Cancer Site | CV Essential Hypertension | CV Complicated Hypertension | CV Coronary Disease | CV Heart Failure | CV Arrhythmias | CV Cerebrovascular Disease | CV Peripheral Vascular Disease | MET Type1 Diabetes | MET Type2 Diabetes | MET Complicated Diabetes | MET Obesity | MET Dyslipidemia | MET Metabolic Syndrome | MET Thyroid Disorders | RESP COPD | RESP Asthma | RESP Chronic Bronchitis | RESP Emphysema | RESP Pulmonary Fibrosis | RESP Pulmonary Hypertension | RESP Sleep Apnea | REN Acute Renal Failure | REN Chronic Renal Failure | REN End Stage Renal | REN Diabetic Nephropathy | HEP Steatosis | HEP Alcoholic Cirrhosis | HEP NonAlcoholic Cirrhosis | HEP Viral Hepatitis | GI Reflux | GI Peptic Ulcer | GI IBD | GI IBS | NEURO Dementia | NEURO Parkinson | NEURO Epilepsy | NEURO Stroke Sequelae | PSI Major Depression | PSI Bipolar | PSI Anxiety | SUST Alcohol | SUST Tobacco | SUST Opioids | INF HIV | INF Tuberculosis | INF Hepatitis C | INF Hepatitis B |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | +2.2 | -5.7 | +5.6 | +4.7 | +1.7 | -5.8 | +31.8 | -9.7 | +1.0 | -12.7 | -3.6 | +0.6 | +16.3 | +6.3 | +6.3 | +3.7 | -13.7 | +28.0 | +18.7 | -13.7 | +94.5 | +7.1 | +8.5 | +18.9 | -13.7 | -1.5 | +4.3 | -13.7 | +0.3 | -13.7 | +8.7 | -13.7 | -13.7 | +1.3 | +4.8 | +11.9 | -8.3 | +19.9 | -13.7 | +0.0 | +11.3 | +6.8 | -13.7 | -1.1 | +34.5 | -8.3 | -13.7 |
| Larynx | +1.2 | +8.9 | +4.0 | +4.0 | +11.0 | +11.1 | +9.3 | -4.6 | -0.4 | +2.5 | +1.8 | -2.3 | +36.7 | -0.9 | +6.0 | -0.9 | -9.5 | +20.5 | -3.4 | +2.9 | -2.2 | +10.2 | +5.7 | +10.2 | +2.1 | +2.0 | +6.6 | -3.5 | -6.8 | +8.6 | +23.1 | -14.5 | -12.5 | +1.1 | -3.7 | +4.5 | +5.0 | +11.1 | +4.9 | +11.2 | +7.7 | +4.5 | -0.5 | +2.4 | +4.4 | -8.8 | -0.5 |
| Nasopharynx | +3.0 | -9.0 | -5.1 | -0.1 | +0.8 | +28.2 | +11.3 | -9.0 | +7.4 | -2.0 | -2.0 | -0.1 | +1.5 | -2.5 | +14.1 | -5.0 | -9.0 | -0.5 | -5.0 | -9.0 | -9.0 | +8.5 | -1.2 | -2.0 | -8.0 | -7.0 | +3.0 | -9.0 | -9.0 | -6.0 | -8.0 | -2.0 | -9.0 | -9.0 | -3.0 | +1.5 | +9.8 | +69.9 | -9.0 | -4.3 | -5.1 | +8.8 | -9.0 | -4.2 | -9.0 | -9.0 | -9.0 |
| Oral Cavity | +0.6 | +3.1 | +2.2 | +2.7 | +6.6 | +1.3 | +8.4 | +1.7 | +0.9 | -3.6 | +3.2 | -0.6 | +7.3 | +0.9 | +3.3 | +1.1 | +14.9 | +11.0 | +11.9 | -1.2 | +28.6 | +8.3 | +4.5 | +5.3 | -0.3 | +1.9 | +3.6 | +6.7 | -3.4 | -4.8 | +4.0 | -3.6 | -1.6 | +0.1 | +3.0 | +2.6 | -0.0 | +6.8 | +11.7 | +15.8 | +3.7 | +3.7 | -9.6 | +7.9 | -5.1 | -9.6 | -3.4 |
| Oropharynx | +2.0 | +3.6 | +6.5 | +10.0 | +6.5 | +4.9 | +11.2 | -1.4 | +0.7 | -1.1 | +0.2 | -1.6 | +18.6 | +0.1 | +13.0 | +9.1 | +3.6 | +6.9 | +13.4 | -3.4 | +2.6 | +10.7 | -0.1 | +1.8 | -2.4 | +17.6 | +20.5 | -8.4 | -8.4 | +3.6 | +18.0 | -7.4 | -8.4 | +28.7 | +5.0 | +4.0 | +4.6 | +13.5 | +13.2 | +7.3 | +4.3 | +3.9 | +46.7 | -3.3 | +36.2 | -8.4 | -8.4 |
| Other Pharynx | +1.0 | +15.2 | +6.0 | +7.5 | +18.8 | +7.2 | +14.6 | +8.3 | +2.7 | -10.8 | -2.1 | -3.4 | -10.8 | +1.6 | +4.0 | -8.2 | -10.8 | +13.6 | -4.9 | -10.8 | -5.2 | +3.0 | +8.0 | -0.6 | -10.8 | -4.2 | -6.8 | -10.8 | -10.8 | -5.2 | -0.2 | -8.8 | -3.8 | +15.3 | -10.8 | +6.6 | +2.6 | +4.5 | -4.8 | +25.4 | +11.6 | +6.4 | -10.8 | -7.4 | +1.2 | -10.8 | -10.8 |
| Salivary Glands | +1.3 | +5.9 | +6.8 | +5.8 | +10.7 | +6.7 | +4.3 | +0.1 | +0.5 | +6.6 | -1.0 | -0.7 | -3.3 | -0.6 | +4.3 | +0.1 | -6.9 | +11.6 | +7.1 | +13.1 | -0.1 | +8.9 | +4.0 | -0.5 | -1.0 | +6.2 | +12.6 | +1.1 | -6.9 | -2.0 | +12.1 | -5.0 | -6.9 | +3.6 | +5.9 | -2.1 | +5.1 | +3.4 | -5.0 | -4.7 | +0.9 | +3.0 | -6.9 | +3.7 | -2.2 | -6.9 | -6.9 |
| Sinonasal | +2.2 | +11.0 | +0.2 | +8.5 | +11.9 | +9.7 | +9.8 | +7.9 | +1.3 | -6.7 | +1.1 | -0.6 | +9.3 | +0.0 | +0.1 | +0.1 | -6.7 | -6.7 | +1.9 | -6.7 | +0.5 | +32.0 | +5.4 | +4.6 | -4.7 | +3.5 | +3.8 | -6.7 | -6.7 | +3.7 | -5.4 | -6.7 | -6.7 | +0.4 | -2.9 | +2.6 | +2.4 | +0.5 | +54.4 | -0.0 | +3.3 | +0.2 | -6.7 | -3.2 | -6.7 | -6.7 | -6.7 |
fig, axes = plt.subplots(1, 2, figsize=(12, 6))
# Box plot
ax = axes[0]
data_discharged = []
data_deceased = []
labels = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
data_discharged.append(df_site[df_site['egresar_fallecido'] == 0]['dias_estancia'].values)
data_deceased.append(df_site[df_site['egresar_fallecido'] == 1]['dias_estancia'].values)
labels.append(site)
positions_discharged = np.arange(len(labels)) * 2.5
positions_deceased = positions_discharged + 1
bp1 = ax.boxplot(data_discharged, positions=positions_discharged, widths=0.8, patch_artist=True, label='Discharged')
bp2 = ax.boxplot(data_deceased, positions=positions_deceased, widths=0.8, patch_artist=True, label='Deceased')
for patch in bp1['boxes']:
patch.set_facecolor('lightgreen')
for patch in bp2['boxes']:
patch.set_facecolor('lightcoral')
ax.set_xticks((positions_discharged + positions_deceased) / 2)
ax.set_xticklabels(labels, rotation=30, ha='right', fontsize=9)
ax.set_ylabel('Length of Stay (days)')
ax.set_title('LOS Distribution by Site and Discharge Status')
ax.set_ylim([0, 100])
ax.legend()
ax.grid(True, alpha=0.3, axis='y')
# Bar plot of means
ax = axes[1]
means_discharged = []
means_deceased = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
means_discharged.append(df_site[df_site['egresar_fallecido'] == 0]['dias_estancia'].mean())
means_deceased.append(df_site[df_site['egresar_fallecido'] == 1]['dias_estancia'].mean())
x = np.arange(len(labels))
width = 0.35
ax.bar(x - width/2, means_discharged, width, label='Discharged', color='lightgreen')
ax.bar(x + width/2, means_deceased, width, label='Deceased', color='lightcoral')
ax.set_ylabel('Mean Length of Stay (days)')
ax.set_xlabel('Cancer Site')
ax.set_title('Mean LOS by Site and Discharge Status')
ax.set_xticks(x)
ax.set_xticklabels(labels, rotation=30, ha='right', fontsize=9)
ax.legend()
ax.grid(True, alpha=0.3, axis='y')
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '05_los_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-los-by-site.png", dpi=300, bbox_inches="tight")
plt.show()
heatmap_los_impact = []
for site in sorted(df_head_neck['cancer_site_group'].unique()):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row = []
for col in comorbidity_cols:
df_with = df_site[df_site[col] == 1]
df_without = df_site[df_site[col] == 0]
los_with = df_with['dias_estancia'].mean() if len(df_with) > 0 else 0
los_without = df_without['dias_estancia'].mean() if len(df_without) > 0 else 0
difference = los_with - los_without
row.append(difference)
heatmap_los_impact.append(row)
heatmap_los_impact_df = pd.DataFrame(heatmap_los_impact,
index=sorted(df_head_neck['cancer_site_group'].unique()),
columns=[format_name(col) for col in comorbidity_cols])
fig, ax = plt.subplots(figsize=(22, 8))
sns.heatmap(heatmap_los_impact_df, annot=True, fmt='.1f', cmap='RdBu_r', center=0,
cbar_kws={'label': 'LOS Difference (days)'}, ax=ax,
annot_kws={'rotation': 90, 'fontsize': 8})
ax.set_title('Length of Stay Impact: WITH - WITHOUT Comorbidity (days)')
ax.set_xlabel('Comorbidity')
ax.set_ylabel('Cancer Site')
plt.setp(ax.get_xticklabels(), rotation=45, ha='right', rotation_mode='anchor', fontsize=8)
plt.setp(ax.get_yticklabels(), rotation=0)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '06_los_impact_heatmap.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-los-impact-heatmap.png", dpi=300, bbox_inches="tight")
plt.show()
fig, axes = plt.subplots(2, 2, figsize=(14, 8))
sites = sorted(df_head_neck['cancer_site_group'].unique())
colors = ['steelblue', 'forestgreen', 'coral', 'mediumpurple']
# Row 1: Cases by year per site
ax = axes[0, 0]
for site, color in zip(sites, colors):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
cases_by_year = df_site['year'].value_counts().sort_index()
ax.plot(cases_by_year.index, cases_by_year.values, marker='o', linewidth=2, label=site, color=color)
ax.set_xlabel('Year')
ax.set_ylabel('Number of Cases')
ax.set_title('Trend: Cases by Year')
ax.legend()
ax.grid(True, alpha=0.3)
# Row 1: Mortality rate by year per site
ax = axes[0, 1]
for site, color in zip(sites, colors):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
mortality_by_year = df_site.groupby('year')['egresar_fallecido'].mean() * 100
ax.plot(mortality_by_year.index, mortality_by_year.values, marker='o', linewidth=2, label=site, color=color)
ax.set_xlabel('Year')
ax.set_ylabel('Mortality Rate (%)')
ax.set_title('Trend: Mortality Rate by Year')
ax.legend()
ax.grid(True, alpha=0.3)
# Row 2: Mean LOS by year per site
ax = axes[1, 0]
for site, color in zip(sites, colors):
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
los_by_year = df_site.groupby('year')['dias_estancia'].mean()
ax.plot(los_by_year.index, los_by_year.values, marker='o', linewidth=2, label=site, color=color)
ax.set_xlabel('Year')
ax.set_ylabel('Mean Length of Stay (days)')
ax.set_title('Trend: Mean LOS by Year')
ax.legend()
ax.grid(True, alpha=0.3)
# Row 2: Overall mortality trend
ax = axes[1, 1]
overall_mortality = df_head_neck.groupby('year')['egresar_fallecido'].mean() * 100
ax.plot(overall_mortality.index, overall_mortality.values, marker='o', linewidth=3, markersize=10, color='darkred')
ax.fill_between(overall_mortality.index, overall_mortality.values, alpha=0.3, color='darkred')
ax.set_xlabel('Year')
ax.set_ylabel('Mortality Rate (%)')
ax.set_title('Trend: Overall Mortality Rate')
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, '07_temporal_trends.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-temporal-trends.png", dpi=300, bbox_inches="tight")
plt.show()
from scipy.stats import chi2_contingency, ranksums
# Bivariate analysis: chi-square for categorical association with mortality
all_site_names = sorted(df_head_neck['cancer_site_group'].unique())
df_bivariate = []
for site in all_site_names:
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
if len(df_site) < 30:
continue
for col in comorbidity_cols:
# Chi-square test for mortality association
contingency = pd.crosstab(df_site[col], df_site['egresar_fallecido'])
try:
chi2, pval_mort, dof, expected = chi2_contingency(contingency)
except:
pval_mort = np.nan
# Wilcoxon rank-sum test (Mann-Whitney U) for LOS difference
group_with = df_site[df_site[col] == 1]['dias_estancia'].dropna()
group_without = df_site[df_site[col] == 0]['dias_estancia'].dropna()
try:
if len(group_with) >= 5 and len(group_without) >= 5:
stat, pval_los = ranksums(group_with, group_without)
else:
pval_los = np.nan
except:
pval_los = np.nan
df_bivariate.append({
'Cancer Site': site,
'Comorbidity': format_name(col),
'Chi-square p-value': pval_mort,
'LOS p-value (Wilcoxon)': pval_los,
'N with': len(group_with),
'N without': len(group_without)
})
df_bivariate = pd.DataFrame(df_bivariate)
show_table(df_bivariate, "Table 20: Bivariate Analysis by Cancer Site", index=False)| Cancer Site | Comorbidity | Chi-square p-value | LOS p-value (Wilcoxon) | N with | N without |
|---|---|---|---|---|---|
| Hypopharynx | CV Essential Hypertension | 1.00 | 1.00 | 147 | 253 |
| Hypopharynx | CV Complicated Hypertension | 1.00 | — | 1 | 399 |
| Hypopharynx | CV Coronary Disease | 0.27 | 0.55 | 16 | 384 |
| Hypopharynx | CV Heart Failure | 0.13 | 0.44 | 6 | 394 |
| Hypopharynx | CV Arrhythmias | 0.00 | 0.63 | 15 | 385 |
| Hypopharynx | CV Cerebrovascular Disease | 0.59 | — | 3 | 397 |
| Hypopharynx | CV Peripheral Vascular Disease | 1.00 | 0.04 | 6 | 394 |
| Hypopharynx | MET Type1 Diabetes | 1.00 | — | 1 | 399 |
| Hypopharynx | MET Type2 Diabetes | 0.92 | 0.67 | 57 | 343 |
| Hypopharynx | MET Complicated Diabetes | 1.00 | — | 1 | 399 |
| Hypopharynx | MET Obesity | 1.00 | 0.95 | 6 | 394 |
| Hypopharynx | MET Dyslipidemia | 1.00 | 0.82 | 38 | 362 |
| Hypopharynx | MET Metabolic Syndrome | 1.00 | — | 1 | 399 |
| Hypopharynx | MET Thyroid Disorders | 0.55 | 0.41 | 64 | 336 |
| Hypopharynx | RESP COPD | 0.28 | 0.00 | 25 | 375 |
| Hypopharynx | RESP Asthma | 0.89 | 0.26 | 5 | 395 |
| Hypopharynx | RESP Chronic Bronchitis | 1.00 | — | 0 | 400 |
| Hypopharynx | RESP Emphysema | 1.00 | 0.01 | 8 | 392 |
| Hypopharynx | RESP Pulmonary Fibrosis | 1.00 | — | 4 | 396 |
| Hypopharynx | RESP Pulmonary Hypertension | 1.00 | — | 0 | 400 |
| Hypopharynx | RESP Sleep Apnea | 1.00 | — | 1 | 399 |
| Hypopharynx | REN Acute Renal Failure | 0.99 | 0.05 | 6 | 394 |
| Hypopharynx | REN Chronic Renal Failure | 0.39 | 0.04 | 18 | 382 |
| Hypopharynx | REN End Stage Renal | 0.91 | 0.02 | 7 | 393 |
| Hypopharynx | REN Diabetic Nephropathy | 1.00 | — | 0 | 400 |
| Hypopharynx | HEP Steatosis | 1.00 | — | 4 | 396 |
| Hypopharynx | HEP Alcoholic Cirrhosis | 0.76 | — | 4 | 396 |
| Hypopharynx | HEP NonAlcoholic Cirrhosis | 1.00 | — | 0 | 400 |
| Hypopharynx | HEP Viral Hepatitis | 1.00 | — | 3 | 397 |
| Hypopharynx | GI Reflux | 1.00 | — | 0 | 400 |
| Hypopharynx | GI Peptic Ulcer | 0.59 | — | 3 | 397 |
| Hypopharynx | GI IBD | 1.00 | — | 0 | 400 |
| Hypopharynx | GI IBS | 1.00 | — | 0 | 400 |
| Hypopharynx | NEURO Dementia | 1.00 | 0.44 | 5 | 395 |
| Hypopharynx | NEURO Parkinson | 0.39 | — | 2 | 398 |
| Hypopharynx | NEURO Epilepsy | 1.00 | 0.02 | 7 | 393 |
| Hypopharynx | NEURO Stroke Sequelae | 1.00 | — | 2 | 398 |
| Hypopharynx | PSI Major Depression | 1.00 | 0.24 | 8 | 392 |
| Hypopharynx | PSI Bipolar | 1.00 | — | 0 | 400 |
| Hypopharynx | PSI Anxiety | 0.76 | — | 4 | 396 |
| Hypopharynx | SUST Alcohol | 0.23 | 0.00 | 26 | 374 |
| Hypopharynx | SUST Tobacco | 0.62 | 0.02 | 67 | 333 |
| Hypopharynx | SUST Opioids | 1.00 | — | 0 | 400 |
| Hypopharynx | INF HIV | 1.00 | 0.84 | 6 | 394 |
| Hypopharynx | INF Tuberculosis | 0.39 | — | 2 | 398 |
| Hypopharynx | INF Hepatitis C | 1.00 | — | 2 | 398 |
| Hypopharynx | INF Hepatitis B | 1.00 | — | 0 | 400 |
| Larynx | CV Essential Hypertension | 1.00 | 0.00 | 1202 | 1423 |
| Larynx | CV Complicated Hypertension | 0.58 | 0.01 | 13 | 2612 |
| Larynx | CV Coronary Disease | 0.34 | 0.03 | 152 | 2473 |
| Larynx | CV Heart Failure | 0.00 | 0.00 | 92 | 2533 |
| Larynx | CV Arrhythmias | 0.00 | 0.00 | 182 | 2443 |
| Larynx | CV Cerebrovascular Disease | 0.11 | 0.00 | 77 | 2548 |
| Larynx | CV Peripheral Vascular Disease | 0.00 | 0.00 | 81 | 2544 |
| Larynx | MET Type1 Diabetes | 0.82 | 0.95 | 9 | 2616 |
| Larynx | MET Type2 Diabetes | 0.12 | 0.77 | 520 | 2105 |
| Larynx | MET Complicated Diabetes | 1.00 | 0.23 | 6 | 2619 |
| Larynx | MET Obesity | 0.07 | 0.14 | 131 | 2494 |
| Larynx | MET Dyslipidemia | 0.27 | 0.38 | 221 | 2404 |
| Larynx | MET Metabolic Syndrome | 0.08 | 0.00 | 12 | 2613 |
| Larynx | MET Thyroid Disorders | 0.74 | 0.54 | 293 | 2332 |
| Larynx | RESP COPD | 0.05 | 0.00 | 362 | 2263 |
| Larynx | RESP Asthma | 0.79 | 0.99 | 95 | 2530 |
| Larynx | RESP Chronic Bronchitis | 1.00 | — | 1 | 2624 |
| Larynx | RESP Emphysema | 0.00 | 0.00 | 46 | 2579 |
| Larynx | RESP Pulmonary Fibrosis | 0.07 | 0.84 | 27 | 2598 |
| Larynx | RESP Pulmonary Hypertension | 0.08 | 0.12 | 12 | 2613 |
| Larynx | RESP Sleep Apnea | 1.00 | 0.71 | 26 | 2599 |
| Larynx | REN Acute Renal Failure | 0.00 | 0.00 | 160 | 2465 |
| Larynx | REN Chronic Renal Failure | 0.00 | 0.01 | 136 | 2489 |
| Larynx | REN End Stage Renal | 0.02 | 0.46 | 29 | 2596 |
| Larynx | REN Diabetic Nephropathy | 1.00 | 0.86 | 8 | 2617 |
| Larynx | HEP Steatosis | 0.19 | 0.72 | 24 | 2601 |
| Larynx | HEP Alcoholic Cirrhosis | 0.00 | 0.00 | 34 | 2591 |
| Larynx | HEP NonAlcoholic Cirrhosis | 0.00 | — | 2 | 2623 |
| Larynx | HEP Viral Hepatitis | 1.00 | — | 4 | 2621 |
| Larynx | GI Reflux | 0.54 | 0.17 | 15 | 2610 |
| Larynx | GI Peptic Ulcer | 0.00 | 0.00 | 30 | 2595 |
| Larynx | GI IBD | 1.00 | — | 0 | 2625 |
| Larynx | GI IBS | 0.11 | — | 1 | 2624 |
| Larynx | NEURO Dementia | 0.58 | 0.44 | 35 | 2590 |
| Larynx | NEURO Parkinson | 1.00 | 0.53 | 19 | 2606 |
| Larynx | NEURO Epilepsy | 0.15 | 0.01 | 41 | 2584 |
| Larynx | NEURO Stroke Sequelae | 0.49 | 0.01 | 55 | 2570 |
| Larynx | PSI Major Depression | 0.35 | 0.01 | 61 | 2564 |
| Larynx | PSI Bipolar | 0.97 | 0.28 | 7 | 2618 |
| Larynx | PSI Anxiety | 0.33 | 0.00 | 23 | 2602 |
| Larynx | SUST Alcohol | 0.37 | 0.00 | 119 | 2506 |
| Larynx | SUST Tobacco | 0.14 | 0.00 | 637 | 1988 |
| Larynx | SUST Opioids | 1.00 | — | 1 | 2624 |
| Larynx | INF HIV | 1.00 | 0.14 | 17 | 2608 |
| Larynx | INF Tuberculosis | 1.00 | 0.38 | 7 | 2618 |
| Larynx | INF Hepatitis C | 1.00 | — | 3 | 2622 |
| Larynx | INF Hepatitis B | 1.00 | — | 1 | 2624 |
| Nasopharynx | CV Essential Hypertension | 0.63 | 0.21 | 104 | 264 |
| Nasopharynx | CV Complicated Hypertension | 1.00 | — | 0 | 368 |
| Nasopharynx | CV Coronary Disease | 1.00 | 0.71 | 8 | 360 |
| Nasopharynx | CV Heart Failure | 1.00 | 0.30 | 6 | 362 |
| Nasopharynx | CV Arrhythmias | 0.30 | 0.01 | 11 | 357 |
| Nasopharynx | CV Cerebrovascular Disease | 0.06 | 0.00 | 6 | 362 |
| Nasopharynx | CV Peripheral Vascular Disease | 0.09 | 0.01 | 15 | 353 |
| Nasopharynx | MET Type1 Diabetes | 1.00 | — | 0 | 368 |
| Nasopharynx | MET Type2 Diabetes | 1.00 | 0.02 | 46 | 322 |
| Nasopharynx | MET Complicated Diabetes | 1.00 | — | 1 | 367 |
| Nasopharynx | MET Obesity | 0.34 | 0.38 | 26 | 342 |
| Nasopharynx | MET Dyslipidemia | 0.24 | 0.12 | 33 | 335 |
| Nasopharynx | MET Metabolic Syndrome | 0.27 | — | 2 | 366 |
| Nasopharynx | MET Thyroid Disorders | 0.82 | 0.21 | 35 | 333 |
| Nasopharynx | RESP COPD | 1.00 | — | 3 | 365 |
| Nasopharynx | RESP Asthma | 0.72 | 0.90 | 13 | 355 |
| Nasopharynx | RESP Chronic Bronchitis | 1.00 | — | 0 | 368 |
| Nasopharynx | RESP Emphysema | 0.27 | — | 2 | 366 |
| Nasopharynx | RESP Pulmonary Fibrosis | 1.00 | — | 4 | 364 |
| Nasopharynx | RESP Pulmonary Hypertension | 1.00 | — | 0 | 368 |
| Nasopharynx | RESP Sleep Apnea | 1.00 | — | 1 | 367 |
| Nasopharynx | REN Acute Renal Failure | 0.03 | 0.01 | 20 | 348 |
| Nasopharynx | REN Chronic Renal Failure | 0.76 | 0.20 | 12 | 356 |
| Nasopharynx | REN End Stage Renal | 1.00 | — | 2 | 366 |
| Nasopharynx | REN Diabetic Nephropathy | 1.00 | — | 1 | 367 |
| Nasopharynx | HEP Steatosis | 1.00 | — | 1 | 367 |
| Nasopharynx | HEP Alcoholic Cirrhosis | 1.00 | — | 1 | 367 |
| Nasopharynx | HEP NonAlcoholic Cirrhosis | 1.00 | — | 0 | 368 |
| Nasopharynx | HEP Viral Hepatitis | 1.00 | — | 0 | 368 |
| Nasopharynx | GI Reflux | 1.00 | — | 1 | 367 |
| Nasopharynx | GI Peptic Ulcer | 1.00 | — | 1 | 367 |
| Nasopharynx | GI IBD | 1.00 | — | 1 | 367 |
| Nasopharynx | GI IBS | 1.00 | — | 0 | 368 |
| Nasopharynx | NEURO Dementia | 1.00 | — | 0 | 368 |
| Nasopharynx | NEURO Parkinson | 0.07 | — | 1 | 367 |
| Nasopharynx | NEURO Epilepsy | 1.00 | — | 4 | 364 |
| Nasopharynx | NEURO Stroke Sequelae | 1.00 | — | 3 | 365 |
| Nasopharynx | PSI Major Depression | 1.00 | — | 2 | 366 |
| Nasopharynx | PSI Bipolar | 1.00 | — | 0 | 368 |
| Nasopharynx | PSI Anxiety | 0.67 | 0.97 | 14 | 354 |
| Nasopharynx | SUST Alcohol | 0.63 | 0.03 | 15 | 353 |
| Nasopharynx | SUST Tobacco | 0.18 | 0.41 | 29 | 339 |
| Nasopharynx | SUST Opioids | 1.00 | — | 0 | 368 |
| Nasopharynx | INF HIV | 1.00 | 0.94 | 5 | 363 |
| Nasopharynx | INF Tuberculosis | 1.00 | — | 0 | 368 |
| Nasopharynx | INF Hepatitis C | 1.00 | — | 0 | 368 |
| Nasopharynx | INF Hepatitis B | 1.00 | — | 0 | 368 |
| Oral Cavity | CV Essential Hypertension | 0.32 | 0.00 | 1316 | 1685 |
| Oral Cavity | CV Complicated Hypertension | 0.49 | 0.49 | 15 | 2986 |
| Oral Cavity | CV Coronary Disease | 0.25 | 0.00 | 114 | 2887 |
| Oral Cavity | CV Heart Failure | 0.05 | 0.00 | 65 | 2936 |
| Oral Cavity | CV Arrhythmias | 0.00 | 0.00 | 169 | 2832 |
| Oral Cavity | CV Cerebrovascular Disease | 0.03 | 0.01 | 73 | 2928 |
| Oral Cavity | CV Peripheral Vascular Disease | 0.00 | 0.00 | 63 | 2938 |
| Oral Cavity | MET Type1 Diabetes | 1.00 | 0.90 | 11 | 2990 |
| Oral Cavity | MET Type2 Diabetes | 0.90 | 0.14 | 560 | 2441 |
| Oral Cavity | MET Complicated Diabetes | 0.76 | 0.40 | 13 | 2988 |
| Oral Cavity | MET Obesity | 0.20 | 0.00 | 156 | 2845 |
| Oral Cavity | MET Dyslipidemia | 0.52 | 0.73 | 343 | 2658 |
| Oral Cavity | MET Metabolic Syndrome | 0.01 | 0.04 | 18 | 2983 |
| Oral Cavity | MET Thyroid Disorders | 0.06 | 0.01 | 368 | 2633 |
| Oral Cavity | RESP COPD | 0.38 | 0.00 | 154 | 2847 |
| Oral Cavity | RESP Asthma | 0.05 | 0.82 | 53 | 2948 |
| Oral Cavity | RESP Chronic Bronchitis | 1.00 | — | 4 | 2997 |
| Oral Cavity | RESP Emphysema | 0.85 | 0.00 | 22 | 2979 |
| Oral Cavity | RESP Pulmonary Fibrosis | 0.00 | 0.94 | 22 | 2979 |
| Oral Cavity | RESP Pulmonary Hypertension | 0.12 | 0.16 | 8 | 2993 |
| Oral Cavity | RESP Sleep Apnea | 1.00 | 0.00 | 9 | 2992 |
| Oral Cavity | REN Acute Renal Failure | 0.00 | 0.00 | 136 | 2865 |
| Oral Cavity | REN Chronic Renal Failure | 0.01 | 0.00 | 137 | 2864 |
| Oral Cavity | REN End Stage Renal | 0.04 | 0.00 | 31 | 2970 |
| Oral Cavity | REN Diabetic Nephropathy | 1.00 | 0.23 | 7 | 2994 |
| Oral Cavity | HEP Steatosis | 0.01 | 0.13 | 41 | 2960 |
| Oral Cavity | HEP Alcoholic Cirrhosis | 0.17 | 0.11 | 19 | 2982 |
| Oral Cavity | HEP NonAlcoholic Cirrhosis | 1.00 | 0.50 | 9 | 2992 |
| Oral Cavity | HEP Viral Hepatitis | 1.00 | 0.46 | 5 | 2996 |
| Oral Cavity | GI Reflux | 0.76 | 0.09 | 13 | 2988 |
| Oral Cavity | GI Peptic Ulcer | 0.55 | 0.01 | 16 | 2985 |
| Oral Cavity | GI IBD | 0.57 | — | 4 | 2997 |
| Oral Cavity | GI IBS | 1.00 | — | 2 | 2999 |
| Oral Cavity | NEURO Dementia | 0.14 | 0.95 | 52 | 2949 |
| Oral Cavity | NEURO Parkinson | 1.00 | 0.55 | 20 | 2981 |
| Oral Cavity | NEURO Epilepsy | 0.78 | 0.04 | 36 | 2965 |
| Oral Cavity | NEURO Stroke Sequelae | 0.07 | 0.03 | 57 | 2944 |
| Oral Cavity | PSI Major Depression | 0.90 | 0.01 | 89 | 2912 |
| Oral Cavity | PSI Bipolar | 0.89 | 0.69 | 7 | 2994 |
| Oral Cavity | PSI Anxiety | 1.00 | 0.00 | 21 | 2980 |
| Oral Cavity | SUST Alcohol | 0.68 | 0.00 | 143 | 2858 |
| Oral Cavity | SUST Tobacco | 0.60 | 0.00 | 445 | 2556 |
| Oral Cavity | SUST Opioids | 1.00 | — | 0 | 3001 |
| Oral Cavity | INF HIV | 0.85 | 0.01 | 22 | 2979 |
| Oral Cavity | INF Tuberculosis | 1.00 | — | 2 | 2999 |
| Oral Cavity | INF Hepatitis C | 1.00 | — | 0 | 3001 |
| Oral Cavity | INF Hepatitis B | 1.00 | 0.46 | 5 | 2996 |
| Oropharynx | CV Essential Hypertension | 0.74 | 0.00 | 352 | 599 |
| Oropharynx | CV Complicated Hypertension | 1.00 | — | 1 | 950 |
| Oropharynx | CV Coronary Disease | 1.00 | 0.02 | 23 | 928 |
| Oropharynx | CV Heart Failure | 0.01 | 0.13 | 10 | 941 |
| Oropharynx | CV Arrhythmias | 0.00 | 0.00 | 30 | 921 |
| Oropharynx | CV Cerebrovascular Disease | 0.26 | 0.04 | 22 | 929 |
| Oropharynx | CV Peripheral Vascular Disease | 1.00 | 0.00 | 26 | 925 |
| Oropharynx | MET Type1 Diabetes | 1.00 | — | 1 | 950 |
| Oropharynx | MET Type2 Diabetes | 0.42 | 0.13 | 130 | 821 |
| Oropharynx | MET Complicated Diabetes | 1.00 | — | 3 | 948 |
| Oropharynx | MET Obesity | 1.00 | 0.23 | 40 | 911 |
| Oropharynx | MET Dyslipidemia | 0.84 | 0.47 | 67 | 884 |
| Oropharynx | MET Metabolic Syndrome | 0.06 | — | 1 | 950 |
| Oropharynx | MET Thyroid Disorders | 0.26 | 0.58 | 83 | 868 |
| Oropharynx | RESP COPD | 0.01 | 0.00 | 35 | 916 |
| Oropharynx | RESP Asthma | 1.00 | 0.92 | 17 | 934 |
| Oropharynx | RESP Chronic Bronchitis | 1.00 | — | 1 | 950 |
| Oropharynx | RESP Emphysema | 0.92 | 0.11 | 10 | 941 |
| Oropharynx | RESP Pulmonary Fibrosis | 0.08 | 0.03 | 7 | 944 |
| Oropharynx | RESP Pulmonary Hypertension | 1.00 | — | 2 | 949 |
| Oropharynx | RESP Sleep Apnea | 1.00 | — | 3 | 948 |
| Oropharynx | REN Acute Renal Failure | 0.00 | 0.00 | 29 | 922 |
| Oropharynx | REN Chronic Renal Failure | 0.06 | 0.30 | 24 | 927 |
| Oropharynx | REN End Stage Renal | 0.79 | 0.56 | 6 | 945 |
| Oropharynx | REN Diabetic Nephropathy | 1.00 | — | 2 | 949 |
| Oropharynx | HEP Steatosis | 0.00 | 0.00 | 14 | 937 |
| Oropharynx | HEP Alcoholic Cirrhosis | 0.95 | 0.06 | 8 | 943 |
| Oropharynx | HEP NonAlcoholic Cirrhosis | 1.00 | — | 0 | 951 |
| Oropharynx | HEP Viral Hepatitis | 1.00 | — | 0 | 951 |
| Oropharynx | GI Reflux | 1.00 | — | 2 | 949 |
| Oropharynx | GI Peptic Ulcer | 1.00 | — | 3 | 948 |
| Oropharynx | GI IBD | 1.00 | — | 1 | 950 |
| Oropharynx | GI IBS | 1.00 | — | 0 | 951 |
| Oropharynx | NEURO Dementia | 1.00 | — | 4 | 947 |
| Oropharynx | NEURO Parkinson | 1.00 | 0.12 | 6 | 945 |
| Oropharynx | NEURO Epilepsy | 1.00 | 0.06 | 21 | 930 |
| Oropharynx | NEURO Stroke Sequelae | 0.69 | 0.08 | 19 | 932 |
| Oropharynx | PSI Major Depression | 0.26 | 0.01 | 22 | 929 |
| Oropharynx | PSI Bipolar | 0.04 | 0.03 | 6 | 945 |
| Oropharynx | PSI Anxiety | 0.86 | 0.92 | 11 | 940 |
| Oropharynx | SUST Alcohol | 0.92 | 0.03 | 46 | 905 |
| Oropharynx | SUST Tobacco | 0.75 | 0.00 | 126 | 825 |
| Oropharynx | SUST Opioids | 1.00 | — | 1 | 950 |
| Oropharynx | INF HIV | 0.74 | 0.05 | 20 | 931 |
| Oropharynx | INF Tuberculosis | 0.24 | — | 2 | 949 |
| Oropharynx | INF Hepatitis C | 1.00 | — | 0 | 951 |
| Oropharynx | INF Hepatitis B | 1.00 | — | 0 | 951 |
| Other Pharynx | CV Essential Hypertension | 0.16 | 0.62 | 94 | 187 |
| Other Pharynx | CV Complicated Hypertension | 0.20 | — | 1 | 280 |
| Other Pharynx | CV Coronary Disease | 0.85 | 0.13 | 6 | 275 |
| Other Pharynx | CV Heart Failure | 0.35 | 0.13 | 7 | 274 |
| Other Pharynx | CV Arrhythmias | 0.08 | 0.00 | 14 | 267 |
| Other Pharynx | CV Cerebrovascular Disease | 0.65 | 0.11 | 10 | 271 |
| Other Pharynx | CV Peripheral Vascular Disease | 0.75 | 0.00 | 11 | 270 |
| Other Pharynx | MET Type1 Diabetes | 0.51 | — | 2 | 279 |
| Other Pharynx | MET Type2 Diabetes | 0.96 | 0.92 | 43 | 238 |
| Other Pharynx | MET Complicated Diabetes | 1.00 | — | 0 | 281 |
| Other Pharynx | MET Obesity | 1.00 | 0.57 | 8 | 273 |
| Other Pharynx | MET Dyslipidemia | 0.29 | 0.91 | 17 | 264 |
| Other Pharynx | MET Metabolic Syndrome | 1.00 | — | 0 | 281 |
| Other Pharynx | MET Thyroid Disorders | 0.40 | 0.53 | 26 | 255 |
| Other Pharynx | RESP COPD | 0.37 | 0.07 | 14 | 267 |
| Other Pharynx | RESP Asthma | 1.00 | — | 3 | 278 |
| Other Pharynx | RESP Chronic Bronchitis | 1.00 | — | 0 | 281 |
| Other Pharynx | RESP Emphysema | 0.91 | — | 4 | 277 |
| Other Pharynx | RESP Pulmonary Fibrosis | 1.00 | — | 3 | 278 |
| Other Pharynx | RESP Pulmonary Hypertension | 1.00 | — | 0 | 281 |
| Other Pharynx | RESP Sleep Apnea | 0.61 | 0.92 | 9 | 272 |
| Other Pharynx | REN Acute Renal Failure | 0.00 | 0.12 | 10 | 271 |
| Other Pharynx | REN Chronic Renal Failure | 0.83 | 0.03 | 12 | 269 |
| Other Pharynx | REN End Stage Renal | 0.91 | — | 4 | 277 |
| Other Pharynx | REN Diabetic Nephropathy | 1.00 | — | 0 | 281 |
| Other Pharynx | HEP Steatosis | 1.00 | — | 3 | 278 |
| Other Pharynx | HEP Alcoholic Cirrhosis | 0.20 | — | 1 | 280 |
| Other Pharynx | HEP NonAlcoholic Cirrhosis | 1.00 | — | 0 | 281 |
| Other Pharynx | HEP Viral Hepatitis | 1.00 | — | 0 | 281 |
| Other Pharynx | GI Reflux | 1.00 | — | 3 | 278 |
| Other Pharynx | GI Peptic Ulcer | 1.00 | 0.52 | 5 | 276 |
| Other Pharynx | GI IBD | 1.00 | — | 1 | 280 |
| Other Pharynx | GI IBS | 1.00 | — | 1 | 280 |
| Other Pharynx | NEURO Dementia | 1.00 | — | 2 | 279 |
| Other Pharynx | NEURO Parkinson | 1.00 | — | 0 | 281 |
| Other Pharynx | NEURO Epilepsy | 0.65 | 0.13 | 10 | 271 |
| Other Pharynx | NEURO Stroke Sequelae | 1.00 | 0.25 | 5 | 276 |
| Other Pharynx | PSI Major Depression | 0.25 | 0.71 | 6 | 275 |
| Other Pharynx | PSI Bipolar | 0.20 | — | 1 | 280 |
| Other Pharynx | PSI Anxiety | 1.00 | — | 2 | 279 |
| Other Pharynx | SUST Alcohol | 1.00 | 0.12 | 5 | 276 |
| Other Pharynx | SUST Tobacco | 0.46 | 0.00 | 31 | 250 |
| Other Pharynx | SUST Opioids | 1.00 | — | 0 | 281 |
| Other Pharynx | INF HIV | 1.00 | — | 4 | 277 |
| Other Pharynx | INF Tuberculosis | 1.00 | — | 1 | 280 |
| Other Pharynx | INF Hepatitis C | 1.00 | — | 0 | 281 |
| Other Pharynx | INF Hepatitis B | 1.00 | — | 0 | 281 |
| Salivary Glands | CV Essential Hypertension | 0.58 | 0.00 | 337 | 561 |
| Salivary Glands | CV Complicated Hypertension | 0.88 | 0.02 | 8 | 890 |
| Salivary Glands | CV Coronary Disease | 0.06 | 0.03 | 27 | 871 |
| Salivary Glands | CV Heart Failure | 0.01 | 0.00 | 20 | 878 |
| Salivary Glands | CV Arrhythmias | 0.00 | 0.00 | 42 | 856 |
| Salivary Glands | CV Cerebrovascular Disease | 0.04 | 0.00 | 15 | 883 |
| Salivary Glands | CV Peripheral Vascular Disease | 1.00 | 0.58 | 8 | 890 |
| Salivary Glands | MET Type1 Diabetes | 1.00 | — | 3 | 895 |
| Salivary Glands | MET Type2 Diabetes | 0.52 | 0.16 | 180 | 718 |
| Salivary Glands | MET Complicated Diabetes | 0.20 | — | 2 | 896 |
| Salivary Glands | MET Obesity | 0.22 | 0.56 | 72 | 826 |
| Salivary Glands | MET Dyslipidemia | 0.52 | 0.54 | 94 | 804 |
| Salivary Glands | MET Metabolic Syndrome | 1.00 | 0.60 | 6 | 892 |
| Salivary Glands | MET Thyroid Disorders | 1.00 | 0.60 | 102 | 796 |
| Salivary Glands | RESP COPD | 0.76 | 0.01 | 37 | 861 |
| Salivary Glands | RESP Asthma | 0.32 | 0.30 | 34 | 864 |
| Salivary Glands | RESP Chronic Bronchitis | 1.00 | — | 0 | 898 |
| Salivary Glands | RESP Emphysema | 0.20 | — | 2 | 896 |
| Salivary Glands | RESP Pulmonary Fibrosis | 1.00 | 0.01 | 7 | 891 |
| Salivary Glands | RESP Pulmonary Hypertension | 1.00 | — | 2 | 896 |
| Salivary Glands | RESP Sleep Apnea | 1.00 | 0.85 | 7 | 891 |
| Salivary Glands | REN Acute Renal Failure | 0.00 | 0.00 | 38 | 860 |
| Salivary Glands | REN Chronic Renal Failure | 0.24 | 0.00 | 38 | 860 |
| Salivary Glands | REN End Stage Renal | 0.81 | 0.50 | 7 | 891 |
| Salivary Glands | REN Diabetic Nephropathy | 1.00 | — | 1 | 897 |
| Salivary Glands | HEP Steatosis | 0.81 | 0.02 | 7 | 891 |
| Salivary Glands | HEP Alcoholic Cirrhosis | 0.20 | — | 2 | 896 |
| Salivary Glands | HEP NonAlcoholic Cirrhosis | 1.00 | — | 1 | 897 |
| Salivary Glands | HEP Viral Hepatitis | 1.00 | — | 0 | 898 |
| Salivary Glands | GI Reflux | 1.00 | — | 2 | 896 |
| Salivary Glands | GI Peptic Ulcer | 1.00 | — | 4 | 894 |
| Salivary Glands | GI IBD | 1.00 | — | 2 | 896 |
| Salivary Glands | GI IBS | 1.00 | — | 0 | 898 |
| Salivary Glands | NEURO Dementia | 0.81 | 0.31 | 7 | 891 |
| Salivary Glands | NEURO Parkinson | 0.93 | 0.32 | 11 | 887 |
| Salivary Glands | NEURO Epilepsy | 0.71 | 0.53 | 16 | 882 |
| Salivary Glands | NEURO Stroke Sequelae | 1.00 | 0.01 | 12 | 886 |
| Salivary Glands | PSI Major Depression | 1.00 | 0.19 | 28 | 870 |
| Salivary Glands | PSI Bipolar | 1.00 | — | 4 | 894 |
| Salivary Glands | PSI Anxiety | 1.00 | 0.06 | 9 | 889 |
| Salivary Glands | SUST Alcohol | 0.53 | 0.98 | 18 | 880 |
| Salivary Glands | SUST Tobacco | 1.00 | 0.07 | 83 | 815 |
| Salivary Glands | SUST Opioids | 1.00 | — | 0 | 898 |
| Salivary Glands | INF HIV | 1.00 | 0.27 | 5 | 893 |
| Salivary Glands | INF Tuberculosis | 1.00 | 0.62 | 5 | 893 |
| Salivary Glands | INF Hepatitis C | 1.00 | — | 0 | 898 |
| Salivary Glands | INF Hepatitis B | 1.00 | — | 0 | 898 |
| Sinonasal | CV Essential Hypertension | 0.74 | 0.06 | 293 | 594 |
| Sinonasal | CV Complicated Hypertension | 0.57 | 0.11 | 5 | 882 |
| Sinonasal | CV Coronary Disease | 0.99 | 0.81 | 11 | 876 |
| Sinonasal | CV Heart Failure | 1.00 | 0.00 | 12 | 875 |
| Sinonasal | CV Arrhythmias | 1.00 | 0.00 | 15 | 872 |
| Sinonasal | CV Cerebrovascular Disease | 0.00 | 0.03 | 19 | 868 |
| Sinonasal | CV Peripheral Vascular Disease | 1.00 | 0.01 | 7 | 880 |
| Sinonasal | MET Type1 Diabetes | 0.45 | — | 4 | 883 |
| Sinonasal | MET Type2 Diabetes | 0.40 | 0.02 | 142 | 745 |
| Sinonasal | MET Complicated Diabetes | 1.00 | — | 2 | 885 |
| Sinonasal | MET Obesity | 0.57 | 0.05 | 57 | 830 |
| Sinonasal | MET Dyslipidemia | 1.00 | 0.93 | 65 | 822 |
| Sinonasal | MET Metabolic Syndrome | 0.17 | — | 2 | 885 |
| Sinonasal | MET Thyroid Disorders | 1.00 | 0.46 | 75 | 812 |
| Sinonasal | RESP COPD | 1.00 | 0.80 | 16 | 871 |
| Sinonasal | RESP Asthma | 0.57 | 0.73 | 23 | 864 |
| Sinonasal | RESP Chronic Bronchitis | 1.00 | — | 0 | 887 |
| Sinonasal | RESP Emphysema | 1.00 | — | 0 | 887 |
| Sinonasal | RESP Pulmonary Fibrosis | 1.00 | 0.56 | 5 | 882 |
| Sinonasal | RESP Pulmonary Hypertension | 1.00 | — | 0 | 887 |
| Sinonasal | RESP Sleep Apnea | 1.00 | 0.42 | 6 | 881 |
| Sinonasal | REN Acute Renal Failure | 0.32 | 0.00 | 15 | 872 |
| Sinonasal | REN Chronic Renal Failure | 1.00 | 0.05 | 18 | 869 |
| Sinonasal | REN End Stage Renal | 1.00 | 0.03 | 7 | 880 |
| Sinonasal | REN Diabetic Nephropathy | 1.00 | — | 2 | 885 |
| Sinonasal | HEP Steatosis | 0.83 | 0.09 | 8 | 879 |
| Sinonasal | HEP Alcoholic Cirrhosis | 1.00 | — | 2 | 885 |
| Sinonasal | HEP NonAlcoholic Cirrhosis | 1.00 | — | 0 | 887 |
| Sinonasal | HEP Viral Hepatitis | 1.00 | — | 0 | 887 |
| Sinonasal | GI Reflux | 1.00 | — | 3 | 884 |
| Sinonasal | GI Peptic Ulcer | 1.00 | — | 3 | 884 |
| Sinonasal | GI IBD | 1.00 | — | 0 | 887 |
| Sinonasal | GI IBS | 1.00 | — | 0 | 887 |
| Sinonasal | NEURO Dementia | 1.00 | 0.18 | 8 | 879 |
| Sinonasal | NEURO Parkinson | 1.00 | 0.89 | 6 | 881 |
| Sinonasal | NEURO Epilepsy | 1.00 | 0.10 | 14 | 873 |
| Sinonasal | NEURO Stroke Sequelae | 0.95 | 0.95 | 10 | 877 |
| Sinonasal | PSI Major Depression | 1.00 | 0.51 | 18 | 869 |
| Sinonasal | PSI Bipolar | 1.00 | — | 1 | 886 |
| Sinonasal | PSI Anxiety | 0.99 | 0.87 | 11 | 876 |
| Sinonasal | SUST Alcohol | 1.00 | 0.08 | 29 | 858 |
| Sinonasal | SUST Tobacco | 0.74 | 0.29 | 84 | 803 |
| Sinonasal | SUST Opioids | 1.00 | — | 0 | 887 |
| Sinonasal | INF HIV | 1.00 | — | 4 | 883 |
| Sinonasal | INF Tuberculosis | 1.00 | — | 0 | 887 |
| Sinonasal | INF Hepatitis C | 1.00 | — | 0 | 887 |
| Sinonasal | INF Hepatitis B | 1.00 | — | 0 | 887 |
# Build dictionaries of significant comorbidities per site (p < 0.05)
display_to_col = {format_name(c): c for c in comorbidity_cols}
sig_comorbidities_mortality = {}
sig_comorbidities_los = {}
for site in all_site_names:
site_biv = df_bivariate[df_bivariate['Cancer Site'] == site]
sig_mort = []
sig_los = []
for _, row in site_biv.iterrows():
comorb_display = row['Comorbidity']
col_name = display_to_col.get(comorb_display, None)
if col_name is None:
continue
if not np.isnan(row['Chi-square p-value']) and row['Chi-square p-value'] < 0.05:
sig_mort.append(col_name)
if not np.isnan(row['LOS p-value (Wilcoxon)']) and row['LOS p-value (Wilcoxon)'] < 0.05:
sig_los.append(col_name)
sig_comorbidities_mortality[site] = sig_mort
sig_comorbidities_los[site] = sig_los
sig_summary = []
for site in all_site_names:
sig_summary.append({
'Cancer Site': site,
'Sig. Comorbidities (Mortality)': len(sig_comorbidities_mortality.get(site, [])),
'Sig. Comorbidities (LOS)': len(sig_comorbidities_los.get(site, [])),
'Mortality List': ', '.join([format_name(c) for c in sig_comorbidities_mortality.get(site, [])]) or '—',
'LOS List': ', '.join([format_name(c) for c in sig_comorbidities_los.get(site, [])]) or '—'
})
sig_summary_df = pd.DataFrame(sig_summary)
show_table(sig_summary_df, "Table 21: Significant Comorbidities by Cancer Site (p < 0.05 in Bivariate Analysis)", index=False)| Cancer Site | Sig. Comorbidities (Mortality) | Sig. Comorbidities (LOS) | Mortality List | LOS List |
|---|---|---|---|---|
| Hypopharynx | 1 | 9 | CV Arrhythmias | CV Peripheral Vascular Disease, RESP COPD, RESP Emphysema, REN Acute Renal Failure, REN Chronic Renal Failure, REN End Stage Renal, NEURO Epilepsy, SUST Alcohol, SUST Tobacco |
| Larynx | 11 | 20 | CV Heart Failure, CV Arrhythmias, CV Peripheral Vascular Disease, RESP COPD, RESP Emphysema, REN Acute Renal Failure, REN Chronic Renal Failure, REN End Stage Renal, HEP Alcoholic Cirrhosis, HEP NonAlcoholic Cirrhosis, GI Peptic Ulcer | CV Essential Hypertension, CV Complicated Hypertension, CV Coronary Disease, CV Heart Failure, CV Arrhythmias, CV Cerebrovascular Disease, CV Peripheral Vascular Disease, MET Metabolic Syndrome, RESP COPD, RESP Emphysema, REN Acute Renal Failure, REN Chronic Renal Failure, HEP Alcoholic Cirrhosis, GI Peptic Ulcer, NEURO Epilepsy, NEURO Stroke Sequelae, PSI Major Depression, PSI Anxiety, SUST Alcohol, SUST Tobacco |
| Nasopharynx | 1 | 6 | REN Acute Renal Failure | CV Arrhythmias, CV Cerebrovascular Disease, CV Peripheral Vascular Disease, MET Type2 Diabetes, REN Acute Renal Failure, SUST Alcohol |
| Oral Cavity | 11 | 23 | CV Heart Failure, CV Arrhythmias, CV Cerebrovascular Disease, CV Peripheral Vascular Disease, MET Metabolic Syndrome, RESP Asthma, RESP Pulmonary Fibrosis, REN Acute Renal Failure, REN Chronic Renal Failure, REN End Stage Renal, HEP Steatosis | CV Essential Hypertension, CV Coronary Disease, CV Heart Failure, CV Arrhythmias, CV Cerebrovascular Disease, CV Peripheral Vascular Disease, MET Obesity, MET Metabolic Syndrome, MET Thyroid Disorders, RESP COPD, RESP Emphysema, RESP Sleep Apnea, REN Acute Renal Failure, REN Chronic Renal Failure, REN End Stage Renal, GI Peptic Ulcer, NEURO Epilepsy, NEURO Stroke Sequelae, PSI Major Depression, PSI Anxiety, SUST Alcohol, SUST Tobacco, INF HIV |
| Oropharynx | 6 | 14 | CV Heart Failure, CV Arrhythmias, RESP COPD, REN Acute Renal Failure, HEP Steatosis, PSI Bipolar | CV Essential Hypertension, CV Coronary Disease, CV Arrhythmias, CV Cerebrovascular Disease, CV Peripheral Vascular Disease, RESP COPD, RESP Pulmonary Fibrosis, REN Acute Renal Failure, HEP Steatosis, PSI Major Depression, PSI Bipolar, SUST Alcohol, SUST Tobacco, INF HIV |
| Other Pharynx | 1 | 4 | REN Acute Renal Failure | CV Arrhythmias, CV Peripheral Vascular Disease, REN Chronic Renal Failure, SUST Tobacco |
| Salivary Glands | 4 | 12 | CV Heart Failure, CV Arrhythmias, CV Cerebrovascular Disease, REN Acute Renal Failure | CV Essential Hypertension, CV Complicated Hypertension, CV Coronary Disease, CV Heart Failure, CV Arrhythmias, CV Cerebrovascular Disease, RESP COPD, RESP Pulmonary Fibrosis, REN Acute Renal Failure, REN Chronic Renal Failure, HEP Steatosis, NEURO Stroke Sequelae |
| Sinonasal | 1 | 7 | CV Cerebrovascular Disease | CV Heart Failure, CV Arrhythmias, CV Cerebrovascular Disease, CV Peripheral Vascular Disease, MET Type2 Diabetes, REN Acute Renal Failure, REN End Stage Renal |
from statsmodels.genmod.generalized_estimating_equations import GEE
from statsmodels.genmod.families import Binomial, Poisson, NegativeBinomial
from statsmodels.genmod.cov_struct import Exchangeable, Independence
import statsmodels.api as sm
from sklearn.metrics import (accuracy_score, precision_score, recall_score,
f1_score, matthews_corrcoef, brier_score_loss,
roc_auc_score, roc_curve, confusion_matrix,
classification_report)
# Use ALL cancer site groups
all_sites = sorted(df_head_neck['cancer_site_group'].unique())
# Store GEE model results
gee_mortality_models = {}
gee_mortality_results = {}
def find_optimal_threshold(y_true, y_proba):
"""Find optimal classification threshold using Youden's J statistic"""
fpr, tpr, thresholds = roc_curve(y_true, y_proba)
j_scores = tpr - fpr
best_idx = np.argmax(j_scores)
return thresholds[best_idx]
def safe_fmt(val, fmt='.4f'):
"""Format a value safely, returning '—' for NaN/Inf"""
if isinstance(val, float) and (np.isnan(val) or np.isinf(val)):
return '—'
try:
return f'{val:{fmt}}'
except:
return '—'
def build_gee_mortality_model(df_site, site_name):
"""Build GEE logistic regression (mixed-effects approximation) for mortality.
Uses hospital_id as cluster group, year_factor as fixed effect,
and only significant comorbidities from bivariate analysis."""
df_model = df_site.copy()
# Code categorical variables
df_model['SEXO_coded'] = (df_model['SEXO'] == 'Female').astype(int)
# Create service dummies
service_dummies = pd.get_dummies(df_model['service_category'], prefix='service', drop_first=True)
for col in service_dummies.columns:
df_model[col] = service_dummies[col]
service_dummy_cols = list(service_dummies.columns)
# Create year dummies (year_factor as fixed effect)
year_dummies = pd.get_dummies(df_model['year_factor'], prefix='year', drop_first=True)
for col in year_dummies.columns:
df_model[col] = year_dummies[col]
year_dummy_cols = list(year_dummies.columns)
# Ensure numeric
df_model['edad'] = pd.to_numeric(df_model['edad'], errors='coerce')
# Base predictors: age, sex, service, year dummies
predictor_cols = ['edad', 'SEXO_coded'] + service_dummy_cols + year_dummy_cols
# Add severity and mortality risk as CATEGORICAL dummies (Low=ref, Medium, High)
for col in ['IR_29301_SEVERIDAD', 'IR_29301_MORTALIDAD']:
if col in df_model.columns:
df_model[col] = pd.to_numeric(df_model[col], errors='coerce')
df_model[col] = df_model[col].fillna(1)
df_model[col] = df_model[col].clip(lower=1).astype(int)
prefix = 'severity' if 'SEVERIDAD' in col else 'mort_risk'
col_dummies = pd.get_dummies(df_model[col], prefix=prefix, drop_first=True)
for dc in col_dummies.columns:
df_model[dc] = col_dummies[dc]
predictor_cols.append(dc)
# Add ONLY significant comorbidities from bivariate analysis (p < 0.05)
sig_comorbs = sig_comorbidities_mortality.get(site_name, [])
used_comorbs = []
for col in sig_comorbs:
if col in df_model.columns:
df_model[col] = pd.to_numeric(df_model[col], errors='coerce').fillna(0).astype(int)
if df_model[col].sum() >= 3:
predictor_cols.append(col)
used_comorbs.append(col)
# Prepare groups (hospital_id for clustering)
df_model['_group'] = df_model['hospital_id']
# Remove missing values
cols_for_model = ['egresar_fallecido', '_group'] + predictor_cols
df_gee = df_model[cols_for_model].dropna()
# Need at least 2 clusters and sufficient data
n_clusters = df_gee['_group'].nunique()
if len(df_gee) < 50 or n_clusters < 2:
return None, None, [], []
# Build design matrix
y = df_gee['egresar_fallecido'].astype(float)
X = df_gee[predictor_cols].astype(float)
X = sm.add_constant(X)
groups = df_gee['_group']
# Fit GEE with exchangeable correlation structure
try:
gee_model = GEE(y, X, groups=groups,
family=Binomial(),
cov_struct=Exchangeable())
gee_result = gee_model.fit(maxiter=200)
except Exception:
# Fallback to independence correlation
try:
gee_model = GEE(y, X, groups=groups,
family=Binomial(),
cov_struct=Independence())
gee_result = gee_model.fit(maxiter=200)
except Exception:
return None, None, [], []
# Check for extreme coefficients and remove problematic variables
max_retries = 5
current_predictors = list(predictor_cols)
for _ in range(max_retries):
params = gee_result.params
bad_vars = []
for i, var in enumerate(current_predictors):
idx = i + 1 # +1 for constant
if idx < len(params) and (np.isinf(params.iloc[idx]) or abs(params.iloc[idx]) > 50):
bad_vars.append(var)
if not bad_vars:
break
current_predictors = [p for p in current_predictors if p not in bad_vars]
if len(current_predictors) < 2:
return None, None, [], []
X = df_gee[current_predictors].astype(float)
X = sm.add_constant(X)
try:
gee_model = GEE(y, X, groups=groups, family=Binomial(), cov_struct=Exchangeable())
gee_result = gee_model.fit(maxiter=200)
except:
try:
gee_model = GEE(y, X, groups=groups, family=Binomial(), cov_struct=Independence())
gee_result = gee_model.fit(maxiter=200)
except:
return None, None, [], []
return gee_result, df_gee, current_predictors, used_comorbs
def get_gee_fit_metrics(gee_result, y, family_type='binomial'):
"""Extract model fit metrics from a GEE result object."""
metrics = {}
metrics['N_obs'] = int(gee_result.nobs)
metrics['N_clusters'] = int(len(set(gee_result.model.groups)))
metrics['N_params'] = int(len(gee_result.params))
# QIC / QICu (quasi-information criteria)
try:
qic, qicu = gee_result.qic()
metrics['QIC'] = qic
metrics['QICu'] = qicu
except Exception:
metrics['QIC'] = np.nan
metrics['QICu'] = np.nan
# Quasi log-likelihood (scale parameter)
try:
metrics['Scale'] = float(gee_result.scale)
except Exception:
metrics['Scale'] = np.nan
# Pseudo R² (McFadden-style for logistic, deviance-based for count)
try:
y_pred = gee_result.fittedvalues
if family_type == 'binomial':
y_arr = np.asarray(y, dtype=float)
p_hat = np.clip(np.asarray(y_pred, dtype=float), 1e-10, 1 - 1e-10)
ll_model = np.sum(y_arr * np.log(p_hat) + (1 - y_arr) * np.log(1 - p_hat))
p_null = np.mean(y_arr)
ll_null = np.sum(y_arr * np.log(p_null) + (1 - y_arr) * np.log(1 - p_null))
metrics['Log-Lik (model)'] = ll_model
metrics['Log-Lik (null)'] = ll_null
metrics['Pseudo R² (McFadden)'] = 1 - ll_model / ll_null if ll_null != 0 else np.nan
else:
y_arr = np.asarray(y, dtype=float)
y_pred_arr = np.asarray(y_pred, dtype=float)
ss_res = np.sum((y_arr - y_pred_arr) ** 2)
ss_tot = np.sum((y_arr - np.mean(y_arr)) ** 2)
metrics['Pseudo R²'] = 1 - ss_res / ss_tot if ss_tot > 0 else np.nan
except Exception:
if family_type == 'binomial':
metrics['Log-Lik (model)'] = np.nan
metrics['Log-Lik (null)'] = np.nan
metrics['Pseudo R² (McFadden)'] = np.nan
else:
metrics['Pseudo R²'] = np.nan
# Correlation structure
try:
cov_name = gee_result.cov_struct.__class__.__name__
metrics['Correlation'] = cov_name
except Exception:
metrics['Correlation'] = '—'
return metrics
# Fit GEE mortality models for EACH cancer site
for site in all_sites:
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site].copy()
n_events = df_site['egresar_fallecido'].sum()
n_total = len(df_site)
if n_events < 10 or n_total < 50:
continue
gee_result, df_gee, used_cols, used_comorbs = build_gee_mortality_model(df_site, site)
if gee_result is not None:
gee_mortality_models[site] = (gee_result, df_gee, used_cols)
gee_mortality_results[site] = gee_result# Table: GEE Logistic Regression - OR [95% CI] by Cancer Site
gee_or_table = {}
for site in all_sites:
if site not in gee_mortality_results:
continue
result = gee_mortality_results[site]
or_ci_list = {}
param_names = list(result.params.index)
ci = result.conf_int()
def _fmt_or(v):
if np.isinf(v) or abs(v) > 1e6:
return 'Inf' if v > 0 else '-Inf'
return f"{v:.3f}"
for var in param_names:
coef = result.params[var]
or_val = np.exp(coef)
ci_lower = np.exp(ci.loc[var, 0] if var in ci.index else ci.iloc[param_names.index(var), 0])
ci_upper = np.exp(ci.loc[var, 1] if var in ci.index else ci.iloc[param_names.index(var), 1])
pval = result.pvalues[var] if var in result.pvalues.index else result.pvalues.iloc[param_names.index(var)]
if np.isnan(or_val) or np.isnan(pval):
or_ci_list[format_name(var)] = '—'
continue
sig_stars = ''
if pval < 0.001:
sig_stars = '***'
elif pval < 0.01:
sig_stars = '**'
elif pval < 0.05:
sig_stars = '*'
or_ci_str = f"{_fmt_or(or_val)} [{_fmt_or(ci_lower)}, {_fmt_or(ci_upper)}]{sig_stars}"
or_ci_list[format_name(var)] = or_ci_str
gee_or_table[site] = or_ci_list
gee_or_df = pd.DataFrame(gee_or_table).fillna('—')
show_table(gee_or_df, "Table 9: GEE Mixed-Effects Logistic Regression - Odds Ratios [95% CI] for In-Hospital Mortality by Cancer Site\n(Hospital as cluster, year as fixed effect, only significant comorbidities from bivariate analysis)\n* p<0.05, ** p<0.01, *** p<0.001", index=True)| Hypopharynx | Larynx | Nasopharynx | Oral Cavity | Oropharynx | Other Pharynx | Salivary Glands | Sinonasal | |
|---|---|---|---|---|---|---|---|---|
| Intercept | 0.000 [0.000, 0.025]*** | 0.004 [0.001, 0.014]*** | — | 0.002 [0.000, 0.007]*** | 0.012 [0.000, 0.329]** | 0.001 [0.000, 0.172]** | 0.015 [0.001, 0.206]** | 0.045 [0.007, 0.290]** |
| Age | 1.032 [0.989, 1.078] | 1.027 [1.013, 1.041]*** | — | 1.018 [1.002, 1.035]* | 1.022 [0.982, 1.064] | 1.052 [1.005, 1.101]* | 0.990 [0.969, 1.011] | 1.002 [0.985, 1.018] |
| Sex (Female) | 1.076 [0.447, 2.590] | 0.982 [0.580, 1.662] | — | 1.014 [0.701, 1.468] | 0.927 [0.408, 2.108] | 0.863 [0.353, 2.109] | 0.485 [0.234, 1.005] | 0.887 [0.440, 1.784] |
| Service ITU | 2.796 [0.357, 21.901] | 0.451 [0.209, 0.973]* | — | 0.583 [0.259, 1.311] | 0.463 [0.131, 1.641] | 0.489 [0.053, 4.518] | 0.969 [0.150, 6.276] | 0.223 [0.043, 1.165] |
| Service MS | 0.812 [0.172, 3.835] | 0.819 [0.491, 1.366] | — | 0.884 [0.524, 1.492] | 0.367 [0.150, 0.899]* | 0.670 [0.167, 2.681] | 0.521 [0.164, 1.660] | 0.284 [0.084, 0.957]* |
| Service Other | 1.449 [0.417, 5.041] | 0.786 [0.496, 1.244] | — | 1.183 [0.655, 2.136] | 0.276 [0.118, 0.648]** | 0.443 [0.140, 1.405] | 0.714 [0.218, 2.337] | 0.335 [0.131, 0.852]* |
| Year 2020 | 3.417 [0.564, 20.696] | 0.585 [0.347, 0.987]* | — | 0.553 [0.304, 1.004] | 0.830 [0.305, 2.257] | 0.622 [0.100, 3.866] | 1.476 [0.463, 4.707] | 1.494 [0.430, 5.189] |
| Year 2021 | 5.211 [0.774, 35.084] | 0.693 [0.380, 1.263] | — | 1.251 [0.669, 2.340] | 0.454 [0.142, 1.450] | 2.348 [0.575, 9.596] | 1.180 [0.399, 3.488] | 0.594 [0.164, 2.152] |
| Year 2022 | 4.763 [0.770, 29.484] | 0.392 [0.235, 0.654]*** | — | 0.935 [0.514, 1.699] | 0.383 [0.136, 1.084] | 0.596 [0.156, 2.284] | 1.001 [0.306, 3.278] | 1.095 [0.290, 4.133] |
| Year 2023 | 2.285 [0.314, 16.609] | 0.460 [0.275, 0.769]** | — | 0.734 [0.411, 1.312] | 0.309 [0.129, 0.744]** | 1.883 [0.476, 7.445] | 0.470 [0.106, 2.083] | 0.444 [0.136, 1.451] |
| Year 2024 | 3.120 [0.391, 24.884] | 0.332 [0.206, 0.534]*** | — | 0.934 [0.514, 1.697] | 0.232 [0.076, 0.711]* | 1.430 [0.305, 6.697] | 1.361 [0.393, 4.709] | 0.720 [0.213, 2.435] |
| Severity Medium | 1.532 [0.170, 13.757] | 1.099 [0.597, 2.024] | — | 2.686 [1.031, 7.000]* | 3.020 [1.179, 7.733]* | 0.329 [0.068, 1.588] | 1.370 [0.103, 18.236] | 3.409 [1.298, 8.952]* |
| Severity High | 3.470 [0.545, 22.076] | 3.300 [1.795, 6.067]*** | — | 11.927 [4.378, 32.496]*** | 8.455 [2.361, 30.286]** | 0.481 [0.073, 3.168] | 5.837 [0.353, 96.645] | 12.691 [3.231, 49.846]*** |
| Mortality Risk Medium | 1.412 [0.300, 6.649] | 1.296 [0.777, 2.161] | — | 1.412 [0.644, 3.099] | 0.769 [0.301, 1.961] | 6.715 [0.469, 96.243] | 1.272 [0.101, 16.017] | 0.416 [0.150, 1.154] |
| Mortality Risk High | 8.937 [2.100, 38.033]** | 7.286 [4.452, 11.923]*** | — | 7.506 [3.505, 16.075]*** | 4.280 [1.524, 12.019]** | 90.776 [5.197, 1585.558]** | 16.732 [0.945, 296.340] | 2.286 [0.568, 9.196] |
| CV Arrhythmias | 2.703 [0.461, 15.855] | 2.508 [1.640, 3.835]*** | — | 1.878 [1.105, 3.193]* | 2.270 [0.440, 11.709] | — | 1.805 [0.597, 5.461] | — |
| CV Heart Failure | — | 1.551 [0.765, 3.148] | — | 0.680 [0.357, 1.297] | 4.117 [0.670, 25.303] | — | 0.755 [0.147, 3.868] | — |
| CV Peripheral Vascular Disease | — | 1.384 [0.700, 2.736] | — | 1.882 [0.948, 3.738] | — | — | — | — |
| RESP COPD | — | 0.865 [0.516, 1.449] | — | — | 1.142 [0.532, 2.453] | — | — | — |
| RESP Emphysema | — | 1.350 [0.628, 2.905] | — | — | — | — | — | — |
| REN Acute Renal Failure | — | 1.157 [0.745, 1.798] | — | 0.465 [0.256, 0.843]* | 1.225 [0.384, 3.910] | 0.700 [0.134, 3.664] | 0.574 [0.222, 1.487] | — |
| REN Chronic Renal Failure | — | 0.801 [0.476, 1.349] | — | 0.877 [0.377, 2.039] | — | — | — | — |
| REN End Stage Renal | — | 1.210 [0.440, 3.328] | — | 1.107 [0.233, 5.269] | — | — | — | — |
| HEP Alcoholic Cirrhosis | — | 4.357 [2.123, 8.944]*** | — | — | — | — | — | — |
| GI Peptic Ulcer | — | 2.035 [0.554, 7.479] | — | — | — | — | — | — |
| CV Cerebrovascular Disease | — | — | — | 1.038 [0.471, 2.287] | — | — | 3.283 [0.465, 23.189] | 1.566 [0.492, 4.982] |
| MET Metabolic Syndrome | — | — | — | 2.562 [0.724, 9.068] | — | — | — | — |
| RESP Asthma | — | — | — | 3.147 [1.268, 7.806]* | — | — | — | — |
| RESP Pulmonary Fibrosis | — | — | — | 2.534 [0.609, 10.534] | — | — | — | — |
| HEP Steatosis | — | — | — | 1.926 [1.002, 3.701]* | 6.214 [1.237, 31.206]* | — | — | — |
| PSI Bipolar | — | — | — | — | 1.445 [0.193, 10.842] | — | — | — |
# Table 10: Performance Metrics - GEE Logistic Models (ALL sites)
gee_perf = []
for site in all_sites:
if site not in gee_mortality_results:
continue
result = gee_mortality_results[site]
df_gee = gee_mortality_models[site][1]
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
n_total = len(df_site)
n_events = df_site['egresar_fallecido'].sum()
event_rate = n_events / n_total * 100 if n_total > 0 else 0
try:
y_pred_proba = result.fittedvalues
y_actual = df_gee['egresar_fallecido'].values
auc = roc_auc_score(y_actual, y_pred_proba) if len(np.unique(y_actual)) > 1 else np.nan
brier = brier_score_loss(y_actual, y_pred_proba)
opt_threshold = find_optimal_threshold(y_actual, y_pred_proba)
y_pred = (y_pred_proba >= opt_threshold).astype(int)
accuracy = accuracy_score(y_actual, y_pred)
sensitivity = recall_score(y_actual, y_pred, zero_division=0)
specificity = recall_score(y_actual, y_pred, pos_label=0, zero_division=0)
ppv = precision_score(y_actual, y_pred, zero_division=0)
tn = ((1 - y_actual) * (1 - y_pred)).sum()
fn = (y_actual * (1 - y_pred)).sum()
npv = tn / (tn + fn) if (tn + fn) > 0 else 0
f1 = f1_score(y_actual, y_pred, zero_division=0)
mcc = matthews_corrcoef(y_actual, y_pred)
youdens_j = sensitivity + specificity - 1
except Exception:
accuracy = sensitivity = specificity = ppv = npv = f1 = mcc = brier = auc = np.nan
opt_threshold = np.nan
youdens_j = np.nan
n_sig_comorbs = len(sig_comorbidities_mortality.get(site, []))
fit_metrics = get_gee_fit_metrics(result, df_gee['egresar_fallecido'], family_type='binomial')
gee_perf.append({
'Cancer Site': site,
'N obs': fit_metrics['N_obs'],
'N clusters': fit_metrics['N_clusters'],
'N params': fit_metrics['N_params'],
'Events': f'{n_events} ({event_rate:.1f}%)',
'Sig. Comorb.': n_sig_comorbs,
'Correlation': fit_metrics['Correlation'],
'QIC': safe_fmt(fit_metrics['QIC'], '.1f'),
'QICu': safe_fmt(fit_metrics['QICu'], '.1f'),
'Log-Lik': safe_fmt(fit_metrics['Log-Lik (model)'], '.1f'),
'Pseudo R²': safe_fmt(fit_metrics['Pseudo R² (McFadden)']),
'Scale': safe_fmt(fit_metrics['Scale']),
'Threshold': safe_fmt(opt_threshold),
'AUC': safe_fmt(auc),
'Brier': safe_fmt(brier),
'Sensitivity': safe_fmt(sensitivity),
'Specificity': safe_fmt(specificity),
'PPV': safe_fmt(ppv),
'NPV': safe_fmt(npv),
'Bal. Accuracy': safe_fmt((sensitivity + specificity) / 2 if not np.isnan(sensitivity) else np.nan),
'F1': safe_fmt(f1),
'MCC': safe_fmt(mcc),
"Youden's J": safe_fmt(youdens_j),
})
gee_perf_df = pd.DataFrame(gee_perf)
show_table(gee_perf_df, "Table 10: Performance & Fit Metrics - GEE Logistic Regression for Mortality\n(QIC = Quasi-Information Criterion, Pseudo R² = McFadden)", index=False)| Cancer Site | N obs | N clusters | N params | Events | Sig. Comorb. | Correlation | QIC | QICu | Log-Lik | Pseudo R² | Scale | Threshold | AUC | Brier | Sensitivity | Specificity | PPV | NPV | Bal. Accuracy | F1 | MCC | Youden's J |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 400 | 49 | 16 | 33 (8.2%) | 1 | Exchangeable | 174.8 | 202.4 | -85.2 | 0.2523 | 1.0000 | 0.0556 | 0.8623 | 0.0624 | 0.9394 | 0.7193 | 0.2313 | 0.9925 | 0.8294 | 0.3713 | 0.3840 | 0.6587 |
| Larynx | 2625 | 64 | 25 | 198 (7.5%) | 11 | Exchangeable | 1041.6 | 1084.5 | -517.2 | 0.2633 | 1.0000 | 0.0915 | 0.8590 | 0.0554 | 0.7929 | 0.7973 | 0.2419 | 0.9793 | 0.7951 | 0.3707 | 0.3613 | 0.5902 |
| Nasopharynx | 368 | 50 | 16 | 23 (6.2%) | 1 | Exchangeable | — | — | — | — | 1.0000 | — | — | — | — | — | — | — | — | — | — | — |
| Oral Cavity | 3001 | 68 | 26 | 176 (5.9%) | 11 | Exchangeable | 915.3 | 959.3 | -453.7 | 0.3228 | 1.0000 | 0.1031 | 0.8971 | 0.0426 | 0.7898 | 0.8683 | 0.2720 | 0.9851 | 0.8290 | 0.4047 | 0.4114 | 0.6581 |
| Oropharynx | 951 | 59 | 21 | 55 (5.8%) | 6 | Exchangeable | 298.6 | 335.1 | -146.5 | 0.3027 | 1.0000 | 0.1115 | 0.8876 | 0.0430 | 0.7636 | 0.8884 | 0.2958 | 0.9839 | 0.8260 | 0.4264 | 0.4271 | 0.6520 |
| Other Pharynx | 281 | 51 | 16 | 30 (10.7%) | 1 | Exchangeable | 128.4 | 155.8 | -61.9 | 0.3514 | 1.0000 | 0.2047 | 0.8981 | 0.0653 | 0.8000 | 0.8805 | 0.4444 | 0.9736 | 0.8402 | 0.5714 | 0.5333 | 0.6805 |
| Salivary Glands | 898 | 59 | 19 | 46 (5.1%) | 4 | Exchangeable | 229.6 | 262.1 | -112.1 | 0.3826 | 1.0000 | 0.0214 | 0.9350 | 0.0378 | 0.9783 | 0.8075 | 0.2153 | 0.9985 | 0.8929 | 0.3529 | 0.4099 | 0.7858 |
| Sinonasal | 887 | 58 | 16 | 41 (4.6%) | 1 | Exchangeable | 235.2 | 263.7 | -115.8 | 0.3025 | 1.0000 | 0.0453 | 0.8984 | 0.0360 | 0.8780 | 0.8298 | 0.2000 | 0.9929 | 0.8539 | 0.3258 | 0.3695 | 0.7078 |
fig, ax = plt.subplots(figsize=(10, 7))
colors_roc = plt.cm.tab10(np.linspace(0, 1, len(gee_mortality_results)))
for idx, (site, result) in enumerate(gee_mortality_results.items()):
try:
y_pred_proba = result.fittedvalues
df_gee = gee_mortality_models[site][1]
y_actual = df_gee['egresar_fallecido'].values
fpr, tpr, _ = roc_curve(y_actual, y_pred_proba)
auc_val = roc_auc_score(y_actual, y_pred_proba)
ax.plot(fpr, tpr, label=f'{site} (AUC={auc_val:.3f})', linewidth=2, color=colors_roc[idx])
except:
pass
ax.plot([0, 1], [0, 1], 'k--', label='Random', linewidth=1)
ax.set_xlabel('False Positive Rate', fontsize=11)
ax.set_ylabel('True Positive Rate', fontsize=11)
ax.set_title('ROC Curves - GEE Logistic Regression by Cancer Site', fontsize=12, fontweight='bold')
ax.legend(loc='lower right', fontsize=9)
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'roc_gee_logistic_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-roc-gee-logistic.png", dpi=300, bbox_inches="tight")
plt.show()
def _extract_or_ci(result, drop_intercept=True, cap=(0.01, 100.0)):
"""Return DataFrame with var, OR, lo, hi, p — finite & capped for plotting."""
params = result.params
ci = result.conf_int()
pvals = result.pvalues
rows = []
for var in params.index:
if drop_intercept and var.lower() in {'const', 'intercept'}:
continue
coef = params[var]
lo, hi = (ci.loc[var, 0], ci.loc[var, 1]) if var in ci.index else (np.nan, np.nan)
p = pvals[var] if var in pvals.index else np.nan
or_v, or_lo, or_hi = np.exp(coef), np.exp(lo), np.exp(hi)
# Skip non-finite
if not np.isfinite(or_v) or not np.isfinite(or_lo) or not np.isfinite(or_hi):
continue
# Cap extreme CIs for display only
or_v_d = float(np.clip(or_v, cap[0], cap[1]))
or_lo_d = float(np.clip(or_lo, cap[0], cap[1]))
or_hi_d = float(np.clip(or_hi, cap[0], cap[1]))
rows.append({
'var': format_name(var),
'OR': or_v_d, 'lo': or_lo_d, 'hi': or_hi_d,
'p': p,
})
return pd.DataFrame(rows)
# Build per-site dataframes
forest_data_logit = {site: _extract_or_ci(res) for site, res in gee_mortality_results.items()}
forest_data_logit = {s: d for s, d in forest_data_logit.items() if not d.empty}
n_sites = len(forest_data_logit)
n_cols = 3
n_rows = int(np.ceil(n_sites / n_cols))
fig, axes = plt.subplots(n_rows, n_cols, figsize=(16, 4 * n_rows), squeeze=False)
for ax_idx, (site, df_f) in enumerate(forest_data_logit.items()):
r, c = divmod(ax_idx, n_cols)
ax = axes[r][c]
df_f = df_f.sort_values('OR').reset_index(drop=True)
y = np.arange(len(df_f))
sig_mask = (df_f['p'] < 0.05)
# Colors: red OR>1 sig, blue OR<1 sig, gray ns
colors = np.where(~sig_mask, '#9e9e9e',
np.where(df_f['OR'] > 1, '#c0392b', '#1f6f8b'))
# Error bars
ax.hlines(y, df_f['lo'], df_f['hi'], color=colors, linewidth=1.5, alpha=0.9)
ax.scatter(df_f['OR'], y,
s=55, marker='o',
facecolors=np.where(sig_mask, colors, 'white'),
edgecolors=colors, linewidths=1.5, zorder=3)
ax.axvline(1, color='#444', linestyle='--', linewidth=0.8, alpha=0.7)
ax.set_xscale('log')
ax.set_xlim(0.05, 50)
ax.set_yticks(y)
ax.set_yticklabels(df_f['var'], fontsize=8)
ax.set_xlabel('Odds Ratio (log scale)', fontsize=9)
ax.set_title(f'{site} (n predictors = {len(df_f)})', fontsize=10, fontweight='bold')
ax.grid(axis='x', linestyle=':', alpha=0.4)
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
# Hide empty subplots
for ax_idx in range(n_sites, n_rows * n_cols):
r, c = divmod(ax_idx, n_cols)
axes[r][c].set_visible(False)
# Legend
from matplotlib.lines import Line2D
legend_elems = [
Line2D([0], [0], marker='o', linestyle='None', markersize=7,
markerfacecolor='#c0392b', markeredgecolor='#c0392b', label='OR > 1, p < 0.05 (harmful)'),
Line2D([0], [0], marker='o', linestyle='None', markersize=7,
markerfacecolor='#1f6f8b', markeredgecolor='#1f6f8b', label='OR < 1, p < 0.05 (protective)'),
Line2D([0], [0], marker='o', linestyle='None', markersize=7,
markerfacecolor='white', markeredgecolor='#9e9e9e', label='Not significant (p ≥ 0.05)'),
Line2D([0], [0], color='#444', linestyle='--', linewidth=1, label='OR = 1 (null)'),
]
fig.legend(handles=legend_elems, loc='lower center', ncol=4, frameon=False,
bbox_to_anchor=(0.5, -0.01), fontsize=10)
fig.suptitle('Forest plot — GEE Logistic Mortality Models: Odds Ratios [95% CI] by Cancer Site',
fontsize=13, fontweight='bold', y=1.00)
plt.tight_layout(rect=[0, 0.02, 1, 0.98])
plt.savefig(os.path.join(OUTPUT_DIR, 'forest_gee_logistic_mortality.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-forest-gee-logistic.png", dpi=300, bbox_inches="tight")
plt.show()
# Store GEE LOS model results
gee_los_models = {}
gee_los_results = {}
def build_gee_los_model(df_site, site_name):
"""Build GEE Negative Binomial regression (mixed-effects approximation) for LOS.
Uses hospital_id as cluster group, year_factor as fixed effect,
and only significant comorbidities from bivariate analysis."""
df_model = df_site[df_site['dias_estancia'] > 0].copy()
df_model['SEXO_coded'] = (df_model['SEXO'] == 'Female').astype(int)
service_dummies = pd.get_dummies(df_model['service_category'], prefix='service', drop_first=True)
for col in service_dummies.columns:
df_model[col] = service_dummies[col]
service_dummy_cols = list(service_dummies.columns)
year_dummies = pd.get_dummies(df_model['year_factor'], prefix='year', drop_first=True)
for col in year_dummies.columns:
df_model[col] = year_dummies[col]
year_dummy_cols = list(year_dummies.columns)
df_model['edad'] = pd.to_numeric(df_model['edad'], errors='coerce')
predictor_cols = ['edad', 'SEXO_coded'] + service_dummy_cols + year_dummy_cols
# Add severity and mortality risk as CATEGORICAL dummies
for col in ['IR_29301_SEVERIDAD', 'IR_29301_MORTALIDAD']:
if col in df_model.columns:
df_model[col] = pd.to_numeric(df_model[col], errors='coerce')
df_model[col] = df_model[col].fillna(1)
df_model[col] = df_model[col].clip(lower=1).astype(int)
prefix = 'severity' if 'SEVERIDAD' in col else 'mort_risk'
col_dummies = pd.get_dummies(df_model[col], prefix=prefix, drop_first=True)
for dc in col_dummies.columns:
df_model[dc] = col_dummies[dc]
predictor_cols.append(dc)
# Add ONLY significant comorbidities from bivariate Wilcoxon test (p < 0.05)
sig_comorbs = sig_comorbidities_los.get(site_name, [])
used_comorbs = []
for col in sig_comorbs:
if col in df_model.columns:
df_model[col] = pd.to_numeric(df_model[col], errors='coerce').fillna(0).astype(int)
if df_model[col].sum() >= 3:
predictor_cols.append(col)
used_comorbs.append(col)
df_model['_group'] = df_model['hospital_id']
cols_for_model = ['dias_estancia', '_group'] + predictor_cols
df_gee = df_model[cols_for_model].dropna()
n_clusters = df_gee['_group'].nunique()
if len(df_gee) < 50 or n_clusters < 2:
return None, None, [], []
y = df_gee['dias_estancia'].astype(float)
X = df_gee[predictor_cols].astype(float)
X = sm.add_constant(X)
groups = df_gee['_group']
# Fit GEE with Negative Binomial family and exchangeable correlation
try:
gee_model = GEE(y, X, groups=groups,
family=NegativeBinomial(alpha=1.0),
cov_struct=Exchangeable())
gee_result = gee_model.fit(maxiter=200)
except Exception:
try:
gee_model = GEE(y, X, groups=groups,
family=NegativeBinomial(alpha=1.0),
cov_struct=Independence())
gee_result = gee_model.fit(maxiter=200)
except Exception:
try:
gee_model = GEE(y, X, groups=groups,
family=Poisson(),
cov_struct=Exchangeable())
gee_result = gee_model.fit(maxiter=200)
except Exception:
return None, None, [], []
# Remove problematic variables
max_retries = 5
current_predictors = list(predictor_cols)
for _ in range(max_retries):
params = gee_result.params
bad_vars = []
for i, var in enumerate(current_predictors):
idx = i + 1
if idx < len(params) and (np.isinf(params.iloc[idx]) or abs(params.iloc[idx]) > 50):
bad_vars.append(var)
if not bad_vars:
break
current_predictors = [p for p in current_predictors if p not in bad_vars]
if len(current_predictors) < 2:
return None, None, [], []
X = df_gee[current_predictors].astype(float)
X = sm.add_constant(X)
try:
gee_model = GEE(y, X, groups=groups, family=NegativeBinomial(alpha=1.0), cov_struct=Exchangeable())
gee_result = gee_model.fit(maxiter=200)
except:
try:
gee_model = GEE(y, X, groups=groups, family=NegativeBinomial(alpha=1.0), cov_struct=Independence())
gee_result = gee_model.fit(maxiter=200)
except:
try:
gee_model = GEE(y, X, groups=groups, family=Poisson(), cov_struct=Exchangeable())
gee_result = gee_model.fit(maxiter=200)
except:
return None, None, [], []
return gee_result, df_gee, current_predictors, used_comorbs
# Fit GEE LOS models for EACH cancer site
for site in all_sites:
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site].copy()
if len(df_site) < 50:
continue
gee_result, df_gee, used_cols, used_comorbs = build_gee_los_model(df_site, site)
if gee_result is not None:
gee_los_models[site] = (gee_result, df_gee, used_cols)
gee_los_results[site] = gee_result# Table 11: GEE Regression - IRR [95% CI] by Cancer Site
gee_irr_table = {}
for site in all_sites:
if site not in gee_los_results:
continue
result = gee_los_results[site]
irr_ci_list = {}
param_names = list(result.params.index)
ci = result.conf_int()
def _fmt_irr(v):
if np.isinf(v) or abs(v) > 1e6:
return 'Inf' if v > 0 else '-Inf'
return f"{v:.3f}"
for var in param_names:
coef = result.params[var]
irr_val = np.exp(coef)
ci_lower = np.exp(ci.loc[var, 0] if var in ci.index else ci.iloc[param_names.index(var), 0])
ci_upper = np.exp(ci.loc[var, 1] if var in ci.index else ci.iloc[param_names.index(var), 1])
pval = result.pvalues[var] if var in result.pvalues.index else result.pvalues.iloc[param_names.index(var)]
if np.isnan(irr_val) or np.isnan(pval):
irr_ci_list[format_name(var)] = '—'
continue
sig_stars = ''
if pval < 0.001:
sig_stars = '***'
elif pval < 0.01:
sig_stars = '**'
elif pval < 0.05:
sig_stars = '*'
irr_ci_str = f"{_fmt_irr(irr_val)} [{_fmt_irr(ci_lower)}, {_fmt_irr(ci_upper)}]{sig_stars}"
irr_ci_list[format_name(var)] = irr_ci_str
gee_irr_table[site] = irr_ci_list
gee_irr_df = pd.DataFrame(gee_irr_table).fillna('—')
show_table(gee_irr_df, "Table 11: GEE Mixed-Effects Regression - IRR [95% CI] for Length of Stay by Cancer Site\n(Hospital as cluster, year as fixed effect, Negative Binomial family, only significant comorbidities)\n* p<0.05, ** p<0.01, *** p<0.001", index=True)| Hypopharynx | Larynx | Nasopharynx | Oral Cavity | Oropharynx | Other Pharynx | Salivary Glands | Sinonasal | |
|---|---|---|---|---|---|---|---|---|
| Intercept | 2.530 [1.002, 6.390]* | 9.526 [6.003, 15.115]*** | 6.734 [3.032, 14.953]*** | 5.628 [3.551, 8.920]*** | 5.804 [2.892, 11.649]*** | 7.964 [4.455, 14.234]*** | 3.986 [2.629, 6.045]*** | 4.623 [2.875, 7.433]*** |
| Age | 1.009 [1.000, 1.019] | 0.998 [0.993, 1.003] | 1.005 [0.998, 1.013] | 1.001 [0.997, 1.004] | 1.001 [0.992, 1.009] | 1.002 [0.997, 1.007] | 1.002 [0.995, 1.008] | 0.999 [0.994, 1.003] |
| Sex (Female) | 0.828 [0.644, 1.065] | 0.781 [0.684, 0.893]*** | 0.760 [0.546, 1.057] | 0.856 [0.784, 0.934]*** | 0.953 [0.744, 1.222] | 1.288 [0.952, 1.743] | 0.819 [0.687, 0.975]* | 0.848 [0.741, 0.970]* |
| Service ITU | 2.636 [1.467, 4.734]** | 1.069 [0.799, 1.429] | 0.554 [0.342, 0.897]* | 1.014 [0.753, 1.366] | 0.850 [0.504, 1.433] | 0.924 [0.286, 2.985] | 0.988 [0.727, 1.342] | 1.556 [0.887, 2.728] |
| Service MS | 1.475 [0.887, 2.451] | 0.731 [0.578, 0.924]** | 1.132 [0.695, 1.845] | 0.884 [0.645, 1.211] | 0.663 [0.489, 0.899]** | 0.645 [0.442, 0.943]* | 1.053 [0.808, 1.371] | 1.315 [0.922, 1.876] |
| Service Other | 1.368 [0.825, 2.269] | 0.707 [0.544, 0.919]** | 0.985 [0.666, 1.459] | 0.773 [0.550, 1.087] | 0.842 [0.558, 1.269] | 0.611 [0.419, 0.890]* | 1.039 [0.811, 1.333] | 1.168 [0.810, 1.684] |
| Year 2020 | 0.913 [0.618, 1.349] | 1.199 [1.008, 1.425]* | 0.878 [0.338, 2.282] | 1.218 [0.894, 1.659] | 1.122 [0.878, 1.434] | 1.114 [0.725, 1.714] | 0.987 [0.741, 1.316] | 1.023 [0.784, 1.333] |
| Year 2021 | 1.111 [0.886, 1.395] | 1.129 [0.939, 1.359] | 0.530 [0.324, 0.865]* | 1.096 [0.854, 1.407] | 1.020 [0.797, 1.305] | 0.949 [0.605, 1.487] | 1.171 [0.814, 1.684] | 1.035 [0.783, 1.367] |
| Year 2022 | 1.179 [0.735, 1.893] | 1.248 [0.971, 1.604] | 0.712 [0.404, 1.253] | 1.047 [0.813, 1.347] | 1.090 [0.840, 1.415] | 0.610 [0.414, 0.900]* | 0.822 [0.603, 1.121] | 1.043 [0.829, 1.313] |
| Year 2023 | 1.053 [0.711, 1.558] | 0.928 [0.776, 1.110] | 0.611 [0.342, 1.092] | 1.010 [0.743, 1.373] | 1.097 [0.855, 1.407] | 0.381 [0.279, 0.521]*** | 0.866 [0.592, 1.267] | 0.897 [0.653, 1.234] |
| Year 2024 | 0.695 [0.480, 1.006] | 1.105 [0.912, 1.339] | 0.843 [0.535, 1.327] | 0.977 [0.802, 1.190] | 1.191 [0.896, 1.582] | 0.890 [0.614, 1.290] | 0.896 [0.615, 1.306] | 0.890 [0.690, 1.148] |
| Severity Medium | 1.394 [1.120, 1.734]** | 1.898 [1.466, 2.457]*** | 1.410 [0.995, 1.997] | 1.862 [1.555, 2.228]*** | 1.679 [1.314, 2.146]*** | 1.743 [1.241, 2.448]** | 1.446 [1.121, 1.866]** | 1.682 [1.286, 2.201]*** |
| Severity High | 1.982 [1.420, 2.766]*** | 2.259 [1.732, 2.947]*** | 3.194 [2.005, 5.087]*** | 3.003 [2.408, 3.745]*** | 2.435 [1.839, 3.225]*** | 2.350 [1.641, 3.364]*** | 3.575 [2.434, 5.250]*** | 3.174 [2.028, 4.969]*** |
| Mortality Risk Medium | 1.153 [0.853, 1.559] | 0.868 [0.770, 0.978]* | 0.929 [0.587, 1.470] | 1.113 [0.930, 1.333] | 0.891 [0.732, 1.085] | 0.850 [0.645, 1.120] | 1.195 [0.963, 1.485] | 0.931 [0.735, 1.180] |
| Mortality Risk High | 1.463 [1.038, 2.062]* | 1.306 [1.109, 1.539]** | 0.905 [0.514, 1.591] | 1.152 [0.877, 1.512] | 1.033 [0.775, 1.376] | 1.112 [0.731, 1.691] | 1.254 [0.928, 1.695] | 1.030 [0.721, 1.472] |
| CV Peripheral Vascular Disease | 2.125 [0.977, 4.623] | 1.069 [0.830, 1.378] | 1.107 [0.613, 1.998] | 1.371 [1.064, 1.768]* | 1.361 [0.877, 2.112] | 1.706 [0.990, 2.940] | — | 0.995 [0.595, 1.663] |
| RESP COPD | 1.070 [0.627, 1.825] | 1.076 [0.911, 1.270] | — | 1.078 [0.859, 1.354] | 1.713 [1.009, 2.908]* | — | 1.016 [0.638, 1.616] | — |
| RESP Emphysema | 2.468 [1.125, 5.416]* | 1.396 [1.120, 1.741]** | — | 1.097 [0.597, 2.016] | — | — | — | — |
| REN Acute Renal Failure | 0.989 [0.486, 2.014] | 0.897 [0.719, 1.120] | 0.667 [0.356, 1.248] | 0.884 [0.690, 1.134] | 1.179 [0.726, 1.916] | — | 0.837 [0.523, 1.341] | 2.194 [1.302, 3.698]** |
| REN Chronic Renal Failure | 1.003 [0.596, 1.688] | 1.044 [0.867, 1.256] | — | 1.138 [0.921, 1.406] | — | 1.549 [0.858, 2.794] | 0.909 [0.631, 1.310] | — |
| REN End Stage Renal | 0.870 [0.526, 1.440] | — | — | 0.966 [0.615, 1.518] | — | — | — | 1.025 [0.632, 1.664] |
| NEURO Epilepsy | 2.212 [1.453, 3.366]*** | 1.064 [0.748, 1.513] | — | 0.949 [0.662, 1.362] | — | — | — | — |
| SUST Alcohol | 1.378 [1.010, 1.882]* | 1.223 [0.938, 1.593] | 0.941 [0.405, 2.187] | 1.218 [0.961, 1.544] | 1.209 [0.946, 1.546] | — | — | — |
| SUST Tobacco | 1.158 [0.778, 1.723] | 1.176 [1.055, 1.312]** | — | 1.153 [0.994, 1.338] | 1.116 [0.872, 1.427] | 1.745 [1.194, 2.550]** | — | — |
| CV Essential Hypertension | — | 1.012 [0.913, 1.121] | — | 0.933 [0.861, 1.011] | 1.054 [0.933, 1.190] | — | 1.045 [0.910, 1.200] | — |
| CV Complicated Hypertension | — | 1.421 [1.026, 1.969]* | — | — | — | — | 1.231 [0.805, 1.882] | — |
| CV Coronary Disease | — | 0.909 [0.764, 1.081] | — | 1.110 [0.839, 1.468] | 1.229 [0.816, 1.851] | — | 0.833 [0.487, 1.426] | — |
| CV Heart Failure | — | 0.941 [0.675, 1.310] | — | 0.727 [0.563, 0.938]* | — | — | 0.935 [0.498, 1.753] | 1.395 [0.878, 2.216] |
| CV Arrhythmias | — | 1.104 [0.934, 1.305] | 0.986 [0.636, 1.528] | 1.099 [0.922, 1.311] | 1.240 [0.881, 1.745] | 2.644 [1.821, 3.838]*** | 1.236 [0.802, 1.906] | 1.563 [1.129, 2.163]** |
| CV Cerebrovascular Disease | — | 1.311 [0.812, 2.116] | 1.964 [0.821, 4.699] | 0.900 [0.491, 1.647] | 1.093 [0.637, 1.876] | — | 1.972 [0.630, 6.169] | 0.829 [0.562, 1.223] |
| MET Metabolic Syndrome | — | 2.376 [1.386, 4.076]** | — | 1.285 [0.750, 2.201] | — | — | — | — |
| HEP Alcoholic Cirrhosis | — | 1.176 [0.806, 1.716] | — | — | — | — | — | — |
| GI Peptic Ulcer | — | 1.693 [1.345, 2.130]*** | — | 0.899 [0.632, 1.279] | — | — | — | — |
| NEURO Stroke Sequelae | — | 0.855 [0.484, 1.509] | — | 1.008 [0.520, 1.954] | — | — | 0.579 [0.169, 1.984] | — |
| PSI Major Depression | — | 1.374 [1.121, 1.684]** | — | 1.383 [1.012, 1.888]* | 1.804 [1.078, 3.019]* | — | — | — |
| PSI Anxiety | — | 1.595 [1.007, 2.528]* | — | 2.083 [1.335, 3.252]** | — | — | — | — |
| MET Type2 Diabetes | — | — | 1.222 [0.759, 1.966] | — | — | — | — | 1.110 [0.924, 1.332] |
| MET Obesity | — | — | — | 1.264 [1.061, 1.507]** | — | — | — | — |
| MET Thyroid Disorders | — | — | — | 1.029 [0.897, 1.181] | — | — | — | — |
| RESP Sleep Apnea | — | — | — | 2.220 [1.306, 3.772]** | — | — | — | — |
| INF HIV | — | — | — | 1.481 [0.786, 2.793] | 0.521 [0.221, 1.230] | — | — | — |
| RESP Pulmonary Fibrosis | — | — | — | — | 1.157 [0.629, 2.127] | — | 0.939 [0.523, 1.687] | — |
| HEP Steatosis | — | — | — | — | 1.687 [0.787, 3.618] | — | 2.245 [0.851, 5.924] | — |
| PSI Bipolar | — | — | — | — | 1.135 [0.551, 2.337] | — | — | — |
# Table 12: Performance Metrics - GEE LOS Models
gee_los_perf = []
for site in all_sites:
if site not in gee_los_results:
continue
result = gee_los_results[site]
df_gee = gee_los_models[site][1]
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
n_total = len(df_site)
try:
y_pred = result.fittedvalues
y_actual = df_gee['dias_estancia'].values
rmse = np.sqrt(np.mean((y_actual - y_pred) ** 2))
mae = np.mean(np.abs(y_actual - y_pred))
ss_res = np.sum((y_actual - y_pred) ** 2)
ss_tot = np.sum((y_actual - np.mean(y_actual)) ** 2)
pseudo_r2 = 1 - ss_res / ss_tot if ss_tot > 0 else np.nan
except Exception:
rmse = mae = pseudo_r2 = np.nan
n_sig_comorbs = len(sig_comorbidities_los.get(site, []))
fit_metrics = get_gee_fit_metrics(result, df_gee['dias_estancia'], family_type='count')
gee_los_perf.append({
'Cancer Site': site,
'N obs': fit_metrics['N_obs'],
'N clusters': fit_metrics['N_clusters'],
'N params': fit_metrics['N_params'],
'Mean LOS': f'{df_site["dias_estancia"].mean():.1f}',
'Median LOS': f'{df_site["dias_estancia"].median():.0f}',
'Sig. Comorb.': n_sig_comorbs,
'Correlation': fit_metrics['Correlation'],
'QIC': safe_fmt(fit_metrics['QIC'], '.1f'),
'QICu': safe_fmt(fit_metrics['QICu'], '.1f'),
'Pseudo R²': safe_fmt(fit_metrics['Pseudo R²']),
'Scale': safe_fmt(fit_metrics['Scale']),
'RMSE': f'{rmse:.2f}' if not np.isnan(rmse) else '—',
'MAE': f'{mae:.2f}' if not np.isnan(mae) else '—',
})
gee_los_perf_df = pd.DataFrame(gee_los_perf)
show_table(gee_los_perf_df, "Table 12: Performance & Fit Metrics - GEE Regression Models for LOS\n(QIC = Quasi-Information Criterion, Negative Binomial/Poisson family)", index=False)| Cancer Site | N obs | N clusters | N params | Mean LOS | Median LOS | Sig. Comorb. | Correlation | QIC | QICu | Pseudo R² | Scale | RMSE | MAE |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypopharynx | 388 | 49 | 24 | 13.7 | 7 | 9 | Exchangeable | 23454.4 | 384.7 | 0.2763 | 1.0000 | 16.26 | 10.26 |
| Larynx | 2398 | 64 | 35 | 14.5 | 7 | 20 | Exchangeable | 56825.0 | 2587.5 | 0.1611 | 1.0000 | 20.50 | 12.44 |
| Nasopharynx | 304 | 49 | 21 | 9.0 | 4 | 6 | Exchangeable | 16628.3 | 330.6 | 0.1315 | 1.0000 | 18.14 | 9.21 |
| Oral Cavity | 2575 | 67 | 38 | 9.6 | 4 | 23 | Exchangeable | 38797.2 | 2530.5 | 0.1196 | 1.0000 | 16.14 | 8.95 |
| Oropharynx | 822 | 59 | 29 | 8.4 | 4 | 14 | Exchangeable | 21087.4 | 767.1 | 0.1652 | 1.0000 | 13.02 | 7.56 |
| Other Pharynx | 252 | 48 | 19 | 10.8 | 5 | 4 | Exchangeable | 14079.0 | 235.0 | 0.1606 | 1.0000 | 15.03 | 9.22 |
| Salivary Glands | 800 | 58 | 27 | 6.9 | 4 | 12 | Exchangeable | 11993.7 | 615.4 | 0.1378 | 1.0000 | 11.20 | 5.84 |
| Sinonasal | 727 | 55 | 22 | 6.7 | 3 | 7 | Exchangeable | 10252.5 | 577.6 | 0.2372 | 1.0000 | 10.48 | 5.93 |
# Reuses _extract_or_ci defined in §6.2 — coefficients exponentiate to IRRs for log-link NB models.
forest_data_los = {site: _extract_or_ci(res) for site, res in gee_los_results.items()}
forest_data_los = {s: d for s, d in forest_data_los.items() if not d.empty}
n_sites = len(forest_data_los)
n_cols = 3
n_rows = int(np.ceil(n_sites / n_cols))
fig, axes = plt.subplots(n_rows, n_cols, figsize=(16, 4 * n_rows), squeeze=False)
for ax_idx, (site, df_f) in enumerate(forest_data_los.items()):
r, c = divmod(ax_idx, n_cols)
ax = axes[r][c]
df_f = df_f.sort_values('OR').reset_index(drop=True) # OR field carries IRR here
y = np.arange(len(df_f))
sig_mask = (df_f['p'] < 0.05)
# IRR > 1 means longer LOS (worse), IRR < 1 means shorter
colors = np.where(~sig_mask, '#9e9e9e',
np.where(df_f['OR'] > 1, '#b8860b', '#2e7d32'))
ax.hlines(y, df_f['lo'], df_f['hi'], color=colors, linewidth=1.5, alpha=0.9)
ax.scatter(df_f['OR'], y,
s=55, marker='s',
facecolors=np.where(sig_mask, colors, 'white'),
edgecolors=colors, linewidths=1.5, zorder=3)
ax.axvline(1, color='#444', linestyle='--', linewidth=0.8, alpha=0.7)
ax.set_xscale('log')
ax.set_xlim(0.3, 8)
ax.set_yticks(y)
ax.set_yticklabels(df_f['var'], fontsize=8)
ax.set_xlabel('Incidence Rate Ratio (log scale)', fontsize=9)
ax.set_title(f'{site} (n predictors = {len(df_f)})', fontsize=10, fontweight='bold')
ax.grid(axis='x', linestyle=':', alpha=0.4)
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
for ax_idx in range(n_sites, n_rows * n_cols):
r, c = divmod(ax_idx, n_cols)
axes[r][c].set_visible(False)
from matplotlib.lines import Line2D
legend_elems = [
Line2D([0], [0], marker='s', linestyle='None', markersize=7,
markerfacecolor='#b8860b', markeredgecolor='#b8860b', label='IRR > 1, p < 0.05 (longer stay)'),
Line2D([0], [0], marker='s', linestyle='None', markersize=7,
markerfacecolor='#2e7d32', markeredgecolor='#2e7d32', label='IRR < 1, p < 0.05 (shorter stay)'),
Line2D([0], [0], marker='s', linestyle='None', markersize=7,
markerfacecolor='white', markeredgecolor='#9e9e9e', label='Not significant (p ≥ 0.05)'),
Line2D([0], [0], color='#444', linestyle='--', linewidth=1, label='IRR = 1 (null)'),
]
fig.legend(handles=legend_elems, loc='lower center', ncol=4, frameon=False,
bbox_to_anchor=(0.5, -0.01), fontsize=10)
fig.suptitle('Forest plot — GEE Negative-Binomial LOS Models: IRR [95% CI] by Cancer Site',
fontsize=13, fontweight='bold', y=1.00)
plt.tight_layout(rect=[0, 0.02, 1, 0.98])
plt.savefig(os.path.join(OUTPUT_DIR, 'forest_gee_nb_los.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-forest-gee-los.png", dpi=300, bbox_inches="tight")
plt.show()
Machine learning models for per-site mortality prediction using multiple class balancing techniques.
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.svm import SVC
from sklearn.neural_network import MLPClassifier
from sklearn.linear_model import LogisticRegression as LR_sklearn
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import (
roc_auc_score, roc_curve, confusion_matrix as cm_func,
accuracy_score, f1_score, precision_score, recall_score,
brier_score_loss, matthews_corrcoef
)
from collections import OrderedDict
try:
from imblearn.over_sampling import SMOTE, ADASYN, RandomOverSampler
from imblearn.under_sampling import RandomUnderSampler
from imblearn.combine import SMOTETomek
IMBLEARN_AVAILABLE = True
except ImportError:
IMBLEARN_AVAILABLE = False
# Feature columns: age, sex, service dummies, ALL comorbidities
ml_comorbidity_cols = [c for c in df_head_neck.columns if c in comorbidity_cols]
ALL_BALANCING_NAMES = ['None (Original)', 'SMOTE', 'ADASYN', 'SMOTETomek',
'Random Oversampling', 'Random Undersampling']
def get_balancing_methods(y_train):
"""Return ALL class balancing methods, adapting k_neighbors to minority class size."""
methods = OrderedDict()
methods['None (Original)'] = None
if not IMBLEARN_AVAILABLE:
for name in ALL_BALANCING_NAMES[1:]:
methods[name] = 'imblearn not installed'
return methods
n_minority = int(y_train.sum())
k = min(5, n_minority - 1) if n_minority > 1 else 0
if k >= 1:
methods['SMOTE'] = SMOTE(random_state=42, k_neighbors=k)
methods['ADASYN'] = ADASYN(random_state=42, n_neighbors=min(k, n_minority - 1))
methods['SMOTETomek'] = SMOTETomek(random_state=42, smote=SMOTE(random_state=42, k_neighbors=k))
else:
methods['SMOTE'] = f'Insufficient minority samples (n={n_minority}, need k>=1)'
methods['ADASYN'] = f'Insufficient minority samples (n={n_minority}, need k>=1)'
methods['SMOTETomek'] = f'Insufficient minority samples (n={n_minority}, need k>=1)'
methods['Random Oversampling'] = RandomOverSampler(random_state=42)
methods['Random Undersampling'] = RandomUnderSampler(random_state=42)
return methods
def _failed_balancing_entry(reason):
"""Return a placeholder result dict for a failed balancing technique."""
return {
'model': None,
'y_pred': None,
'y_proba': None,
'y_test': None,
'opt_threshold': np.nan,
'cm': None,
'failed': True,
'fail_reason': reason,
'metrics': {
'AUC': np.nan, 'Bal. Accuracy': np.nan, 'Sensitivity': np.nan,
'Specificity': np.nan, 'PPV': np.nan, 'NPV': np.nan,
'F1': np.nan, 'MCC': np.nan, 'Brier': np.nan,
"Youden's J": np.nan, 'Threshold': np.nan,
'Train Size': '—', 'Minority %': '—'
}
}
def run_ml_mortality_multi_balancing(df_site, site_name):
"""Train ML classifiers with multiple class balancing techniques + optimal threshold"""
df_ml = df_site.copy()
df_ml['sex_female'] = (df_ml['SEXO'] == 'Female').astype(int)
df_ml['edad'] = pd.to_numeric(df_ml['edad'], errors='coerce')
svc_dummies = pd.get_dummies(df_ml['service_category'], prefix='service', drop_first=True)
for c in svc_dummies.columns:
df_ml[c] = svc_dummies[c]
feature_names = ['edad', 'sex_female'] + list(svc_dummies.columns) + ml_comorbidity_cols
target = 'egresar_fallecido'
df_ml = df_ml[feature_names + [target]].dropna()
n_pos = int(df_ml[target].sum())
n_neg = int((df_ml[target] == 0).sum())
if len(df_ml) < 30 or n_pos < 5 or n_neg < 5:
return None, None, None
X = df_ml[feature_names].values
y = df_ml[target].values
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
try:
X_train, X_test, y_train, y_test = train_test_split(
X_scaled, y, test_size=0.2, random_state=42, stratify=y
)
except ValueError:
X_train, X_test, y_train, y_test = train_test_split(
X_scaled, y, test_size=0.2, random_state=42
)
balancing_methods = get_balancing_methods(y_train)
base_model_class = GradientBoostingClassifier
base_model_params = {'n_estimators': 150, 'random_state': 42, 'max_depth': 4}
balancing_results = OrderedDict()
for bal_name, balancer in balancing_methods.items():
if isinstance(balancer, str):
balancing_results[bal_name] = _failed_balancing_entry(balancer)
continue
try:
if balancer is not None:
X_res, y_res = balancer.fit_resample(X_train, y_train)
else:
X_res, y_res = X_train, y_train
model = base_model_class(**base_model_params)
model.fit(X_res, y_res)
y_proba = model.predict_proba(X_test)[:, 1]
fpr_t, tpr_t, thresholds_t = roc_curve(y_test, y_proba)
j_scores = tpr_t - fpr_t
best_idx = np.argmax(j_scores)
opt_threshold = thresholds_t[best_idx]
y_pred = (y_proba >= opt_threshold).astype(int)
tn, fp, fn, tp = cm_func(y_test, y_pred).ravel()
sens = tp / (tp + fn) if (tp + fn) > 0 else 0
spec = tn / (tn + fp) if (tn + fp) > 0 else 0
ppv_val = tp / (tp + fp) if (tp + fp) > 0 else 0
npv_val = tn / (tn + fn) if (tn + fn) > 0 else 0
balancing_results[bal_name] = {
'model': model,
'y_pred': y_pred,
'y_proba': y_proba,
'y_test': y_test,
'opt_threshold': opt_threshold,
'cm': cm_func(y_test, y_pred),
'failed': False,
'metrics': {
'AUC': roc_auc_score(y_test, y_proba) if len(np.unique(y_test)) > 1 else 0,
'Bal. Accuracy': (sens + spec) / 2,
'Sensitivity': sens,
'Specificity': spec,
'PPV': ppv_val,
'NPV': npv_val,
'F1': f1_score(y_test, y_pred, zero_division=0),
'MCC': matthews_corrcoef(y_test, y_pred),
'Brier': brier_score_loss(y_test, y_proba),
"Youden's J": sens + spec - 1,
'Threshold': opt_threshold,
'Train Size': len(y_res),
'Minority %': f'{y_res.sum()/len(y_res)*100:.1f}'
}
}
except Exception as e:
balancing_results[bal_name] = _failed_balancing_entry(str(e))
successful_results = {k: v for k, v in balancing_results.items() if not v.get('failed', False)}
if not successful_results:
return balancing_results, None, None
best_balancing = max(successful_results, key=lambda k: successful_results[k]['metrics']['AUC'])
best_balancer = balancing_methods.get(best_balancing)
if best_balancer is not None:
try:
X_best, y_best = best_balancer.fit_resample(X_train, y_train)
except:
X_best, y_best = X_train, y_train
else:
X_best, y_best = X_train, y_train
all_models = {
'Logistic Regression': LR_sklearn(max_iter=1000, random_state=42, class_weight='balanced'),
'Random Forest': RandomForestClassifier(n_estimators=200, random_state=42, n_jobs=-1, class_weight='balanced'),
'Gradient Boosting': GradientBoostingClassifier(n_estimators=150, random_state=42, max_depth=4),
'SVM': SVC(kernel='rbf', probability=True, random_state=42, class_weight='balanced'),
'MLP': MLPClassifier(hidden_layer_sizes=(64, 32), max_iter=1000, random_state=42)
}
model_results = {}
for name, model in all_models.items():
try:
model.fit(X_best, y_best)
y_proba = model.predict_proba(X_test)[:, 1]
fpr_t, tpr_t, thresholds_t = roc_curve(y_test, y_proba)
j_scores = tpr_t - fpr_t
best_idx = np.argmax(j_scores)
opt_threshold = thresholds_t[best_idx]
y_pred = (y_proba >= opt_threshold).astype(int)
tn, fp, fn, tp = cm_func(y_test, y_pred).ravel()
sens = tp / (tp + fn) if (tp + fn) > 0 else 0
spec = tn / (tn + fp) if (tn + fp) > 0 else 0
ppv_val = tp / (tp + fp) if (tp + fp) > 0 else 0
npv_val = tn / (tn + fn) if (tn + fn) > 0 else 0
model_results[name] = {
'model': model,
'y_pred': y_pred,
'y_proba': y_proba,
'y_test': y_test,
'opt_threshold': opt_threshold,
'cm': cm_func(y_test, y_pred),
'metrics': {
'AUC': roc_auc_score(y_test, y_proba) if len(np.unique(y_test)) > 1 else 0,
'Bal. Accuracy': (sens + spec) / 2,
'Sensitivity': sens,
'Specificity': spec,
'PPV': ppv_val,
'NPV': npv_val,
'F1': f1_score(y_test, y_pred, zero_division=0),
'MCC': matthews_corrcoef(y_test, y_pred),
'Brier': brier_score_loss(y_test, y_proba),
"Youden's J": sens + spec - 1,
'Threshold': opt_threshold
}
}
except:
pass
return balancing_results, model_results, (feature_names, X_test, y_test, best_balancing)
# Run ML mortality models for EACH cancer site
ml_balancing_by_site = {}
ml_models_by_site = {}
ml_meta_by_site = {}
for site in all_sites:
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site].copy()
bal_results, model_results, meta = run_ml_mortality_multi_balancing(df_site, site)
if bal_results is not None and len(bal_results) > 0:
ml_balancing_by_site[site] = bal_results
if model_results is not None:
ml_models_by_site[site] = model_results
if meta is not None:
ml_meta_by_site[site] = meta# Table: Class balancing technique comparison for each site (Gradient Boosting as base model)
for site in all_sites:
if site not in ml_balancing_by_site:
continue
results = ml_balancing_by_site[site]
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
n_total = len(df_site)
n_events = df_site['egresar_fallecido'].sum()
metrics_df = pd.DataFrame({name: d['metrics'] for name, d in results.items()}).T
metrics_df.index.name = 'Balancing'
status_col = []
for name in metrics_df.index:
entry = results[name]
if entry.get('failed', False):
reason = entry.get('fail_reason', 'Error')
status_col.append(reason[:40])
else:
status_col.append('OK')
metrics_df.insert(0, 'Status', status_col)
for col in metrics_df.columns:
if col in ['Train Size', 'Minority %', 'Status']:
metrics_df[col] = metrics_df[col].apply(lambda x: '—' if (isinstance(x, float) and np.isnan(x)) else x)
else:
metrics_df[col] = metrics_df[col].apply(lambda x: '—' if (isinstance(x, float) and (np.isnan(x) or np.isinf(x))) else f'{x:.4f}')
show_table(metrics_df,
f"Class Balancing Comparison (Gradient Boosting): {site} (N={n_total:,}, Events={n_events}, {n_events/n_total*100:.1f}%)",
index=True)| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.5978 | 0.6918 | 1.0000 | 0.3836 | 0.1346 | 1.0000 | 0.2373 | 0.2272 | 0.1059 | 0.3836 | 0.0043 | 320 | 8.1 |
| SMOTE | OK | 0.6213 | 0.7094 | 0.8571 | 0.5616 | 0.1579 | 0.9762 | 0.2667 | 0.2370 | 0.1132 | 0.4188 | 0.0303 | 588 | 50.0 |
| ADASYN | OK | 0.6204 | 0.6996 | 0.7143 | 0.6849 | 0.1786 | 0.9615 | 0.2857 | 0.2365 | 0.1121 | 0.3992 | 0.0892 | 584 | 49.7 |
| SMOTETomek | OK | 0.6115 | 0.6996 | 0.7143 | 0.6849 | 0.1786 | 0.9615 | 0.2857 | 0.2365 | 0.1157 | 0.3992 | 0.0848 | 580 | 50.0 |
| Random Oversampling | OK | 0.7006 | 0.7162 | 0.8571 | 0.5753 | 0.1622 | 0.9767 | 0.2727 | 0.2451 | 0.1124 | 0.4325 | 0.0272 | 588 | 50.0 |
| Random Undersampling | OK | 0.4403 | 0.6233 | 1.0000 | 0.2466 | 0.1129 | 1.0000 | 0.2029 | 0.1669 | 0.4881 | 0.2466 | 0.0050 | 52 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.6275 | 0.6255 | 0.3500 | 0.9010 | 0.2258 | 0.9438 | 0.2745 | 0.2064 | 0.0749 | 0.2510 | 0.1495 | 2100 | 7.5 |
| SMOTE | OK | 0.5740 | 0.5933 | 0.5000 | 0.6866 | 0.1163 | 0.9433 | 0.1887 | 0.1055 | 0.0915 | 0.1866 | 0.1589 | 3884 | 50.0 |
| ADASYN | OK | 0.5877 | 0.5872 | 0.9250 | 0.2495 | 0.0923 | 0.9758 | 0.1678 | 0.1090 | 0.0939 | 0.1745 | 0.0457 | 3875 | 49.9 |
| SMOTETomek | OK | 0.5847 | 0.5955 | 0.5250 | 0.6660 | 0.1148 | 0.9444 | 0.1883 | 0.1063 | 0.0864 | 0.1910 | 0.1306 | 3744 | 50.0 |
| Random Oversampling | OK | 0.5728 | 0.5729 | 0.5500 | 0.5959 | 0.1009 | 0.9414 | 0.1705 | 0.0785 | 0.1381 | 0.1459 | 0.2486 | 3884 | 50.0 |
| Random Undersampling | OK | 0.6431 | 0.6343 | 0.4500 | 0.8186 | 0.1698 | 0.9475 | 0.2466 | 0.1775 | 0.2642 | 0.2686 | 0.8045 | 316 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.6391 | 0.6681 | 0.8000 | 0.5362 | 0.1111 | 0.9737 | 0.1951 | 0.1689 | 0.0792 | 0.3362 | 0.0028 | 294 | 6.1 |
| SMOTE | OK | 0.7130 | 0.7130 | 0.6000 | 0.8261 | 0.2000 | 0.9661 | 0.3000 | 0.2660 | 0.0721 | 0.4261 | 0.0615 | 552 | 50.0 |
| ADASYN | OK | 0.7130 | 0.7203 | 0.6000 | 0.8406 | 0.2143 | 0.9667 | 0.3158 | 0.2824 | 0.0788 | 0.4406 | 0.0720 | 552 | 50.0 |
| SMOTETomek | OK | 0.4986 | 0.6130 | 0.4000 | 0.8261 | 0.1429 | 0.9500 | 0.2105 | 0.1449 | 0.0894 | 0.2261 | 0.0526 | 548 | 50.0 |
| Random Oversampling | OK | 0.5855 | 0.6739 | 1.0000 | 0.3478 | 0.1000 | 1.0000 | 0.1818 | 0.1865 | 0.1080 | 0.3478 | 0.0028 | 552 | 50.0 |
| Random Undersampling | OK | 0.7188 | 0.7203 | 0.6000 | 0.8406 | 0.2143 | 0.9667 | 0.3158 | 0.2824 | 0.2460 | 0.4406 | 0.9935 | 36 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.7222 | 0.6762 | 0.5714 | 0.7809 | 0.1389 | 0.9672 | 0.2235 | 0.1933 | 0.0602 | 0.3523 | 0.0448 | 2400 | 5.9 |
| SMOTE | OK | 0.5746 | 0.6083 | 0.6000 | 0.6166 | 0.0882 | 0.9614 | 0.1538 | 0.1037 | 0.0974 | 0.2166 | 0.2029 | 4518 | 50.0 |
| ADASYN | OK | 0.5835 | 0.6074 | 0.7714 | 0.4435 | 0.0789 | 0.9691 | 0.1432 | 0.1016 | 0.0952 | 0.2149 | 0.1380 | 4497 | 49.8 |
| SMOTETomek | OK | 0.5723 | 0.6197 | 0.7429 | 0.4965 | 0.0836 | 0.9690 | 0.1503 | 0.1122 | 0.0953 | 0.2393 | 0.1244 | 4254 | 50.0 |
| Random Oversampling | OK | 0.6285 | 0.6327 | 0.3714 | 0.8940 | 0.1781 | 0.9583 | 0.2407 | 0.1903 | 0.1181 | 0.2654 | 0.5121 | 4518 | 50.0 |
| Random Undersampling | OK | 0.7183 | 0.6905 | 0.6000 | 0.7809 | 0.1448 | 0.9693 | 0.2333 | 0.2085 | 0.2441 | 0.3809 | 0.6778 | 282 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.5856 | 0.6553 | 0.7273 | 0.5833 | 0.0964 | 0.9722 | 0.1702 | 0.1460 | 0.0679 | 0.3106 | 0.0104 | 760 | 5.8 |
| SMOTE | OK | 0.5144 | 0.6152 | 0.3636 | 0.8667 | 0.1429 | 0.9571 | 0.2051 | 0.1517 | 0.0881 | 0.2303 | 0.3646 | 1432 | 50.0 |
| ADASYN | OK | 0.4437 | 0.5568 | 0.3636 | 0.7500 | 0.0816 | 0.9507 | 0.1333 | 0.0606 | 0.0970 | 0.1136 | 0.2403 | 1439 | 50.2 |
| SMOTETomek | OK | 0.5333 | 0.6141 | 0.2727 | 0.9556 | 0.2727 | 0.9556 | 0.2727 | 0.2283 | 0.0857 | 0.2283 | 0.5336 | 1366 | 50.0 |
| Random Oversampling | OK | 0.5609 | 0.6348 | 0.6364 | 0.6333 | 0.0959 | 0.9661 | 0.1667 | 0.1293 | 0.0827 | 0.2697 | 0.0871 | 1432 | 50.0 |
| Random Undersampling | OK | 0.6447 | 0.6323 | 0.9091 | 0.3556 | 0.0794 | 0.9846 | 0.1460 | 0.1301 | 0.3244 | 0.2646 | 0.0567 | 88 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.5605 | 0.6275 | 0.6667 | 0.5882 | 0.1600 | 0.9375 | 0.2581 | 0.1576 | 0.1279 | 0.2549 | 0.0028 | 224 | 10.7 |
| SMOTE | OK | 0.5343 | 0.6275 | 0.6667 | 0.5882 | 0.1600 | 0.9375 | 0.2581 | 0.1576 | 0.1219 | 0.2549 | 0.0165 | 400 | 50.0 |
| ADASYN | OK | 0.5245 | 0.6471 | 0.6667 | 0.6275 | 0.1739 | 0.9412 | 0.2759 | 0.1840 | 0.1247 | 0.2941 | 0.0138 | 401 | 50.1 |
| SMOTETomek | OK | 0.5033 | 0.6373 | 0.6667 | 0.6078 | 0.1667 | 0.9394 | 0.2667 | 0.1706 | 0.1251 | 0.2745 | 0.0132 | 398 | 50.0 |
| Random Oversampling | OK | 0.5605 | 0.6618 | 0.8333 | 0.4902 | 0.1613 | 0.9615 | 0.2703 | 0.1993 | 0.1570 | 0.3235 | 0.0134 | 400 | 50.0 |
| Random Undersampling | OK | 0.6961 | 0.7255 | 0.6667 | 0.7843 | 0.2667 | 0.9524 | 0.3810 | 0.3143 | 0.3035 | 0.4510 | 0.9682 | 48 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.6556 | 0.6871 | 0.8889 | 0.4854 | 0.0833 | 0.9881 | 0.1524 | 0.1635 | 0.0597 | 0.3743 | 0.0048 | 718 | 5.2 |
| SMOTE | OK | 0.6699 | 0.6696 | 0.8889 | 0.4503 | 0.0784 | 0.9872 | 0.1441 | 0.1492 | 0.0709 | 0.3392 | 0.0117 | 1362 | 50.0 |
| ADASYN | OK | 0.6322 | 0.6696 | 0.5556 | 0.7836 | 0.1190 | 0.9710 | 0.1961 | 0.1748 | 0.0717 | 0.3392 | 0.0833 | 1362 | 50.0 |
| SMOTETomek | OK | 0.6576 | 0.6550 | 0.6667 | 0.6433 | 0.0896 | 0.9735 | 0.1579 | 0.1397 | 0.0777 | 0.3099 | 0.0224 | 1342 | 50.0 |
| Random Oversampling | OK | 0.6374 | 0.6462 | 0.4444 | 0.8480 | 0.1333 | 0.9667 | 0.2051 | 0.1710 | 0.0924 | 0.2924 | 0.2315 | 1362 | 50.0 |
| Random Undersampling | OK | 0.7609 | 0.7632 | 0.8889 | 0.6374 | 0.1143 | 0.9909 | 0.2025 | 0.2353 | 0.3080 | 0.5263 | 0.5186 | 74 | 50.0 |
| Status | AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | Train Size | Minority % | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Balancing | ||||||||||||||
| None (Original) | OK | 0.5537 | 0.6110 | 0.8750 | 0.3471 | 0.0593 | 0.9833 | 0.1111 | 0.0973 | 0.0572 | 0.2221 | 0.0040 | 709 | 4.7 |
| SMOTE | OK | 0.6118 | 0.6191 | 0.7500 | 0.4882 | 0.0645 | 0.9765 | 0.1188 | 0.0988 | 0.0831 | 0.2382 | 0.0440 | 1352 | 50.0 |
| ADASYN | OK | 0.5882 | 0.6199 | 0.8750 | 0.3647 | 0.0609 | 0.9841 | 0.1138 | 0.1039 | 0.0725 | 0.2397 | 0.0244 | 1346 | 49.8 |
| SMOTETomek | OK | 0.5978 | 0.6110 | 0.8750 | 0.3471 | 0.0593 | 0.9833 | 0.1111 | 0.0973 | 0.0874 | 0.2221 | 0.0229 | 1306 | 50.0 |
| Random Oversampling | OK | 0.6485 | 0.7441 | 1.0000 | 0.4882 | 0.0842 | 1.0000 | 0.1553 | 0.2028 | 0.0780 | 0.4882 | 0.0398 | 1352 | 50.0 |
| Random Undersampling | OK | 0.6504 | 0.6838 | 0.7500 | 0.6176 | 0.0845 | 0.9813 | 0.1519 | 0.1556 | 0.3468 | 0.3676 | 0.7202 | 66 | 50.0 |
# Table: ML model comparison with best balancing for each site
for site in all_sites:
if site not in ml_models_by_site or not ml_models_by_site[site]:
continue
results = ml_models_by_site[site]
meta = ml_meta_by_site.get(site)
best_balancing = meta[3] if meta else 'Unknown'
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
n_total = len(df_site)
n_events = df_site['egresar_fallecido'].sum()
metrics_df = pd.DataFrame({name: d['metrics'] for name, d in results.items()}).T
metrics_df.index.name = 'Model'
for col in metrics_df.columns:
metrics_df[col] = metrics_df[col].apply(lambda x: '—' if (isinstance(x, float) and (np.isnan(x) or np.isinf(x))) else f'{x:.4f}')
show_table(metrics_df,
f"ML Models: {site} (N={n_total:,}, Events={n_events}, Best balancing: {best_balancing})",
index=True)| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.4521 | 0.6135 | 0.8571 | 0.3699 | 0.1154 | 0.9643 | 0.2034 | 0.1345 | 0.2670 | 0.2270 | 0.1141 |
| Random Forest | 0.4843 | 0.5900 | 0.7143 | 0.4658 | 0.1136 | 0.9444 | 0.1961 | 0.1023 | 0.1359 | 0.1800 | 0.0300 |
| Gradient Boosting | 0.7006 | 0.7162 | 0.8571 | 0.5753 | 0.1622 | 0.9767 | 0.2727 | 0.2451 | 0.1124 | 0.4325 | 0.0272 |
| SVM | 0.6047 | 0.7025 | 0.8571 | 0.5479 | 0.1538 | 0.9756 | 0.2609 | 0.2290 | 0.1689 | 0.4051 | 0.1543 |
| MLP | 0.5616 | 0.6644 | 1.0000 | 0.3288 | 0.1250 | 1.0000 | 0.2222 | 0.2027 | 0.2396 | 0.3288 | 0.0000 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.6627 | 0.6673 | 0.4500 | 0.8845 | 0.2432 | 0.9512 | 0.3158 | 0.2551 | 0.2306 | 0.3345 | 0.7886 |
| Random Forest | 0.6885 | 0.6921 | 0.6750 | 0.7093 | 0.1607 | 0.9636 | 0.2596 | 0.2186 | 0.2232 | 0.3843 | 0.5500 |
| Gradient Boosting | 0.6431 | 0.6343 | 0.4500 | 0.8186 | 0.1698 | 0.9475 | 0.2466 | 0.1775 | 0.2642 | 0.2686 | 0.8045 |
| SVM | 0.6565 | 0.6736 | 0.4750 | 0.8722 | 0.2346 | 0.9527 | 0.3140 | 0.2550 | 0.2144 | 0.3472 | 0.7175 |
| MLP | 0.6755 | 0.6771 | 0.5500 | 0.8041 | 0.1880 | 0.9559 | 0.2803 | 0.2258 | 0.3072 | 0.3541 | 0.9930 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.6551 | 0.7420 | 0.6000 | 0.8841 | 0.2727 | 0.9683 | 0.3750 | 0.3415 | 0.2441 | 0.4841 | 0.8362 |
| Random Forest | 0.7594 | 0.7826 | 1.0000 | 0.5652 | 0.1429 | 1.0000 | 0.2500 | 0.2842 | 0.2258 | 0.5652 | 0.4450 |
| Gradient Boosting | 0.7188 | 0.7203 | 0.6000 | 0.8406 | 0.2143 | 0.9667 | 0.3158 | 0.2824 | 0.2460 | 0.4406 | 0.9935 |
| SVM | 0.4145 | 0.6638 | 0.4000 | 0.9275 | 0.2857 | 0.9552 | 0.3333 | 0.2809 | 0.3762 | 0.3275 | 0.7654 |
| MLP | 0.8435 | 0.8551 | 1.0000 | 0.7101 | 0.2000 | 1.0000 | 0.3333 | 0.3769 | 0.4234 | 0.7101 | 0.9973 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.7013 | 0.6648 | 0.4286 | 0.9011 | 0.2113 | 0.9623 | 0.2830 | 0.2392 | 0.1958 | 0.3296 | 0.6417 |
| Random Forest | 0.6390 | 0.6323 | 0.4571 | 0.8074 | 0.1280 | 0.9601 | 0.2000 | 0.1527 | 0.0611 | 0.2646 | 0.0700 |
| Gradient Boosting | 0.7222 | 0.6762 | 0.5714 | 0.7809 | 0.1389 | 0.9672 | 0.2235 | 0.1933 | 0.0602 | 0.3523 | 0.0448 |
| SVM | 0.6413 | 0.6750 | 0.8571 | 0.4929 | 0.0946 | 0.9824 | 0.1705 | 0.1642 | 0.0543 | 0.3501 | 0.0524 |
| MLP | 0.4907 | 0.5626 | 0.1429 | 0.9823 | 0.3333 | 0.9488 | 0.2000 | 0.1879 | 0.0653 | 0.1252 | 0.7080 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.7404 | 0.7098 | 0.6364 | 0.7833 | 0.1522 | 0.9724 | 0.2456 | 0.2287 | 0.2346 | 0.4197 | 0.6146 |
| Random Forest | 0.6662 | 0.6987 | 0.6364 | 0.7611 | 0.1400 | 0.9716 | 0.2295 | 0.2106 | 0.2210 | 0.3975 | 0.6050 |
| Gradient Boosting | 0.6447 | 0.6323 | 0.9091 | 0.3556 | 0.0794 | 0.9846 | 0.1460 | 0.1301 | 0.3244 | 0.2646 | 0.0567 |
| SVM | 0.8288 | 0.8341 | 0.8182 | 0.8500 | 0.2500 | 0.9871 | 0.3830 | 0.3980 | 0.2084 | 0.6682 | 0.6107 |
| MLP | 0.7449 | 0.7341 | 0.8182 | 0.6500 | 0.1250 | 0.9832 | 0.2169 | 0.2251 | 0.3003 | 0.4682 | 0.5443 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.6340 | 0.6863 | 1.0000 | 0.3725 | 0.1579 | 1.0000 | 0.2727 | 0.2425 | 0.2163 | 0.3725 | 0.2220 |
| Random Forest | 0.6830 | 0.7598 | 0.8333 | 0.6863 | 0.2381 | 0.9722 | 0.3704 | 0.3306 | 0.2156 | 0.5196 | 0.5350 |
| Gradient Boosting | 0.6961 | 0.7255 | 0.6667 | 0.7843 | 0.2667 | 0.9524 | 0.3810 | 0.3143 | 0.3035 | 0.4510 | 0.9682 |
| SVM | 0.1961 | 0.5392 | 1.0000 | 0.0784 | 0.1132 | 1.0000 | 0.2034 | 0.0942 | 0.3557 | 0.0784 | 0.4605 |
| MLP | 0.4248 | 0.5392 | 0.3333 | 0.7451 | 0.1333 | 0.9048 | 0.1905 | 0.0547 | 0.3116 | 0.0784 | 0.6873 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.6004 | 0.6637 | 0.5556 | 0.7719 | 0.1136 | 0.9706 | 0.1887 | 0.1661 | 0.3319 | 0.3275 | 0.8033 |
| Random Forest | 0.7840 | 0.7953 | 0.8889 | 0.7018 | 0.1356 | 0.9917 | 0.2353 | 0.2742 | 0.2239 | 0.5906 | 0.5550 |
| Gradient Boosting | 0.7609 | 0.7632 | 0.8889 | 0.6374 | 0.1143 | 0.9909 | 0.2025 | 0.2353 | 0.3080 | 0.5263 | 0.5186 |
| SVM | 0.6283 | 0.6345 | 1.0000 | 0.2690 | 0.0672 | 1.0000 | 0.1259 | 0.1344 | 0.2393 | 0.2690 | 0.3582 |
| MLP | 0.6244 | 0.6696 | 0.5556 | 0.7836 | 0.1190 | 0.9710 | 0.1961 | 0.1748 | 0.3154 | 0.3392 | 0.9770 |
| AUC | Bal. Accuracy | Sensitivity | Specificity | PPV | NPV | F1 | MCC | Brier | Youden's J | Threshold | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | |||||||||||
| Logistic Regression | 0.5618 | 0.6360 | 0.6250 | 0.6471 | 0.0769 | 0.9735 | 0.1370 | 0.1171 | 0.2597 | 0.2721 | 0.4821 |
| Random Forest | 0.5688 | 0.7037 | 0.6250 | 0.7824 | 0.1190 | 0.9779 | 0.2000 | 0.1988 | 0.2729 | 0.4074 | 0.6750 |
| Gradient Boosting | 0.6504 | 0.6838 | 0.7500 | 0.6176 | 0.0845 | 0.9813 | 0.1519 | 0.1556 | 0.3468 | 0.3676 | 0.7202 |
| SVM | 0.5687 | 0.6132 | 0.2500 | 0.9765 | 0.3333 | 0.9651 | 0.2857 | 0.2600 | 0.2503 | 0.2265 | 0.6229 |
| MLP | 0.6037 | 0.6404 | 0.3750 | 0.9059 | 0.1579 | 0.9686 | 0.2222 | 0.1885 | 0.3521 | 0.2809 | 1.0000 |
n_sites_ml = len(ml_models_by_site)
if n_sites_ml > 0:
ncols = min(3, n_sites_ml)
nrows = (n_sites_ml + ncols - 1) // ncols
fig, axes = plt.subplots(nrows, ncols, figsize=(6 * ncols, 5 * nrows))
if n_sites_ml == 1:
axes = np.array([axes])
axes = axes.flatten() if hasattr(axes, 'flatten') else [axes]
for idx, (site, results) in enumerate(ml_models_by_site.items()):
if not results:
continue
ax = axes[idx]
best_balancing = ml_meta_by_site[site][3] if ml_meta_by_site.get(site) else ''
for name, data in results.items():
y_test = data['y_test']
y_proba = data['y_proba']
fpr, tpr, _ = roc_curve(y_test, y_proba)
auc_val = data['metrics']['AUC']
ax.plot(fpr, tpr, label=f'{name} ({auc_val:.3f})', linewidth=2)
ax.plot([0, 1], [0, 1], 'k--', linewidth=1)
ax.set_xlabel('FPR')
ax.set_ylabel('TPR')
ax.set_title(f'{site}\n(Balancing: {best_balancing})', fontweight='bold', fontsize=10)
ax.legend(fontsize=7, loc='lower right')
ax.grid(True, alpha=0.3)
for idx in range(len(ml_models_by_site), len(axes)):
axes[idx].set_visible(False)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'roc_ml_mortality_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-roc-ml-mortality.png", dpi=300, bbox_inches="tight")
plt.show()
if len(ml_balancing_by_site) > 0:
comparison_data = []
for site, results in ml_balancing_by_site.items():
for bal_name, data in results.items():
failed = data.get('failed', False)
comparison_data.append({
'Cancer Site': site,
'Balancing': bal_name,
'AUC': 0.0 if failed else data['metrics']['AUC'],
'Bal. Accuracy': 0.0 if failed else data['metrics']['Bal. Accuracy'],
'F1': 0.0 if failed else data['metrics']['F1'],
'Failed': failed
})
comp_df = pd.DataFrame(comparison_data)
bal_order = [b for b in ALL_BALANCING_NAMES if b in comp_df['Balancing'].unique()]
comp_df['Balancing'] = pd.Categorical(comp_df['Balancing'], categories=bal_order, ordered=True)
fig, axes = plt.subplots(1, 3, figsize=(14, 8))
metrics_to_plot = ['AUC', 'Bal. Accuracy', 'F1']
colors_resamp = plt.cm.Set2(np.linspace(0, 1, len(bal_order)))
for ax_idx, metric in enumerate(metrics_to_plot):
ax = axes[ax_idx]
pivot = comp_df.pivot(index='Cancer Site', columns='Balancing', values=metric)
pivot = pivot[[c for c in bal_order if c in pivot.columns]]
pivot.fillna(0).plot(kind='bar', ax=ax, color=colors_resamp[:len(pivot.columns)])
ax.set_title(metric, fontweight='bold')
ax.set_ylabel(metric)
ax.legend(fontsize=6, loc='lower right')
ax.grid(True, alpha=0.3, axis='y')
plt.setp(ax.get_xticklabels(), rotation=30, ha='right', fontsize=8)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'balancing_comparison_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-balancing-comparison.png", dpi=300, bbox_inches="tight")
plt.show()
n_sites_ml = len(ml_models_by_site)
if n_sites_ml > 0:
ncols = min(4, n_sites_ml)
nrows = (n_sites_ml + ncols - 1) // ncols
fig, axes = plt.subplots(nrows, ncols, figsize=(4 * ncols, 4 * nrows))
if n_sites_ml == 1:
axes = np.array([axes])
axes = axes.flatten() if hasattr(axes, 'flatten') else [axes]
for idx, (site, results) in enumerate(ml_models_by_site.items()):
if not results:
continue
best_name = max(results, key=lambda k: results[k]['metrics']['AUC'])
cm = results[best_name]['cm']
ax = axes[idx]
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues', ax=ax, cbar=False,
xticklabels=['Survived', 'Deceased'], yticklabels=['Survived', 'Deceased'])
ax.set_title(f'{site}\n({best_name})', fontsize=10, fontweight='bold')
ax.set_ylabel('True')
ax.set_xlabel('Predicted')
for idx in range(len(ml_models_by_site), len(axes)):
axes[idx].set_visible(False)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'cm_ml_mortality_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-cm-ml-mortality.png", dpi=300, bbox_inches="tight")
plt.show()
sites_with_gb = [s for s in ml_models_by_site if ml_models_by_site[s] and 'Gradient Boosting' in ml_models_by_site[s]]
n_gb = len(sites_with_gb)
if n_gb > 0:
ncols = min(3, n_gb)
nrows = (n_gb + ncols - 1) // ncols
fig, axes = plt.subplots(nrows, ncols, figsize=(6 * ncols, 5 * nrows))
if n_gb == 1:
axes = np.array([axes])
axes = axes.flatten() if hasattr(axes, 'flatten') else [axes]
for idx, site in enumerate(sites_with_gb):
model = ml_models_by_site[site]['Gradient Boosting']['model']
feat_names = ml_meta_by_site[site][0]
importances = model.feature_importances_
formatted = [format_name(f) for f in feat_names]
imp_df = pd.DataFrame({'Feature': formatted, 'Importance': importances})
imp_df = imp_df.sort_values('Importance', ascending=True).tail(12)
ax = axes[idx]
ax.barh(imp_df['Feature'], imp_df['Importance'], color='steelblue')
ax.set_xlabel('Importance')
ax.set_title(f'{site}', fontweight='bold')
ax.grid(True, alpha=0.3, axis='x')
for idx in range(n_gb, len(axes)):
axes[idx].set_visible(False)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'feat_importance_mortality_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-feat-importance-mortality.png", dpi=300, bbox_inches="tight")
plt.show()
from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor
from sklearn.linear_model import LinearRegression, Ridge, Lasso
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
def run_ml_los_for_site(df_site, site_name):
"""Train 5 ML regressors for LOS on a single cancer site"""
df_ml = df_site[df_site['dias_estancia'] > 0].copy()
df_ml['sex_female'] = (df_ml['SEXO'] == 'Female').astype(int)
df_ml['edad'] = pd.to_numeric(df_ml['edad'], errors='coerce')
svc_dummies = pd.get_dummies(df_ml['service_category'], prefix='service', drop_first=True)
for c in svc_dummies.columns:
df_ml[c] = svc_dummies[c]
feature_names = ['edad', 'sex_female'] + list(svc_dummies.columns) + ml_comorbidity_cols
target = 'dias_estancia'
df_ml = df_ml[feature_names + [target]].dropna()
if len(df_ml) < 80:
return None, None
X = df_ml[feature_names].values
y = df_ml[target].values
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
X_train, X_test, y_train, y_test = train_test_split(
X_scaled, y, test_size=0.2, random_state=42
)
models = {
'Linear Regression': LinearRegression(),
'Ridge': Ridge(alpha=1.0),
'Lasso': Lasso(alpha=0.1, max_iter=5000),
'Random Forest': RandomForestRegressor(n_estimators=100, random_state=42, n_jobs=-1),
'Gradient Boosting': GradientBoostingRegressor(n_estimators=100, random_state=42)
}
results = {}
for name, model in models.items():
try:
model.fit(X_train, y_train)
y_pred = model.predict(X_test)
results[name] = {
'model': model,
'y_pred': y_pred,
'y_test': y_test,
'metrics': {
'RMSE': np.sqrt(mean_squared_error(y_test, y_pred)),
'MAE': mean_absolute_error(y_test, y_pred),
'R2': r2_score(y_test, y_pred),
'MSE': mean_squared_error(y_test, y_pred)
}
}
except:
pass
return results, feature_names
# Run ML LOS models for EACH cancer site
ml_los_by_site = {}
ml_los_features_by_site = {}
for site in all_sites:
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site].copy()
results, feat_names = run_ml_los_for_site(df_site, site)
if results and len(results) > 0:
ml_los_by_site[site] = results
ml_los_features_by_site[site] = feat_names# Table: ML LOS Performance by Site
for site in all_sites:
if site not in ml_los_by_site:
continue
results = ml_los_by_site[site]
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
n_total = len(df_site)
mean_los = df_site['dias_estancia'].mean()
metrics_df = pd.DataFrame({name: d['metrics'] for name, d in results.items()}).T
metrics_df.index.name = 'Model'
for col in metrics_df.columns:
metrics_df[col] = metrics_df[col].apply(lambda x: '—' if (isinstance(x, float) and (np.isnan(x) or np.isinf(x))) else f'{x:.4f}')
show_table(metrics_df,
f"ML LOS Performance: {site} (N={n_total:,}, Mean LOS={mean_los:.1f} days)",
index=True)| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 22.2782 | 13.9487 | -0.4538 | 496.3178 |
| Ridge | 22.2354 | 13.9280 | -0.4483 | 494.4112 |
| Lasso | 21.9121 | 13.6609 | -0.4065 | 480.1414 |
| Random Forest | 19.4278 | 12.5932 | -0.1056 | 377.4393 |
| Gradient Boosting | 20.6044 | 13.4355 | -0.2436 | 424.5431 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 19.5764 | 13.3325 | -0.0275 | 383.2339 |
| Ridge | 19.5748 | 13.3316 | -0.0274 | 383.1739 |
| Lasso | 19.4355 | 13.2411 | -0.0128 | 377.7388 |
| Random Forest | 21.3782 | 14.2926 | -0.2254 | 457.0262 |
| Gradient Boosting | 20.3111 | 13.8083 | -0.1061 | 412.5399 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 10.8355 | 7.9235 | -0.1720 | 117.4088 |
| Ridge | 10.7736 | 7.8920 | -0.1587 | 116.0715 |
| Lasso | 10.2645 | 7.5827 | -0.0517 | 105.3589 |
| Random Forest | 11.1671 | 7.6256 | -0.2449 | 124.7033 |
| Gradient Boosting | 13.0054 | 8.0567 | -0.6884 | 169.1396 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 18.8474 | 10.6064 | -0.0307 | 355.2235 |
| Ridge | 18.8466 | 10.6061 | -0.0307 | 355.1952 |
| Lasso | 18.7448 | 10.5509 | -0.0195 | 351.3677 |
| Random Forest | 20.5552 | 11.3571 | -0.2260 | 422.5151 |
| Gradient Boosting | 19.0218 | 10.6592 | -0.0499 | 361.8286 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 13.9983 | 8.5612 | -0.1791 | 195.9520 |
| Ridge | 13.9887 | 8.5563 | -0.1774 | 195.6842 |
| Lasso | 13.8397 | 8.4215 | -0.1525 | 191.5368 |
| Random Forest | 13.5869 | 8.4933 | -0.1108 | 184.6032 |
| Gradient Boosting | 13.9424 | 8.5364 | -0.1697 | 194.3902 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 12.2293 | 8.8816 | -0.2828 | 149.5557 |
| Ridge | 12.1605 | 8.8576 | -0.2684 | 147.8770 |
| Lasso | 11.7386 | 8.6341 | -0.1819 | 137.7949 |
| Random Forest | 11.3697 | 7.8188 | -0.1088 | 129.2695 |
| Gradient Boosting | 14.7723 | 9.3805 | -0.8717 | 218.2223 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 9.9316 | 6.2590 | -0.2303 | 98.6370 |
| Ridge | 9.9229 | 6.2540 | -0.2282 | 98.4633 |
| Lasso | 9.5267 | 6.0424 | -0.1320 | 90.7571 |
| Random Forest | 9.3527 | 6.2639 | -0.0911 | 87.4735 |
| Gradient Boosting | 9.3368 | 6.1751 | -0.0874 | 87.1758 |
| RMSE | MAE | R2 | MSE | |
|---|---|---|---|---|
| Model | ||||
| Linear Regression | 13.5113 | 6.8173 | 0.0791 | 182.5560 |
| Ridge | 13.5123 | 6.8185 | 0.0790 | 182.5820 |
| Lasso | 13.5369 | 6.8838 | 0.0756 | 183.2484 |
| Random Forest | 13.5982 | 6.7620 | 0.0672 | 184.9117 |
| Gradient Boosting | 13.7705 | 6.8266 | 0.0434 | 189.6273 |
n_sites_los = len(ml_los_by_site)
if n_sites_los > 0:
ncols = min(3, n_sites_los)
nrows = (n_sites_los + ncols - 1) // ncols
fig, axes = plt.subplots(nrows, ncols, figsize=(6 * ncols, 5 * nrows))
if n_sites_los == 1:
axes = np.array([axes])
axes = axes.flatten() if hasattr(axes, 'flatten') else [axes]
for idx, (site, results) in enumerate(ml_los_by_site.items()):
best_name = max(results, key=lambda k: results[k]['metrics']['R2'])
best = results[best_name]
y_test = best['y_test']
y_pred = best['y_pred']
r2 = best['metrics']['R2']
ax = axes[idx]
ax.scatter(y_test, y_pred, alpha=0.4, s=15, color='steelblue')
lims = [min(y_test.min(), y_pred.min()), max(np.percentile(y_test, 95), np.percentile(y_pred, 95))]
ax.plot(lims, lims, 'r--', lw=2)
ax.set_xlabel('Observed LOS (days)')
ax.set_ylabel('Predicted LOS (days)')
ax.set_title(f'{site}\n{best_name} (R²={r2:.3f})', fontweight='bold', fontsize=10)
ax.grid(True, alpha=0.3)
for idx in range(n_sites_los, len(axes)):
axes[idx].set_visible(False)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'pred_vs_obs_los_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-pred-vs-obs-los.png", dpi=300, bbox_inches="tight")
plt.show()
sites_with_rf_los = [s for s in ml_los_by_site if 'Random Forest' in ml_los_by_site[s]]
n_rf_los = len(sites_with_rf_los)
if n_rf_los > 0:
ncols = min(3, n_rf_los)
nrows = (n_rf_los + ncols - 1) // ncols
fig, axes = plt.subplots(nrows, ncols, figsize=(6 * ncols, 5 * nrows))
if n_rf_los == 1:
axes = np.array([axes])
axes = axes.flatten() if hasattr(axes, 'flatten') else [axes]
for idx, site in enumerate(sites_with_rf_los):
model = ml_los_by_site[site]['Random Forest']['model']
feat_names = ml_los_features_by_site[site]
importances = model.feature_importances_
formatted = [format_name(f) for f in feat_names]
imp_df = pd.DataFrame({'Feature': formatted, 'Importance': importances})
imp_df = imp_df.sort_values('Importance', ascending=True).tail(12)
ax = axes[idx]
ax.barh(imp_df['Feature'], imp_df['Importance'], color='forestgreen')
ax.set_xlabel('Importance')
ax.set_title(f'{site}', fontweight='bold')
ax.grid(True, alpha=0.3, axis='x')
for idx in range(n_rf_los, len(axes)):
axes[idx].set_visible(False)
plt.tight_layout()
plt.savefig(os.path.join(OUTPUT_DIR, 'feat_importance_los_by_site.png'), dpi=300, bbox_inches='tight', facecolor='white')
plt.savefig(_FIG / "fig-feat-importance-los.png", dpi=300, bbox_inches="tight")
plt.show()
# Combined summary: best model per site for mortality and LOS
summary_rows = []
for site in all_sites:
row = {'Cancer Site': site}
df_site = df_head_neck[df_head_neck['cancer_site_group'] == site]
row['N'] = f'{len(df_site):,}'
row['Mortality %'] = f'{df_site["egresar_fallecido"].mean()*100:.1f}'
row['Mean LOS'] = f'{df_site["dias_estancia"].mean():.1f}'
# GEE Logistic regression AUC
if site in gee_mortality_results:
try:
result = gee_mortality_results[site]
df_gee = gee_mortality_models[site][1]
y_proba = result.fittedvalues
y_actual = df_gee['egresar_fallecido'].values
row['GEE Mortality AUC'] = f'{roc_auc_score(y_actual, y_proba):.3f}'
except:
row['GEE Mortality AUC'] = '—'
else:
row['GEE Mortality AUC'] = '—'
# Best ML mortality AUC (with class balancing)
if site in ml_models_by_site and ml_models_by_site[site]:
best_name = max(ml_models_by_site[site], key=lambda k: ml_models_by_site[site][k]['metrics']['AUC'])
best_auc = ml_models_by_site[site][best_name]['metrics']['AUC']
best_balancing = ml_meta_by_site[site][3] if ml_meta_by_site.get(site) else 'Unknown'
row['Best ML Mortality AUC'] = f'{best_auc:.3f} ({best_name}, {best_balancing})'
else:
row['Best ML Mortality AUC'] = '—'
# GEE Poisson/NB model RMSE
if site in gee_los_results:
try:
result = gee_los_results[site]
df_gee = gee_los_models[site][1]
y_pred = result.fittedvalues
y_actual = df_gee['dias_estancia'].values
rmse = np.sqrt(np.mean((y_actual - y_pred) ** 2))
row['GEE LOS RMSE'] = f'{rmse:.2f}'
except:
row['GEE LOS RMSE'] = '—'
else:
row['GEE LOS RMSE'] = '—'
# Best ML LOS R²
if site in ml_los_by_site:
best_name = max(ml_los_by_site[site], key=lambda k: ml_los_by_site[site][k]['metrics']['R2'])
best_r2 = ml_los_by_site[site][best_name]['metrics']['R2']
row['Best ML R² (LOS)'] = f'{best_r2:.3f} ({best_name})'
else:
row['Best ML R² (LOS)'] = '—'
summary_rows.append(row)
summary_df = pd.DataFrame(summary_rows)
show_table(summary_df, "Table 13: Summary - GEE & ML Model Performance Across All Cancer Sites\n(GEE: Mixed-Effects with hospital clustering; ML: Multiple Class Balancing Techniques)", index=False)| Cancer Site | N | Mortality % | Mean LOS | GEE Mortality AUC | Best ML Mortality AUC | GEE LOS RMSE | Best ML R² (LOS) |
|---|---|---|---|---|---|---|---|
| Hypopharynx | 400 | 8.2 | 13.7 | 0.862 | 0.701 (Gradient Boosting, Random Oversampling) | 16.26 | -0.106 (Random Forest) |
| Larynx | 2,625 | 7.5 | 14.5 | 0.859 | 0.688 (Random Forest, Random Undersampling) | 20.50 | -0.013 (Lasso) |
| Nasopharynx | 368 | 6.2 | 9.0 | — | 0.843 (MLP, Random Undersampling) | 18.14 | -0.052 (Lasso) |
| Oral Cavity | 3,001 | 5.9 | 9.6 | 0.897 | 0.722 (Gradient Boosting, None (Original)) | 16.14 | -0.020 (Lasso) |
| Oropharynx | 951 | 5.8 | 8.4 | 0.888 | 0.829 (SVM, Random Undersampling) | 13.02 | -0.111 (Random Forest) |
| Other Pharynx | 281 | 10.7 | 10.8 | 0.898 | 0.696 (Gradient Boosting, Random Undersampling) | 15.03 | -0.109 (Random Forest) |
| Salivary Glands | 898 | 5.1 | 6.9 | 0.935 | 0.784 (Random Forest, Random Undersampling) | 11.20 | -0.087 (Gradient Boosting) |
| Sinonasal | 887 | 4.6 | 6.7 | 0.898 | 0.650 (Gradient Boosting, Random Undersampling) | 10.48 | 0.079 (Linear Regression) |
This analysis of head and neck malignancies in Chile (2019-2024) provides independent regression and machine learning models for each cancer site, enabling site-specific risk assessment for both in-hospital mortality and length of stay. Head and neck cancer comprises diverse anatomical subsites (oral cavity, pharynx, larynx, sinonasal) with markedly different epidemiology, treatment complexity, and clinical outcomes.
Mixed-effects models were fit using Generalized Estimating Equations (GEE) to account for the hierarchical structure of observations within hospitals and years. For mortality, GEE logistic regression was used with Binomial family and exchangeable correlation structure. For length of stay, GEE negative binomial regression was used to account for overdispersion in count data, with Poisson as fallback. All models include age, sex, service category, severity/mortality indices, and only significant comorbidities from bivariate analysis (p < 0.05), improving parsimony and interpretability. Hospital was specified as the grouping variable to capture clustering effects.
To address class imbalance and improve generalization, multiple class balancing techniques were evaluated: SMOTE, ADASYN, SMOTETomek, random oversampling, random undersampling, and no balancing (original). For each cancer site, gradient boosting was used as the base model to identify the best-performing balancing method by AUC. Subsequently, five classifiers (Logistic Regression, Random Forest, Gradient Boosting, SVM, MLP) were trained using the optimal balancing strategy. All models used optimal decision thresholds derived from Youden’s J statistic rather than default 0.5 cutoff, improving balanced accuracy and clinical utility.
Head and neck cancer mortality and hospital resource utilization vary significantly by anatomical subsite and comorbidity burden. Treatment decisions (surgery vs. radiation vs. chemoradiation) and patient characteristics (age, performance status, comorbidities) directly influence hospital outcomes. These models provide quantitative tools for risk stratification and outcome prediction, supporting clinical decision-making and resource planning.
This comprehensive analysis of head and neck malignancies in Chilean public hospitals provides an epidemiological profile of in-hospital mortality and length of stay across multiple anatomical subsites (2019-2024). The results highlight the importance of site-specific modeling, the differential impact of comorbidities by location, and temporal patterns that may inform resource allocation and quality improvement initiatives in head and neck cancer care.