generate_dataset.py

#!/usr/bin/env python3
"""
Literature-Informed Synthetic Snakebite Envenomation Dataset Generator
======================================================================

Generates realistic synthetic records of snakebite envenomation cases
presenting to health facilities in sub-Saharan Africa, including snake
species, envenomation syndrome, severity grading, clinical features,
antivenom treatment, complications, and outcomes.

Target population: Snakebite victims aged 1-80 years presenting to
health facilities across SSA.

DAG (Sampling Order):
  1. age, sex, occupation, location (roots)
  2. snake_species (conditional on region/habitat)
  3. envenomation_syndrome: cytotoxic / hemotoxic / neurotoxic / dry bite
  4. severity: mild / moderate / severe
  5. clinical_features: swelling, coagulopathy, neurotoxicity, necrosis
  6. treatment: antivenom, wound care, fasciotomy
  7. complications: compartment syndrome, AKI, necrosis, amputation
  8. outcome: survived / died / disability

References:
-----------
[1]  Chippaux JP (Toxicon 2011). SSA: ~315,000 envenomations/yr,
     ~32,000 deaths, ~7,000 amputations. Case fatality 6-10%.
[2]  WHO (2019). Snakebite envenoming strategy. NTD category A since
     2017. Target: halve deaths/disability by 2030.
[3]  Habib AG, et al. (PLoS NTD 2015). Echis ocellatus envenoming in
     Nigeria: coagulopathy in 70-80%, local swelling 90%+.
     EchiTab antivenom efficacy.
[4]  Warrell DA (Lancet 2010). Snake bite. Clinical features by species:
     Bitis arietans: massive local swelling, necrosis, coagulopathy.
     Naja: necrosis, neurotoxicity. Dendroaspis: rapid neurotoxicity.
     Echis: coagulopathy, haemorrhage, renal failure.
[5]  Harrison RA, et al. (PLoS NTD 2009). Snakebite: disease of poverty.
     Agricultural workers, children at highest risk.
[6]  Kasturiratne A, et al. (PLoS Med 2008). Global burden: 1.2-5.5M
     bites/yr, 421,000-1.8M envenomations, 20,000-94,000 deaths.
[7]  Gutierrez JM, et al. (Nat Rev Dis Primers 2017). Snakebite
     envenoming comprehensive review. Antivenom sole specific treatment.
[8]  Schioldann E, et al. (PLoS NTD 2018). Antivenom availability in
     SSA: <2% of need met. Mean time to treatment 6-24 hours.
"""

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

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

SCENARIOS = {
    'referral_hospital': {
        'description': 'Tertiary/referral hospital with antivenom stock, '
                       'surgical capacity (e.g., Kaltungo Nigeria, Kilifi Kenya)',
        # Snake species distribution [1][3][4]
        'species_probs': {
            'echis_ocellatus': 0.30,     # Carpet viper
            'bitis_arietans': 0.25,      # Puff adder
            'naja_nigricollis': 0.15,    # Black-necked spitting cobra
            'naja_haje': 0.08,           # Egyptian cobra
            'dendroaspis_polylepis': 0.05, # Black mamba
            'dendroaspis_viridis': 0.03, # Green mamba
            'causus_rhombeatus': 0.04,   # Night adder
            'atractaspis_spp': 0.03,     # Burrowing asp
            'unknown_unidentified': 0.07,
        },
        'dry_bite_rate': 0.20,           # No envenomation
        'antivenom_available': 0.70,
        'antivenom_given_if_needed': 0.80,
        'surgical_capacity': 0.90,
        'time_to_facility_hrs_mean': 4,
        'case_fatality': 0.04,
    },
    'district_hospital': {
        'description': 'District hospital, intermittent antivenom supply, '
                       'basic surgical (e.g., district hospitals Ghana, Tanzania)',
        'species_probs': {
            'echis_ocellatus': 0.28,
            'bitis_arietans': 0.28,
            'naja_nigricollis': 0.14,
            'naja_haje': 0.06,
            'dendroaspis_polylepis': 0.04,
            'dendroaspis_viridis': 0.02,
            'causus_rhombeatus': 0.06,
            'atractaspis_spp': 0.03,
            'unknown_unidentified': 0.09,
        },
        'dry_bite_rate': 0.20,
        'antivenom_available': 0.35,
        'antivenom_given_if_needed': 0.60,
        'surgical_capacity': 0.50,
        'time_to_facility_hrs_mean': 8,
        'case_fatality': 0.08,
    },
    'rural_health_centre': {
        'description': 'Rural PHC/health centre, no antivenom, referral '
                       'needed for severe cases (e.g., rural DRC, Niger, Chad)',
        'species_probs': {
            'echis_ocellatus': 0.30,
            'bitis_arietans': 0.30,
            'naja_nigricollis': 0.12,
            'naja_haje': 0.05,
            'dendroaspis_polylepis': 0.05,
            'dendroaspis_viridis': 0.02,
            'causus_rhombeatus': 0.05,
            'atractaspis_spp': 0.02,
            'unknown_unidentified': 0.09,
        },
        'dry_bite_rate': 0.20,
        'antivenom_available': 0.05,
        'antivenom_given_if_needed': 0.20,
        'surgical_capacity': 0.10,
        'time_to_facility_hrs_mean': 16,
        'case_fatality': 0.14,
    },
}

# Envenomation syndrome by species [4]
SYNDROME_BY_SPECIES = {
    'echis_ocellatus': {'hemotoxic': 0.70, 'cytotoxic': 0.25, 'mixed': 0.05},
    'bitis_arietans': {'cytotoxic': 0.60, 'hemotoxic': 0.30, 'mixed': 0.10},
    'naja_nigricollis': {'cytotoxic': 0.55, 'neurotoxic': 0.35, 'mixed': 0.10},
    'naja_haje': {'neurotoxic': 0.60, 'cytotoxic': 0.30, 'mixed': 0.10},
    'dendroaspis_polylepis': {'neurotoxic': 0.85, 'mixed': 0.15},
    'dendroaspis_viridis': {'neurotoxic': 0.80, 'mixed': 0.20},
    'causus_rhombeatus': {'cytotoxic': 0.90, 'hemotoxic': 0.10},
    'atractaspis_spp': {'cytotoxic': 0.70, 'cardiotoxic': 0.30},
    'unknown_unidentified': {'cytotoxic': 0.45, 'hemotoxic': 0.30,
                              'neurotoxic': 0.15, 'mixed': 0.10},
}


def sample_categorical(probs_dict, rng):
    labels = list(probs_dict.keys())
    probs = np.array(list(probs_dict.values()))
    probs = probs / probs.sum()
    return rng.choice(labels, p=probs)


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

    records = []

    for idx in range(n):
        rec = {'id': idx + 1}

        # ── Step 1: Demographics [5] ──
        r = rng.random()
        if r < 0.15:
            rec['age_years'] = rng.integers(1, 10)
        elif r < 0.30:
            rec['age_years'] = rng.integers(10, 18)
        elif r < 0.65:
            rec['age_years'] = rng.integers(18, 40)
        elif r < 0.85:
            rec['age_years'] = rng.integers(40, 60)
        else:
            rec['age_years'] = rng.integers(60, 80)

        rec['sex'] = rng.choice(['M', 'F'], p=[0.65, 0.35])  # Male predominance [5]

        if rec['age_years'] >= 18:
            rec['occupation'] = rng.choice(
                ['farmer', 'herder', 'hunter', 'trader', 'student', 'other'],
                p=[0.45, 0.15, 0.10, 0.10, 0.10, 0.10])
        else:
            rec['occupation'] = rng.choice(['student', 'child', 'herder'],
                                           p=[0.40, 0.45, 0.15])

        rec['bite_location'] = rng.choice(
            ['foot', 'ankle', 'lower_leg', 'hand', 'forearm', 'other'],
            p=[0.30, 0.20, 0.15, 0.20, 0.10, 0.05])

        rec['activity_at_bite'] = rng.choice(
            ['farming', 'walking', 'sleeping', 'herding', 'collecting_firewood', 'other'],
            p=[0.35, 0.25, 0.15, 0.10, 0.10, 0.05])

        # Time to facility [8]
        rec['time_to_facility_hours'] = max(0.5, round(
            rng.exponential(sc['time_to_facility_hrs_mean']), 1))
        rec['time_to_facility_hours'] = min(rec['time_to_facility_hours'], 72)

        # Traditional treatment before hospital
        rec['traditional_treatment_first'] = 1 if rng.random() < 0.40 else 0
        if rec['traditional_treatment_first']:
            rec['time_to_facility_hours'] *= 1.5

        # ── Step 2: Snake Species [1][4] ──
        rec['snake_species'] = sample_categorical(sc['species_probs'], rng)

        # ── Step 3: Envenomation & Syndrome ──
        rec['dry_bite'] = 1 if rng.random() < sc['dry_bite_rate'] else 0

        if rec['dry_bite']:
            rec['envenomation_syndrome'] = 'none'
            rec['severity'] = 'none'
        else:
            syndrome_probs = SYNDROME_BY_SPECIES.get(
                rec['snake_species'],
                {'cytotoxic': 0.50, 'hemotoxic': 0.30, 'neurotoxic': 0.20})
            rec['envenomation_syndrome'] = sample_categorical(syndrome_probs, rng)

            # Severity grading
            sev_roll = rng.random()
            if sev_roll < 0.30:
                rec['severity'] = 'mild'
            elif sev_roll < 0.70:
                rec['severity'] = 'moderate'
            else:
                rec['severity'] = 'severe'

            # Mamba bites more often severe
            if rec['snake_species'].startswith('dendroaspis'):
                if rng.random() < 0.4:
                    rec['severity'] = 'severe'

        # ── Step 4: Clinical Features [3][4] ──
        rec['local_swelling'] = 0
        rec['local_necrosis'] = 0
        rec['coagulopathy'] = 0
        rec['inr_elevated'] = 0
        rec['bleeding'] = 0
        rec['neurotoxicity'] = 0
        rec['ptosis'] = 0
        rec['respiratory_failure'] = 0
        rec['acute_kidney_injury'] = 0
        rec['compartment_syndrome'] = 0

        if not rec['dry_bite']:
            # Local effects
            if rec['envenomation_syndrome'] in ('cytotoxic', 'mixed'):
                rec['local_swelling'] = 1 if rng.random() < 0.92 else 0
                if rec['severity'] == 'severe':
                    rec['local_necrosis'] = 1 if rng.random() < 0.45 else 0
                    rec['compartment_syndrome'] = 1 if rng.random() < 0.15 else 0
                elif rec['severity'] == 'moderate':
                    rec['local_necrosis'] = 1 if rng.random() < 0.15 else 0
            elif rec['envenomation_syndrome'] == 'hemotoxic':
                rec['local_swelling'] = 1 if rng.random() < 0.85 else 0

            # Hemotoxic effects [3]
            if rec['envenomation_syndrome'] in ('hemotoxic', 'mixed'):
                rec['coagulopathy'] = 1 if rng.random() < 0.75 else 0
                if rec['coagulopathy']:
                    rec['inr_elevated'] = 1
                    rec['bleeding'] = 1 if rng.random() < 0.40 else 0
                rec['acute_kidney_injury'] = 1 if rng.random() < 0.20 else 0

            # Neurotoxic effects [4]
            if rec['envenomation_syndrome'] in ('neurotoxic', 'mixed'):
                rec['neurotoxicity'] = 1 if rng.random() < 0.80 else 0
                if rec['neurotoxicity']:
                    rec['ptosis'] = 1 if rng.random() < 0.70 else 0
                    if rec['severity'] == 'severe':
                        rec['respiratory_failure'] = 1 if rng.random() < 0.35 else 0

            # Delayed presentation worsens everything
            if rec['time_to_facility_hours'] > 12:
                if not rec['compartment_syndrome'] and rec['local_swelling']:
                    rec['compartment_syndrome'] = 1 if rng.random() < 0.10 else 0
                if not rec['acute_kidney_injury'] and rec['coagulopathy']:
                    rec['acute_kidney_injury'] = 1 if rng.random() < 0.15 else 0

        # ── Step 5: Treatment [7][8] ──
        rec['antivenom_given'] = 0
        rec['antivenom_vials'] = 0
        rec['antivenom_reaction'] = 0
        rec['wound_care'] = 0
        rec['fasciotomy'] = 0
        rec['blood_transfusion'] = 0
        rec['mechanical_ventilation'] = 0
        rec['dialysis'] = 0
        rec['amputation'] = 0
        rec['antibiotics_given'] = 0

        if not rec['dry_bite']:
            # Antivenom
            needs_antivenom = rec['severity'] in ('moderate', 'severe')
            if needs_antivenom:
                av_available = rng.random() < sc['antivenom_available']
                if av_available:
                    rec['antivenom_given'] = 1 if rng.random() < sc['antivenom_given_if_needed'] else 0

            if rec['antivenom_given']:
                if rec['severity'] == 'severe':
                    rec['antivenom_vials'] = rng.integers(3, 10)
                else:
                    rec['antivenom_vials'] = rng.integers(1, 5)
                rec['antivenom_reaction'] = 1 if rng.random() < 0.15 else 0

            # Wound care
            rec['wound_care'] = 1 if rng.random() < 0.85 else 0
            rec['antibiotics_given'] = 1 if rng.random() < 0.70 else 0

            # Surgery
            if rec['compartment_syndrome'] and rng.random() < sc['surgical_capacity']:
                rec['fasciotomy'] = 1

            if rec['local_necrosis'] and rec['severity'] == 'severe':
                if rec['time_to_facility_hours'] > 24 or not rec['antivenom_given']:
                    rec['amputation'] = 1 if rng.random() < 0.12 else 0

            # Supportive
            if rec['bleeding'] and rng.random() < 0.30:
                rec['blood_transfusion'] = 1
            if rec['respiratory_failure'] and rng.random() < sc['surgical_capacity']:
                rec['mechanical_ventilation'] = 1
            if rec['acute_kidney_injury'] and rec['severity'] == 'severe':
                if rng.random() < 0.15:
                    rec['dialysis'] = 1

        # ── Step 6: Outcome ──
        mort = 0.005  # Dry bite / mild
        if not rec['dry_bite']:
            if rec['severity'] == 'severe':
                mort = sc['case_fatality'] * 2.5
            elif rec['severity'] == 'moderate':
                mort = sc['case_fatality']
            else:
                mort = sc['case_fatality'] * 0.3

            # Antivenom protective
            if rec['antivenom_given']:
                mort *= 0.40

            # Delayed presentation increases mortality
            if rec['time_to_facility_hours'] > 12:
                mort *= 1.5
            if rec['time_to_facility_hours'] > 24:
                mort *= 1.5

            # Specific high-risk features
            if rec['respiratory_failure']:
                if not rec['mechanical_ventilation']:
                    mort = min(mort * 3, 0.80)
            if rec['acute_kidney_injury'] and rec['severity'] == 'severe':
                mort *= 1.5

            # Children more vulnerable
            if rec['age_years'] < 10:
                mort *= 1.5

        if rng.random() < min(mort, 0.70):
            rec['outcome'] = 'died'
        elif rec['amputation']:
            rec['outcome'] = 'survived_with_disability'
        elif rec['local_necrosis'] or rec['compartment_syndrome']:
            rec['outcome'] = 'survived_with_sequelae'
        else:
            rec['outcome'] = 'survived_full_recovery'

        # Hospital days
        if rec['outcome'] == 'died':
            rec['hospital_days'] = max(0, rng.integers(0, 7))
        elif rec['dry_bite'] or rec['severity'] in ('none', 'mild'):
            rec['hospital_days'] = rng.integers(1, 3)
        elif rec['severity'] == 'moderate':
            rec['hospital_days'] = rng.integers(2, 8)
        else:
            rec['hospital_days'] = rng.integers(5, 21)

        records.append(rec)

    df = pd.DataFrame(records)

    # ── Print Summary ──
    print(f"\n{'='*65}")
    print(f"Snakebite Envenomation — {scenario} (n={n}, seed={seed})")
    print(f"{'='*65}")

    print(f"\n  Dry bites: {df['dry_bite'].sum()} ({df['dry_bite'].mean()*100:.1f}%)")
    env = df[df['dry_bite'] == 0]
    print(f"  Envenomated: {len(env)}")

    print(f"\n  Top species:")
    for sp, ct in df['snake_species'].value_counts().head(5).items():
        print(f"    {sp}: {ct} ({ct/n*100:.1f}%)")

    print(f"\n  Syndrome (envenomated):")
    for syn, ct in env['envenomation_syndrome'].value_counts().items():
        print(f"    {syn}: {ct} ({ct/len(env)*100:.1f}%)")

    print(f"\n  Severity (envenomated):")
    for sev in ['mild', 'moderate', 'severe']:
        ct = (env['severity'] == sev).sum()
        print(f"    {sev}: {ct} ({ct/len(env)*100:.1f}%)")

    print(f"\n  Antivenom given: {df['antivenom_given'].sum()} "
          f"({df['antivenom_given'].mean()*100:.1f}%)")
    print(f"  Mean time to facility: {df['time_to_facility_hours'].mean():.1f} hrs")
    print(f"  Traditional treatment first: {df['traditional_treatment_first'].mean()*100:.1f}%")

    died = (df['outcome'] == 'died').sum()
    amp = df['amputation'].sum()
    print(f"\n  Deaths: {died} ({died/n*100:.1f}%)")
    print(f"  Amputations: {amp}")

    return df


if __name__ == '__main__':
    parser = argparse.ArgumentParser(
        description='Generate synthetic snakebite envenomation dataset')
    parser.add_argument('--scenario', type=str, default='district_hospital',
                        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_dataset(n=args.n, seed=args.seed, scenario=sc_name)
            out = os.path.join('data', f'snakebite_{sc_name}.csv')
            df.to_csv(out, index=False)
            print(f"  → Saved to {out}\n")
    else:
        df = generate_dataset(n=args.n, seed=args.seed, scenario=args.scenario)
        out = args.output or os.path.join('data', f'snakebite_{args.scenario}.csv')
        df.to_csv(out, index=False)
        print(f"  → Saved to {out}")