generate_dataset.py

#!/usr/bin/env python3
"""
Literature-Informed Synthetic TB Screening & Symptom Dataset Generator
======================================================================

Generates realistic synthetic datasets of patients undergoing tuberculosis
screening at LMIC health facilities, with demographic data, symptom profiles,
HIV co-infection status, diagnostic results (smear, GeneXpert, chest X-ray),
and TB classification.

Target population: Adults and adolescents (≥15 years) presenting for TB
screening at facility-based care in TB-endemic LMIC settings.

DAG (Sampling Order):
  1. age, sex (roots)
  2. hiv_status (scenario-dependent, age/sex-adjusted)
  3. bmi (conditional on age, TB status)
  4. true_tb_status (from prevalence, conditional on HIV)
  5. tb_type (pulmonary vs extrapulmonary)
  6. symptoms: cough, fever, night_sweats, weight_loss, hemoptysis
  7. smear_result, genexpert_result, cxr_finding (conditional on TB + test performance)
  8. tb_classification (derived)

References:
-----------
[1] WHO (2023). Global Tuberculosis Report 2023. Geneva.
[2] WHO (2021). WHO consolidated guidelines on tuberculosis: Module 3 -
    Diagnosis: Rapid diagnostics for TB detection. Geneva.
[3] Corbett EL, et al. (2003). The growing burden of tuberculosis: global
    trends and interactions with the HIV epidemic. Arch Intern Med, 163:1009.
[4] Steingart KR, et al. (2014). Xpert MTB/RIF assay for pulmonary TB and
    rifampicin resistance in adults. Cochrane Database Syst Rev, 1:CD009593.
[5] Getahun H, et al. (2007). Diagnosis of smear-negative pulmonary TB in
    people with HIV infection. Lancet Infect Dis, 7(4):238-246.
[6] WHO (2013). Systematic screening for active TB: principles and
    recommendations. Geneva.
[7] van't Hoog AH, et al. (2012). Screening strategies for TB prevalence
    surveys. PLoS One, 7(5):e36392.
[8] UNAIDS (2023). Global HIV & AIDS statistics fact sheet.
"""

import numpy as np
import pandas as pd
from scipy.stats import truncnorm
import argparse
import os

# ============================================================
# SECTION 1: Literature-Informed Parameters
# ============================================================

SCENARIOS = {
    'low_tb_burden': {
        'description': 'Lower TB incidence LMIC (e.g., urban Latin America)',
        'tb_prevalence_screened': 0.05,  # WHO 2023: 3-8% among symptomatic screened
        'hiv_prevalence': 0.03,
        'hiv_tb_coinfection': 0.15,      # 10-20% of TB is HIV+ in low-HIV settings
        'mdr_rate': 0.03,               # WHO 2023: 2-5% new cases
        'smear_positive_pct': 0.55,      # ~50-60% of PTB is smear+
    },
    'moderate_tb_burden': {
        'description': 'Moderate TB burden (e.g., Kenya, India, Philippines)',
        'tb_prevalence_screened': 0.12,  # WHO 2023: 8-15% among presumptive TB
        'hiv_prevalence': 0.08,
        'hiv_tb_coinfection': 0.30,      # Corbett 2003: 25-40% in SSA
        'mdr_rate': 0.05,
        'smear_positive_pct': 0.50,
    },
    'high_tb_burden': {
        'description': 'High TB/HIV burden (e.g., South Africa, Mozambique, DRC)',
        'tb_prevalence_screened': 0.22,  # WHO 2023: 15-30% in high-burden
        'hiv_prevalence': 0.20,
        'hiv_tb_coinfection': 0.50,      # Corbett 2003: 40-60% in high HIV
        'mdr_rate': 0.08,               # WHO 2023: up to 10% in hotspots
        'smear_positive_pct': 0.40,      # Lower smear+ in HIV coinfection
    },
}

# --- GeneXpert Performance ---
# Source: Steingart 2014, WHO 2021
XPERT_SENSITIVITY_SMEAR_POS = 0.98   # Smear-positive PTB
XPERT_SENSITIVITY_SMEAR_NEG = 0.68   # Smear-negative PTB
XPERT_SENSITIVITY_HIV = 0.79         # HIV-associated TB (pooled)
XPERT_SPECIFICITY = 0.98
XPERT_RIF_SENSITIVITY = 0.95         # For rifampicin resistance detection

# --- Smear Microscopy Performance ---
# Source: Getahun 2007
SMEAR_SENSITIVITY = 0.60             # Overall
SMEAR_SENSITIVITY_HIV = 0.40         # Reduced in HIV+
SMEAR_SPECIFICITY = 0.98

# --- CXR Performance ---
# Source: van't Hoog 2012
CXR_SENSITIVITY = 0.87               # For active PTB
CXR_SPECIFICITY = 0.70               # Many false positives (old TB, other pathology)

# --- Symptom Profiles ---
# Source: WHO 2013 systematic screening guidelines, van't Hoog 2012
SYMPTOM_PROB_TB = {
    'cough_2weeks': 0.75,     # Cough ≥2 weeks
    'fever': 0.65,
    'night_sweats': 0.55,
    'weight_loss': 0.60,
    'hemoptysis': 0.15,
    'chest_pain': 0.40,
    'fatigue': 0.70,
    'loss_of_appetite': 0.55,
}
SYMPTOM_PROB_NO_TB = {
    'cough_2weeks': 0.10,
    'fever': 0.15,
    'night_sweats': 0.05,
    'weight_loss': 0.08,
    'hemoptysis': 0.01,
    'chest_pain': 0.12,
    'fatigue': 0.20,
    'loss_of_appetite': 0.10,
}

# --- BMI ---
BMI_NORMAL = {'mean': 22.0, 'sd': 3.5}
BMI_TB = {'mean': 18.5, 'sd': 2.5}  # TB patients often underweight


# ============================================================
# SECTION 2: Utility Functions
# ============================================================

def trunc_normal(mean, sd, lo, hi, size, rng):
    a, b = (lo - mean) / sd, (hi - mean) / sd
    return truncnorm.rvs(a, b, loc=mean, scale=sd, size=size,
                         random_state=rng.integers(0, 2**31))


# ============================================================
# SECTION 3: Main Generator
# ============================================================

def generate_tb_dataset(n=10000, seed=42, scenario='moderate_tb_burden'):
    rng = np.random.default_rng(seed)
    sc = SCENARIOS[scenario]

    # ── Step 1: Demographics ──
    sex = rng.choice(['M', 'F'], size=n, p=[0.58, 0.42])  # TB M:F ~1.4:1 (WHO 2023)
    age = trunc_normal(38, 14, 15, 85, n, rng).astype(int)

    # ── Step 2: HIV status ──
    hiv_status = np.zeros(n, dtype=int)
    for i in range(n):
        p_hiv = sc['hiv_prevalence']
        if 20 <= age[i] <= 45:
            p_hiv *= 1.5  # Peak HIV age
        if sex[i] == 'F' and 20 <= age[i] <= 35:
            p_hiv *= 1.3  # Higher in young women in SSA
        hiv_status[i] = 1 if rng.random() < min(p_hiv, 0.50) else 0

    # ── Step 3: True TB status ──
    true_tb = np.zeros(n, dtype=int)
    for i in range(n):
        p_tb = sc['tb_prevalence_screened']
        if hiv_status[i]:
            p_tb *= 3.0  # HIV is strongest risk factor (Corbett 2003)
        if age[i] > 50:
            p_tb *= 1.3
        if sex[i] == 'M':
            p_tb *= 1.2
        true_tb[i] = 1 if rng.random() < min(p_tb, 0.60) else 0

    # ── Step 4: TB type and drug resistance ──
    tb_type = np.array(['none'] * n, dtype=object)
    rifampicin_resistant = np.zeros(n, dtype=int)
    for i in range(n):
        if true_tb[i]:
            # ~85% pulmonary, ~15% extrapulmonary (higher in HIV+)
            p_eptb = 0.25 if hiv_status[i] else 0.12
            tb_type[i] = 'extrapulmonary' if rng.random() < p_eptb else 'pulmonary'
            rifampicin_resistant[i] = 1 if rng.random() < sc['mdr_rate'] else 0

    # ── Step 5: BMI ──
    bmi = np.zeros(n)
    for i in range(n):
        if true_tb[i]:
            bmi[i] = rng.normal(BMI_TB['mean'], BMI_TB['sd'])
        else:
            bmi[i] = rng.normal(BMI_NORMAL['mean'], BMI_NORMAL['sd'])
    bmi = np.clip(np.round(bmi, 1), 12.0, 45.0)

    # ── Step 6: Symptoms ──
    symptoms = {}
    for sym in SYMPTOM_PROB_TB:
        symptoms[sym] = np.zeros(n, dtype=int)
        for i in range(n):
            p = SYMPTOM_PROB_TB[sym] if true_tb[i] else SYMPTOM_PROB_NO_TB[sym]
            # HIV+ TB patients may have fewer respiratory symptoms
            if true_tb[i] and hiv_status[i] and sym in ('cough_2weeks', 'hemoptysis'):
                p *= 0.80
            symptoms[sym][i] = 1 if rng.random() < p else 0

    # Cough duration (weeks)
    cough_duration_weeks = np.zeros(n, dtype=int)
    for i in range(n):
        if symptoms['cough_2weeks'][i]:
            if true_tb[i]:
                cough_duration_weeks[i] = max(2, rng.poisson(5))
            else:
                cough_duration_weeks[i] = max(2, rng.poisson(3))
    cough_duration_weeks = np.clip(cough_duration_weeks, 0, 52)

    # Number of WHO screening symptoms (cough ≥2wk, fever, night sweats, weight loss)
    who_symptom_count = (symptoms['cough_2weeks'] + symptoms['fever'] +
                          symptoms['night_sweats'] + symptoms['weight_loss'])

    # ── Step 7: Smear microscopy ──
    smear_result = np.zeros(n, dtype=int)
    for i in range(n):
        if true_tb[i] and tb_type[i] == 'pulmonary':
            sens = SMEAR_SENSITIVITY_HIV if hiv_status[i] else SMEAR_SENSITIVITY
            smear_result[i] = 1 if rng.random() < sens else 0
        elif true_tb[i] and tb_type[i] == 'extrapulmonary':
            smear_result[i] = 1 if rng.random() < 0.10 else 0  # Very low for EPTB
        else:
            smear_result[i] = 1 if rng.random() < (1 - SMEAR_SPECIFICITY) else 0

    # ── Step 8: GeneXpert MTB/RIF ──
    xpert_mtb = np.zeros(n, dtype=int)
    xpert_rif_resistant = np.array(['N/A'] * n, dtype=object)
    for i in range(n):
        if true_tb[i] and tb_type[i] == 'pulmonary':
            if smear_result[i]:
                sens = XPERT_SENSITIVITY_SMEAR_POS
            elif hiv_status[i]:
                sens = XPERT_SENSITIVITY_HIV
            else:
                sens = XPERT_SENSITIVITY_SMEAR_NEG
            xpert_mtb[i] = 1 if rng.random() < sens else 0
        elif true_tb[i] and tb_type[i] == 'extrapulmonary':
            xpert_mtb[i] = 1 if rng.random() < 0.40 else 0
        else:
            xpert_mtb[i] = 1 if rng.random() < (1 - XPERT_SPECIFICITY) else 0

        if xpert_mtb[i]:
            if rifampicin_resistant[i]:
                xpert_rif_resistant[i] = 'detected' if rng.random() < XPERT_RIF_SENSITIVITY else 'not_detected'
            else:
                xpert_rif_resistant[i] = 'not_detected' if rng.random() < 0.98 else 'detected'

    # ── Step 9: Chest X-ray ──
    cxr_result = np.array(['normal'] * n, dtype=object)
    for i in range(n):
        if true_tb[i] and tb_type[i] == 'pulmonary':
            if rng.random() < CXR_SENSITIVITY:
                cxr_result[i] = rng.choice(['infiltrate', 'cavity', 'miliary',
                                             'pleural_effusion'],
                                            p=[0.50, 0.25, 0.10, 0.15])
        elif true_tb[i] and tb_type[i] == 'extrapulmonary':
            if rng.random() < 0.30:
                cxr_result[i] = rng.choice(['infiltrate', 'pleural_effusion'],
                                            p=[0.40, 0.60])
        else:
            if rng.random() < (1 - CXR_SPECIFICITY):
                cxr_result[i] = rng.choice(['old_tb_scar', 'other_pathology',
                                             'infiltrate'],
                                            p=[0.40, 0.40, 0.20])

    cxr_abnormal = (cxr_result != 'normal').astype(int)

    # ── Step 10: TB classification ──
    tb_classification = np.array(['not_tb'] * n, dtype=object)
    for i in range(n):
        if true_tb[i]:
            if tb_type[i] == 'pulmonary' and smear_result[i]:
                tb_classification[i] = 'smear_positive_ptb'
            elif tb_type[i] == 'pulmonary':
                tb_classification[i] = 'smear_negative_ptb'
            else:
                tb_classification[i] = 'extrapulmonary_tb'

    # ── Step 11: Treatment outcome (simplified) ──
    treatment_outcome = np.array(['not_applicable'] * n, dtype=object)
    for i in range(n):
        if true_tb[i]:
            if rifampicin_resistant[i]:
                outcomes = ['cured', 'completed', 'failed', 'died', 'lost_to_followup']
                probs = [0.45, 0.15, 0.15, 0.15, 0.10]
            elif hiv_status[i]:
                outcomes = ['cured', 'completed', 'failed', 'died', 'lost_to_followup']
                probs = [0.50, 0.20, 0.05, 0.15, 0.10]
            else:
                outcomes = ['cured', 'completed', 'failed', 'died', 'lost_to_followup']
                probs = [0.65, 0.20, 0.03, 0.05, 0.07]
            treatment_outcome[i] = rng.choice(outcomes, p=probs)

    # ── Assemble DataFrame ──
    df = pd.DataFrame({
        'id': np.arange(1, n + 1),
        'age_years': age,
        'sex': sex,
        'bmi': bmi,
        'hiv_status': hiv_status,
        'cough_2weeks': symptoms['cough_2weeks'],
        'cough_duration_weeks': cough_duration_weeks,
        'fever': symptoms['fever'],
        'night_sweats': symptoms['night_sweats'],
        'weight_loss': symptoms['weight_loss'],
        'hemoptysis': symptoms['hemoptysis'],
        'chest_pain': symptoms['chest_pain'],
        'fatigue': symptoms['fatigue'],
        'loss_of_appetite': symptoms['loss_of_appetite'],
        'who_symptom_screen_count': who_symptom_count,
        'smear_result': smear_result,
        'xpert_mtb_detected': xpert_mtb,
        'xpert_rif_resistance': xpert_rif_resistant,
        'cxr_result': cxr_result,
        'cxr_abnormal': cxr_abnormal,
        'true_tb_status': true_tb,
        'tb_type': tb_type,
        'tb_classification': tb_classification,
        'rifampicin_resistant': rifampicin_resistant,
        'treatment_outcome': treatment_outcome,
    })

    # ── Print summary ──
    tb_pos = true_tb.sum()
    print(f"\n{'='*60}")
    print(f"TB Screening — {scenario} (n={n}, seed={seed})")
    print(f"{'='*60}")
    print(f"\nTB prevalence: {tb_pos/n*100:.1f}%")
    print(f"HIV+: {hiv_status.mean()*100:.1f}%")
    print(f"TB-HIV co-infection: {(true_tb & hiv_status).sum()}/{tb_pos} "
          f"({(true_tb & hiv_status).sum()/max(tb_pos,1)*100:.0f}%) of TB cases")
    print(f"Pulmonary TB: {(tb_type=='pulmonary').sum()}")
    print(f"Extrapulmonary TB: {(tb_type=='extrapulmonary').sum()}")
    print(f"Smear+: {smear_result.sum()} ({smear_result.mean()*100:.1f}%)")
    print(f"Xpert MTB+: {xpert_mtb.sum()} ({xpert_mtb.mean()*100:.1f}%)")
    print(f"CXR abnormal: {cxr_abnormal.sum()} ({cxr_abnormal.mean()*100:.1f}%)")
    print(f"MDR-TB: {rifampicin_resistant.sum()} ({rifampicin_resistant[true_tb==1].mean()*100:.1f}% of TB)")
    print(f"Mean BMI (TB): {bmi[true_tb==1].mean():.1f}, (non-TB): {bmi[true_tb==0].mean():.1f}")

    return df


# ============================================================
# SECTION 4: CLI Entry Point
# ============================================================

if __name__ == '__main__':
    parser = argparse.ArgumentParser(
        description='Generate synthetic TB screening dataset')
    parser.add_argument('--scenario', type=str, default='moderate_tb_burden',
                        choices=list(SCENARIOS.keys()))
    parser.add_argument('--n', type=int, default=10000)
    parser.add_argument('--seed', type=int, default=42)
    parser.add_argument('--output', type=str, default=None)
    parser.add_argument('--all-scenarios', action='store_true')
    args = parser.parse_args()

    os.makedirs('data', exist_ok=True)

    if args.all_scenarios:
        for sc_name in SCENARIOS:
            df = generate_tb_dataset(n=args.n, seed=args.seed, scenario=sc_name)
            out = os.path.join('data', f'tb_{sc_name}.csv')
            df.to_csv(out, index=False)
            print(f"  → Saved to {out}\n")
    else:
        df = generate_tb_dataset(n=args.n, seed=args.seed, scenario=args.scenario)
        out = args.output or os.path.join('data', f'tb_{args.scenario}.csv')
        df.to_csv(out, index=False)
        print(f"  → Saved to {out}")