generate_dataset.py

#!/usr/bin/env python3
"""
HIV Viral Load & CD4 Testing Dataset
========================================
Each record = ONE HIV viral load or CD4 test request/result.

Literature-Grounded Parameterization:
[1] WHO (2023). HIV treatment guidelines.
    - VL preferred monitoring; target <1000 copies/mL
    - CD4 for staging and OI prophylaxis
    - VL coverage SSA: ~70% of ART patients

[2] UNAIDS (2023). Global HIV statistics.
    - SSA: 25.6M PLHIV, 21M on ART
    - VL suppression: 76% of those on ART

[3] Lecher SL et al. (2021). Scale-up HIV VL monitoring.
    MMWR. DOI: 10.15585/mmwr.mm7034a2
    - VL coverage 17% to 71% (2015-2019)
    - Centralized testing: TAT weeks

[4] Jani IV et al. (2016). POC CD4 and VL testing.
    Lancet HIV. DOI: 10.1016/S2352-3018(15)00236-X
    - POC reduces loss to follow-up
    - Conventional VL TAT: 2-6 weeks
    - POC VL TAT: 1-3 hours
    - DBS for sample transport
"""
import numpy as np, pandas as pd, argparse, os

SCENARIOS = {
    'vl_accessible': {
        'exemplar': 'South Africa/Botswana/Kenya (urban)',
        'vl_coverage': 0.75,
        'cd4_available': 0.85,
        'poc_vl_available': 0.20,
        'dbs_used': 0.15,
        'conventional_vl': 0.65,
        'vl_suppression_rate': 0.80,
        'cd4_median_on_art': 450,
        'tat_vl_days': 7,
        'tat_cd4_days': 1,
        'eac_for_unsuppressed': 0.60,
        'result_returned_patient': 0.70,
        'reagent_stockout': 0.10,
    },
    'vl_limited': {
        'exemplar': 'Kenya (district)/Ghana/Tanzania/Ethiopia',
        'vl_coverage': 0.35,
        'cd4_available': 0.50,
        'poc_vl_available': 0.05,
        'dbs_used': 0.30,
        'conventional_vl': 0.25,
        'vl_suppression_rate': 0.72,
        'cd4_median_on_art': 380,
        'tat_vl_days': 21,
        'tat_cd4_days': 3,
        'eac_for_unsuppressed': 0.30,
        'result_returned_patient': 0.40,
        'reagent_stockout': 0.30,
    },
    'no_vl_access': {
        'exemplar': 'DRC/CAR/Sierra Leone/Niger (rural)',
        'vl_coverage': 0.08,
        'cd4_available': 0.15,
        'poc_vl_available': 0.01,
        'dbs_used': 0.05,
        'conventional_vl': 0.05,
        'vl_suppression_rate': 0.60,
        'cd4_median_on_art': 300,
        'tat_vl_days': 45,
        'tat_cd4_days': 14,
        'eac_for_unsuppressed': 0.10,
        'result_returned_patient': 0.15,
        'reagent_stockout': 0.55,
    },
}

def generate_dataset(n=10000, seed=42, scenario='vl_limited'):
    rng = np.random.default_rng(seed)
    sc = SCENARIOS[scenario]
    records = []
    for idx in range(n):
        rec = {'id': idx + 1}
        rec['age'] = int(np.clip(rng.normal(38, 12), 1, 80))
        rec['sex'] = rng.choice(['male','female'], p=[0.38, 0.62])
        rec['pregnant'] = 1 if (rec['sex']=='female' and 15<=rec['age']<=45 and rng.random()<0.08) else 0
        rec['facility_level'] = rng.choice(['tertiary','district','health_centre','clinic'],
            p=[0.10, 0.25, 0.35, 0.30])
        rec['on_art'] = 1 if rng.random() < 0.85 else 0
        rec['art_duration_months'] = int(rng.exponential(36)) if rec['on_art'] else 0
        rec['art_regimen'] = rng.choice(['TLD','TLE','AZT_based','PI_based','other'],
            p=[0.40, 0.25, 0.15, 0.10, 0.10]) if rec['on_art'] else 'none'

        # Test type ordered
        rec['vl_ordered'] = 1 if rng.random() < sc['vl_coverage'] else 0
        rec['cd4_ordered'] = 1 if rng.random() < sc['cd4_available'] else 0

        # VL testing
        tier = 1.3 if rec['facility_level'] in ['tertiary','district'] else 0.7
        if rec['vl_ordered']:
            rec['vl_platform'] = rng.choice(['conventional_central','poc_vl','dbs_referred'],
                p=[sc['conventional_vl']*tier/(sc['conventional_vl']*tier+sc['poc_vl_available']+sc['dbs_used']+0.01),
                   sc['poc_vl_available']/(sc['conventional_vl']*tier+sc['poc_vl_available']+sc['dbs_used']+0.01),
                   (sc['dbs_used']+0.01)/(sc['conventional_vl']*tier+sc['poc_vl_available']+sc['dbs_used']+0.01)])
            if rec['on_art']:
                suppressed = rng.random() < sc['vl_suppression_rate']
                if suppressed:
                    rec['vl_result'] = int(np.clip(rng.exponential(50), 0, 999))
                else:
                    rec['vl_result'] = int(np.clip(rng.lognormal(np.log(5000), 1.5), 1000, 500000))
            else:
                rec['vl_result'] = int(np.clip(rng.lognormal(np.log(50000), 1.2), 1000, 1000000))
            rec['vl_suppressed'] = 1 if rec['vl_result'] < 1000 else 0
            rec['vl_tat_days'] = round(max(0.1, rng.lognormal(
                np.log(0.1 if rec['vl_platform']=='poc_vl' else sc['tat_vl_days']), 0.5)), 1)
        else:
            rec['vl_platform'] = 'not_done'
            rec['vl_result'] = np.nan
            rec['vl_suppressed'] = np.nan
            rec['vl_tat_days'] = np.nan

        # CD4 testing
        if rec['cd4_ordered']:
            rec['cd4_platform'] = rng.choice(['flow_cytometry','poc_cd4','manual'],
                p=[0.40, 0.30, 0.30] if sc['cd4_available']>0.5 else [0.15, 0.25, 0.60])
            if rec['on_art']:
                rec['cd4_result'] = int(np.clip(rng.normal(sc['cd4_median_on_art'], 180), 10, 1500))
            else:
                rec['cd4_result'] = int(np.clip(rng.lognormal(np.log(200), 0.7), 5, 800))
            rec['cd4_below_200'] = 1 if rec['cd4_result'] < 200 else 0
            rec['cd4_tat_days'] = round(max(0.1, rng.lognormal(np.log(sc['tat_cd4_days']), 0.5)), 1)
        else:
            rec['cd4_platform'] = 'not_done'
            rec['cd4_result'] = np.nan
            rec['cd4_below_200'] = np.nan
            rec['cd4_tat_days'] = np.nan

        # Clinical actions
        rec['result_returned'] = 1 if rng.random() < sc['result_returned_patient'] else 0
        rec['eac_provided'] = 1 if (rec['vl_ordered'] and rec.get('vl_suppressed')==0 and rng.random() < sc['eac_for_unsuppressed']) else 0
        rec['regimen_switch'] = 1 if (rec['vl_ordered'] and rec.get('vl_suppressed')==0 and rng.random() < 0.15) else 0
        rec['oi_prophylaxis'] = 1 if (rec['cd4_ordered'] and rec.get('cd4_below_200')==1 and rng.random() < 0.65) else 0

        # Quality
        rec['reagent_stockout'] = 1 if rng.random() < sc['reagent_stockout'] else 0
        rec['sample_rejected'] = 1 if rng.random() < 0.04 else 0
        rec['test_not_done_stockout'] = 1 if (rec['reagent_stockout'] and rng.random() < 0.40) else 0
        rec['specimen_transport_issue'] = 1 if (rec.get('vl_platform') in ['dbs_referred','conventional_central'] and rng.random() < 0.12) else 0

        rec['year'] = rng.choice([2019,2020,2021,2022,2023], p=[0.12,0.18,0.20,0.25,0.25])
        records.append(rec)

    df = pd.DataFrame(records)
    print(f"\n{'='*60}\nHIV Viral Load & CD4 — {scenario} ({sc['exemplar']})")
    print(f"  VL ordered: {df['vl_ordered'].mean()*100:.1f}% | CD4 ordered: {df['cd4_ordered'].mean()*100:.1f}%")
    vl_done = df[df['vl_ordered']==1]
    if len(vl_done)>0: print(f"  VL suppressed: {vl_done['vl_suppressed'].mean()*100:.1f}% | VL TAT median: {vl_done['vl_tat_days'].median():.1f}d")
    print(f"  Result returned: {df['result_returned'].mean()*100:.1f}% | Stockout: {df['reagent_stockout'].mean()*100:.1f}%")
    return df

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--all-scenarios', action='store_true')
    parser.add_argument('--n', type=int, default=10000)
    parser.add_argument('--seed', type=int, default=42)
    args = parser.parse_args()
    os.makedirs('data', exist_ok=True)
    if args.all_scenarios:
        for sc in SCENARIOS:
            df = generate_dataset(n=args.n, seed=args.seed, scenario=sc)
            df.to_csv(os.path.join('data', f'hiv_vl_cd4_{sc}.csv'), index=False)
            print(f"  -> Saved\n")
    else:
        df = generate_dataset(n=args.n, seed=args.seed)
        df.to_csv(os.path.join('data', 'hiv_vl_cd4_vl_limited.csv'), index=False)