generate_dataset.py

#!/usr/bin/env python3
"""
Literature-Informed Synthetic Community Health Worker iCCM Triage Dataset
=========================================================================

Generates realistic synthetic datasets of sick child assessments by community
health workers (CHWs) using integrated Community Case Management (iCCM)
protocols. Children aged 2-59 months presenting with acute illness at
community level in LMIC settings.

iCCM covers three core conditions:
  - Malaria (fever + RDT)
  - Pneumonia (cough + fast breathing)
  - Diarrhoea (loose stools + dehydration assessment)
Plus danger sign assessment for referral decisions.

DAG (Sampling Order):
  1. age_months, sex (roots)
  2. presenting_complaint (from prevalence)
  3. true_diagnosis (conditional on complaint + scenario)
  4. symptoms & signs (conditional on diagnosis)
  5. rdt_result (conditional on malaria status + test performance)
  6. respiratory_rate, fast_breathing (conditional on pneumonia)
  7. muac_cm (age-conditional, used for malnutrition screening)
  8. danger_signs (conditional on severity)
  9. chw_classification (iCCM algorithm output)
  10. chw_action (treat/refer)

References:
-----------
[1] WHO/UNICEF (2012). Caring for the sick child in the community. Geneva.
[2] WHO/UNICEF (2014). Integrated Community Case Management (iCCM):
    Evidence review. Geneva.
[3] Marsh DR, et al. (2012). Introduction to a special supplement on iCCM.
    Am J Trop Med Hyg, 87(5 Suppl):1-5.
[4] Druetz T, et al. (2015). Impact of iCCM on child mortality: a systematic
    review. Paediatrics & International Child Health, 35(1):18-29.
[5] WHO (2023). World Malaria Report 2023. Geneva.
[6] UNICEF (2019). Diarrhoea treatment guidelines. New York.
[7] WHO (2014). Revised WHO classification and treatment of pneumonia in
    children at health facilities. Geneva.
[8] DHS/MICS Program. Treatment-seeking and CHW utilization data.
"""

import numpy as np
import pandas as pd
import argparse
import os

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

SCENARIOS = {
    'low_burden': {
        'description': 'Lower burden LMIC community (e.g., urban peri-urban)',
        'malaria_pct': 0.20,        # Of presenting children
        'pneumonia_pct': 0.15,
        'diarrhoea_pct': 0.25,
        'mixed_pct': 0.08,          # Multiple conditions
        'other_febrile_pct': 0.32,  # Other/viral
        'danger_sign_rate': 0.04,   # Need referral
        'malnutrition_sam_rate': 0.02,
        'malnutrition_mam_rate': 0.05,
    },
    'moderate_burden': {
        'description': 'Average LMIC community (e.g., rural Kenya, Senegal)',
        'malaria_pct': 0.35,
        'pneumonia_pct': 0.18,
        'diarrhoea_pct': 0.22,
        'mixed_pct': 0.10,
        'other_febrile_pct': 0.15,
        'danger_sign_rate': 0.07,
        'malnutrition_sam_rate': 0.04,
        'malnutrition_mam_rate': 0.08,
    },
    'high_burden': {
        'description': 'High burden / conflict (e.g., Sahel, DRC)',
        'malaria_pct': 0.45,
        'pneumonia_pct': 0.20,
        'diarrhoea_pct': 0.18,
        'mixed_pct': 0.12,
        'other_febrile_pct': 0.05,
        'danger_sign_rate': 0.12,
        'malnutrition_sam_rate': 0.08,
        'malnutrition_mam_rate': 0.12,
    },
}

# --- RDT Performance (same as malaria dataset) ---
RDT_SENSITIVITY = 0.93
RDT_SPECIFICITY = 0.95

# --- Respiratory rate thresholds (WHO/UNICEF iCCM) ---
# Fast breathing: ≥50 breaths/min (2-11 months), ≥40 (12-59 months)
RR_PNEUMONIA = {'mean': 56, 'sd': 8}   # Pneumonia cases
RR_NORMAL = {'mean': 32, 'sd': 6}       # Non-pneumonia

# --- MUAC reference ---
MUAC_NORMAL = {'mean': 14.8, 'sd': 1.2}
MUAC_MAM = {'mean': 12.0, 'sd': 0.3}   # 11.5-12.4 cm
MUAC_SAM = {'mean': 10.8, 'sd': 0.5}   # <11.5 cm


# ============================================================
# SECTION 2: Main Generator
# ============================================================

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

    # ── Step 1: Demographics ──
    sex = rng.choice(['M', 'F'], size=n, p=[0.512, 0.488])
    age_months = rng.integers(2, 60, size=n)

    # ── Step 2: True diagnosis ──
    diag_probs = [sc['malaria_pct'], sc['pneumonia_pct'], sc['diarrhoea_pct'],
                  sc['mixed_pct'], sc['other_febrile_pct']]
    diagnoses = ['malaria', 'pneumonia', 'diarrhoea', 'mixed', 'other_febrile']
    true_diagnosis = rng.choice(diagnoses, size=n, p=diag_probs)

    # Mixed = malaria + pneumonia or malaria + diarrhoea
    mixed_type = np.array(['none'] * n, dtype=object)
    for i in range(n):
        if true_diagnosis[i] == 'mixed':
            mixed_type[i] = rng.choice(['malaria_pneumonia', 'malaria_diarrhoea'],
                                        p=[0.55, 0.45])

    # ── Step 3: Fever ──
    fever = np.zeros(n, dtype=int)
    temperature = np.zeros(n)
    for i in range(n):
        if true_diagnosis[i] in ('malaria', 'other_febrile'):
            fever[i] = 1 if rng.random() < 0.88 else 0
        elif true_diagnosis[i] == 'pneumonia':
            fever[i] = 1 if rng.random() < 0.65 else 0
        elif true_diagnosis[i] == 'diarrhoea':
            fever[i] = 1 if rng.random() < 0.35 else 0
        elif true_diagnosis[i] == 'mixed':
            fever[i] = 1 if rng.random() < 0.90 else 0

        if fever[i]:
            temperature[i] = rng.normal(38.6, 0.7)
        else:
            temperature[i] = rng.normal(37.0, 0.3)
    temperature = np.clip(np.round(temperature, 1), 35.5, 41.5)
    fever_duration_days = np.where(fever, np.clip(rng.poisson(3, n), 1, 14), 0)

    # ── Step 4: Cough & respiratory signs ──
    cough = np.zeros(n, dtype=int)
    respiratory_rate = np.zeros(n, dtype=int)
    for i in range(n):
        has_resp = true_diagnosis[i] == 'pneumonia' or \
                   (true_diagnosis[i] == 'mixed' and 'pneumonia' in mixed_type[i])
        if has_resp:
            cough[i] = 1 if rng.random() < 0.90 else 0
            respiratory_rate[i] = max(15, int(rng.normal(RR_PNEUMONIA['mean'], RR_PNEUMONIA['sd'])))
        else:
            cough[i] = 1 if rng.random() < 0.25 else 0
            respiratory_rate[i] = max(15, int(rng.normal(RR_NORMAL['mean'], RR_NORMAL['sd'])))

    # Fast breathing classification (WHO iCCM)
    fast_breathing = np.zeros(n, dtype=int)
    for i in range(n):
        threshold = 50 if age_months[i] < 12 else 40
        fast_breathing[i] = 1 if respiratory_rate[i] >= threshold else 0

    # ── Step 5: Diarrhoea signs ──
    diarrhoea = np.zeros(n, dtype=int)
    diarrhoea_days = np.zeros(n, dtype=int)
    blood_in_stool = np.zeros(n, dtype=int)
    dehydration_status = np.array(['none'] * n, dtype=object)

    for i in range(n):
        has_diarr = true_diagnosis[i] == 'diarrhoea' or \
                    (true_diagnosis[i] == 'mixed' and 'diarrhoea' in mixed_type[i])
        if has_diarr:
            diarrhoea[i] = 1
            diarrhoea_days[i] = max(1, rng.poisson(4))
            blood_in_stool[i] = 1 if rng.random() < 0.08 else 0
            r = rng.random()
            if r < 0.10:
                dehydration_status[i] = 'severe'
            elif r < 0.35:
                dehydration_status[i] = 'some'
            else:
                dehydration_status[i] = 'none'
        else:
            diarrhoea[i] = 1 if rng.random() < 0.08 else 0
            if diarrhoea[i]:
                diarrhoea_days[i] = max(1, rng.poisson(2))

    diarrhoea_days = np.clip(diarrhoea_days, 0, 21)

    # ── Step 6: RDT result ──
    has_malaria = np.array([
        true_diagnosis[i] == 'malaria' or
        (true_diagnosis[i] == 'mixed' and 'malaria' in mixed_type[i])
        for i in range(n)
    ])

    rdt_result = np.zeros(n, dtype=int)
    for i in range(n):
        if has_malaria[i]:
            rdt_result[i] = 1 if rng.random() < RDT_SENSITIVITY else 0
        else:
            rdt_result[i] = 1 if rng.random() < (1 - RDT_SPECIFICITY) else 0

    # ── Step 7: MUAC ──
    muac = np.zeros(n)
    nutrition_status = np.array(['normal'] * n, dtype=object)
    for i in range(n):
        r = rng.random()
        if r < sc['malnutrition_sam_rate']:
            muac[i] = rng.normal(MUAC_SAM['mean'], MUAC_SAM['sd'])
            nutrition_status[i] = 'SAM'
        elif r < sc['malnutrition_sam_rate'] + sc['malnutrition_mam_rate']:
            muac[i] = rng.normal(MUAC_MAM['mean'], MUAC_MAM['sd'])
            nutrition_status[i] = 'MAM'
        else:
            muac[i] = rng.normal(MUAC_NORMAL['mean'], MUAC_NORMAL['sd'])
    muac = np.clip(np.round(muac, 1), 7.0, 20.0)
    # Reconcile MUAC with nutrition status
    nutrition_status = np.where(muac < 11.5, 'SAM',
                        np.where(muac < 12.5, 'MAM', 'normal'))

    # ── Step 8: Danger signs (WHO/UNICEF iCCM) ──
    unable_to_drink = np.zeros(n, dtype=int)
    vomiting_everything = np.zeros(n, dtype=int)
    convulsions = np.zeros(n, dtype=int)
    lethargic_unconscious = np.zeros(n, dtype=int)
    chest_indrawing = np.zeros(n, dtype=int)

    for i in range(n):
        is_severe = rng.random() < sc['danger_sign_rate']
        # More likely with younger children, malnourished
        if age_months[i] < 12:
            is_severe = is_severe or rng.random() < sc['danger_sign_rate'] * 0.5
        if nutrition_status[i] == 'SAM':
            is_severe = is_severe or rng.random() < 0.15

        if is_severe:
            unable_to_drink[i] = 1 if rng.random() < 0.45 else 0
            vomiting_everything[i] = 1 if rng.random() < 0.40 else 0
            convulsions[i] = 1 if rng.random() < 0.20 else 0
            lethargic_unconscious[i] = 1 if rng.random() < 0.25 else 0
            chest_indrawing[i] = 1 if rng.random() < 0.35 else 0
        else:
            # Occasional isolated signs in non-severe
            vomiting_everything[i] = 1 if rng.random() < 0.02 else 0
            chest_indrawing[i] = 1 if rng.random() < 0.01 else 0

    any_danger_sign = ((unable_to_drink + vomiting_everything + convulsions +
                        lethargic_unconscious + chest_indrawing) > 0).astype(int)

    # ── Step 9: CHW classification (iCCM algorithm) ──
    chw_classification = np.array(['other'] * n, dtype=object)
    for i in range(n):
        classifications = []
        if rdt_result[i] == 1:
            classifications.append('malaria')
        if cough[i] and fast_breathing[i]:
            classifications.append('pneumonia')
        if diarrhoea[i]:
            classifications.append('diarrhoea')
        if nutrition_status[i] == 'SAM':
            classifications.append('severe_malnutrition')

        if len(classifications) == 0:
            chw_classification[i] = 'other_febrile' if fever[i] else 'well_child'
        elif len(classifications) == 1:
            chw_classification[i] = classifications[0]
        else:
            chw_classification[i] = '+'.join(sorted(classifications))

    # ── Step 10: CHW action ──
    chw_action = np.array(['treat_at_community'] * n, dtype=object)
    for i in range(n):
        if any_danger_sign[i]:
            chw_action[i] = 'refer_urgently'
        elif nutrition_status[i] == 'SAM':
            chw_action[i] = 'refer_urgently'
        elif dehydration_status[i] == 'severe':
            chw_action[i] = 'refer_urgently'
        elif age_months[i] < 2:
            chw_action[i] = 'refer_urgently'  # Very young infant
        elif rdt_result[i] and fast_breathing[i]:
            chw_action[i] = 'treat_and_refer'
        elif blood_in_stool[i]:
            chw_action[i] = 'refer'

    # Treatment given
    act_given = ((rdt_result == 1) & (chw_action != 'refer_urgently')).astype(int)
    amoxicillin_given = ((cough == 1) & (fast_breathing == 1) &
                          (chw_action != 'refer_urgently')).astype(int)
    ors_given = ((diarrhoea == 1) & (chw_action != 'refer_urgently')).astype(int)
    zinc_given = ors_given.copy()

    # ── Assemble DataFrame ──
    df = pd.DataFrame({
        'id': np.arange(1, n + 1),
        'age_months': age_months,
        'sex': sex,
        'fever': fever,
        'temperature_c': temperature,
        'fever_duration_days': fever_duration_days,
        'cough': cough,
        'respiratory_rate_bpm': respiratory_rate,
        'fast_breathing': fast_breathing,
        'chest_indrawing': chest_indrawing,
        'diarrhoea': diarrhoea,
        'diarrhoea_duration_days': diarrhoea_days,
        'blood_in_stool': blood_in_stool,
        'dehydration_status': dehydration_status,
        'rdt_result': rdt_result,
        'muac_cm': muac,
        'nutrition_status': nutrition_status,
        'unable_to_drink': unable_to_drink,
        'vomiting_everything': vomiting_everything,
        'convulsions': convulsions,
        'lethargic_unconscious': lethargic_unconscious,
        'any_danger_sign': any_danger_sign,
        'true_diagnosis': true_diagnosis,
        'chw_classification': chw_classification,
        'chw_action': chw_action,
        'act_given': act_given,
        'amoxicillin_given': amoxicillin_given,
        'ors_given': ors_given,
        'zinc_given': zinc_given,
    })

    # ── Print summary ──
    print(f"\n{'='*60}")
    print(f"iCCM CHW Triage — {scenario} (n={n}, seed={seed})")
    print(f"{'='*60}")
    print(f"\nTrue diagnosis:")
    for d in diagnoses:
        print(f"  {d:20s}: {(true_diagnosis==d).mean()*100:.1f}%")
    print(f"\nRDT+: {rdt_result.mean()*100:.1f}%")
    print(f"Fast breathing: {fast_breathing.mean()*100:.1f}%")
    print(f"Diarrhoea: {(diarrhoea==1).mean()*100:.1f}%")
    print(f"Any danger sign: {any_danger_sign.mean()*100:.1f}%")
    print(f"SAM: {(nutrition_status=='SAM').mean()*100:.1f}%")
    print(f"Referral rate: {np.mean(['refer' in str(a) for a in chw_action])*100:.1f}%")
    print(f"ACT given: {act_given.mean()*100:.1f}%")
    print(f"Amoxicillin given: {amoxicillin_given.mean()*100:.1f}%")
    print(f"ORS given: {ors_given.mean()*100:.1f}%")

    return df


# ============================================================
# SECTION 3: CLI Entry Point
# ============================================================

if __name__ == '__main__':
    parser = argparse.ArgumentParser(
        description='Generate synthetic iCCM CHW triage dataset')
    parser.add_argument('--scenario', type=str, default='moderate_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_iccm_dataset(n=args.n, seed=args.seed, scenario=sc_name)
            out = os.path.join('data', f'iccm_{sc_name}.csv')
            df.to_csv(out, index=False)
            print(f"  → Saved to {out}\n")
    else:
        df = generate_iccm_dataset(n=args.n, seed=args.seed, scenario=args.scenario)
        out = args.output or os.path.join('data', f'iccm_{args.scenario}.csv')
        df.to_csv(out, index=False)
        print(f"  → Saved to {out}")