| |
| """ |
| Literature-Informed Medical Oxygen Supply Dataset |
| ==================================================== |
| |
| Generates realistic synthetic facility-level observations of medical |
| oxygen availability, delivery systems, equipment status, and patient |
| outcomes across three tiers of healthcare in sub-Saharan Africa. |
| |
| Each record represents ONE facility observation for ONE monthly period. |
| |
| Epidemiological Parameterization (web-searched): |
| ------------------------------------------------- |
| [1] Oxygen Hub / ITT (2021). Closing the medical oxygen gap in SSA. |
| Most SSA countries have <10% of volume needed. PSA vs LOX |
| trade-offs. Decentralized hub-and-spoke model proposed. |
| |
| [2] PMC (2022). Oxygen inequity in COVID-19 pandemic and beyond. |
| Ghana, Senegal SPRINT pilot. UNICEF workgroups. PMC9972372. |
| |
| [3] PMC (2022). Comprehensive approach to medical oxygen ecosystem. |
| 165 PSA plants needing repair globally; 151 in SSA. PMC9771461. |
| |
| [4] PMC (2021). Oxygen delivery systems for adults in SSA. Scoping |
| review. Hypoxemia prevalence 11-89%. High mortality among |
| hypoxemic patients. PMC8109278. |
| |
| [5] BMC Health Services Research (2025). Design and maintenance of |
| oxygen concentrators in SSA. WHO distributed >30,000 OCs. |
| Maintenance challenges documented. doi:10.1186/s12913-025-12315-6 |
| |
| [6] PMC (2024). Functional availability of medical oxygen for |
| pneumonia management. ~50% of facilities in resource-limited |
| settings had no or inconsistent oxygen. PMC11082622. |
| |
| [7] WHO (2023). Essential medicines list — medical oxygen included. |
| Pulse oximetry essential for identifying hypoxemia. |
| """ |
|
|
| import numpy as np |
| import pandas as pd |
| import argparse |
| import os |
|
|
| OXYGEN_SOURCES = [ |
| 'PSA_plant_onsite', 'LOX_bulk_tank', 'oxygen_concentrator', |
| 'cylinder_piped', 'cylinder_portable', 'none' |
| ] |
|
|
| PATIENT_CONDITIONS = [ |
| 'pneumonia_child', 'pneumonia_adult', 'neonatal_respiratory', |
| 'COPD_exacerbation', 'COVID19_severe', 'severe_malaria', |
| 'heart_failure', 'surgical_anaesthesia', 'trauma', |
| 'asthma_severe', 'sepsis', 'other_hypoxemia' |
| ] |
|
|
| SHORTAGE_CAUSES = [ |
| 'cylinder_delivery_delay', 'PSA_plant_breakdown', 'concentrator_malfunction', |
| 'power_outage', 'empty_cylinders_not_collected', 'funding_for_refill', |
| 'no_oxygen_source', 'demand_surge', 'supplier_stockout', |
| 'piping_system_leak', 'regulator_valve_failure', |
| ] |
|
|
| SCENARIOS = { |
| 'referral_hospital': { |
| 'description': ( |
| 'Urban referral/teaching hospital with PSA plant or LOX ' |
| 'bulk tank, piped oxygen system, pulse oximetry, ICU, ' |
| 'biomedical technician. Analogous to Muhimbili (TZ), ' |
| 'Kenyatta (KE), Mulago (UG).' |
| ), |
| 'facility_level': 'referral_hospital', |
| 'has_piped_system': True, |
| 'has_icu': True, |
| 'has_pulse_oximeter': True, |
| 'has_biomedical_tech': True, |
| 'oxygen_source_probs': [0.25, 0.30, 0.20, 0.20, 0.05, 0.00], |
| 'oxygen_available_rate': 0.82, |
| 'sufficient_quantity_rate': 0.65, |
| 'concentrator_functional_rate': 0.70, |
| 'pulse_ox_functional_rate': 0.85, |
| 'cylinder_refill_days': 5, |
| 'mortality_hypoxic_untreated': 0.35, |
| }, |
| 'district_hospital': { |
| 'description': ( |
| 'District hospital with oxygen concentrators and/or ' |
| 'cylinders, limited pulse oximetry, no ICU, clinical ' |
| 'officer manages oxygen. Analogous to district hospitals ' |
| 'in Malawi, Rwanda, Mozambique.' |
| ), |
| 'facility_level': 'district_hospital', |
| 'has_piped_system': False, |
| 'has_icu': False, |
| 'has_pulse_oximeter': False, |
| 'has_biomedical_tech': False, |
| 'oxygen_source_probs': [0.02, 0.05, 0.35, 0.15, 0.30, 0.13], |
| 'oxygen_available_rate': 0.50, |
| 'sufficient_quantity_rate': 0.35, |
| 'concentrator_functional_rate': 0.45, |
| 'pulse_ox_functional_rate': 0.40, |
| 'cylinder_refill_days': 21, |
| 'mortality_hypoxic_untreated': 0.45, |
| }, |
| 'rural_health_centre': { |
| 'description': ( |
| 'Rural health centre with no permanent oxygen source, ' |
| 'occasional cylinder if available, no pulse oximetry, ' |
| 'nurse-managed. Analogous to health centres in Niger, ' |
| 'DRC, South Sudan, rural Ethiopia.' |
| ), |
| 'facility_level': 'rural_health_centre', |
| 'has_piped_system': False, |
| 'has_icu': False, |
| 'has_pulse_oximeter': False, |
| 'has_biomedical_tech': False, |
| 'oxygen_source_probs': [0.00, 0.00, 0.08, 0.02, 0.20, 0.70], |
| 'oxygen_available_rate': 0.12, |
| 'sufficient_quantity_rate': 0.08, |
| 'concentrator_functional_rate': 0.15, |
| 'pulse_ox_functional_rate': 0.08, |
| 'cylinder_refill_days': 45, |
| 'mortality_hypoxic_untreated': 0.55, |
| }, |
| } |
|
|
|
|
| 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} |
|
|
| |
| rec['facility_level'] = sc['facility_level'] |
| rec['facility_id'] = f"O2_{rng.integers(1, 200):04d}" |
| rec['region_type'] = rng.choice( |
| ['urban', 'peri_urban', 'rural'], |
| p=[0.10, 0.15, 0.75] if scenario == 'rural_health_centre' |
| else ([0.55, 0.25, 0.20] if scenario == 'referral_hospital' |
| else [0.20, 0.35, 0.45])) |
| rec['bed_count'] = max(5, int(rng.normal( |
| 250 if scenario == 'referral_hospital' else |
| (60 if scenario == 'district_hospital' else 12), 40))) |
| rec['has_icu'] = 1 if sc['has_icu'] else (1 if rng.random() < 0.02 else 0) |
| rec['has_nicu'] = 1 if sc['has_icu'] else (1 if rng.random() < 0.05 else 0) |
| rec['has_piped_oxygen'] = 1 if sc['has_piped_system'] else ( |
| 1 if rng.random() < 0.05 else 0) |
| rec['has_biomedical_technician'] = 1 if sc['has_biomedical_tech'] else ( |
| 1 if rng.random() < 0.05 else 0) |
|
|
| |
| rec['primary_oxygen_source'] = rng.choice( |
| OXYGEN_SOURCES, p=sc['oxygen_source_probs']) |
|
|
| rec['concentrator_count'] = 0 |
| if rec['primary_oxygen_source'] == 'oxygen_concentrator' or rng.random() < 0.20: |
| rec['concentrator_count'] = max(0, int(rng.poisson( |
| 5 if scenario == 'referral_hospital' else |
| (2 if scenario == 'district_hospital' else 0.3)))) |
| rec['concentrator_functional'] = 0 |
| if rec['concentrator_count'] > 0: |
| func_rate = sc['concentrator_functional_rate'] |
| rec['concentrator_functional'] = max(0, min( |
| rec['concentrator_count'], |
| int(rec['concentrator_count'] * rng.normal(func_rate, 0.2)))) |
| rec['concentrator_mean_age_years'] = max(0, round(rng.exponential( |
| 3 if scenario == 'referral_hospital' else |
| (5 if scenario == 'district_hospital' else 8)), 1)) |
| rec['concentrator_maintenance_available'] = 1 if rec['has_biomedical_technician'] else ( |
| 1 if rng.random() < 0.10 else 0) |
|
|
| rec['cylinder_count_full'] = max(0, int(rng.poisson( |
| 10 if scenario == 'referral_hospital' else |
| (3 if scenario == 'district_hospital' else 0.5)))) |
| rec['cylinder_count_empty'] = max(0, int(rng.poisson( |
| 5 if scenario == 'referral_hospital' else |
| (4 if scenario == 'district_hospital' else 1)))) |
| rec['days_since_cylinder_refill'] = max(0, int(rng.exponential( |
| sc['cylinder_refill_days']))) |
|
|
| rec['PSA_plant_functional'] = 0 |
| if rec['primary_oxygen_source'] == 'PSA_plant_onsite': |
| rec['PSA_plant_functional'] = 1 if rng.random() < 0.55 else 0 |
|
|
| |
| rec['pulse_oximeter_available'] = 1 if sc['has_pulse_oximeter'] else ( |
| 1 if rng.random() < (0.35 if scenario == 'district_hospital' else 0.05) else 0) |
| rec['pulse_oximeter_functional'] = 0 |
| if rec['pulse_oximeter_available']: |
| rec['pulse_oximeter_functional'] = 1 if rng.random() < sc['pulse_ox_functional_rate'] else 0 |
| rec['SpO2_screening_routine'] = 0 |
| if rec['pulse_oximeter_functional']: |
| rec['SpO2_screening_routine'] = 1 if rng.random() < ( |
| 0.70 if scenario == 'referral_hospital' else |
| (0.30 if scenario == 'district_hospital' else 0.05)) else 0 |
|
|
| |
| rec['oxygen_available_today'] = 1 if rng.random() < sc['oxygen_available_rate'] else 0 |
| rec['oxygen_sufficient_for_demand'] = 0 |
| if rec['oxygen_available_today']: |
| rec['oxygen_sufficient_for_demand'] = 1 if rng.random() < sc['sufficient_quantity_rate'] / sc['oxygen_available_rate'] else 0 |
| rec['oxygen_stockout_days_last_month'] = 0 |
| if not rec['oxygen_available_today']: |
| rec['oxygen_stockout_days_last_month'] = max(1, min(30, |
| int(rng.exponential(12)))) |
| elif rng.random() < 0.20: |
| rec['oxygen_stockout_days_last_month'] = max(1, min(15, |
| int(rng.exponential(5)))) |
|
|
| rec['shortage_cause'] = 'not_applicable' |
| if not rec['oxygen_available_today'] or rec['oxygen_stockout_days_last_month'] > 0: |
| if scenario == 'rural_health_centre': |
| cause_p = [0.15, 0.02, 0.05, 0.12, 0.05, 0.10, 0.35, 0.08, 0.05, 0.01, 0.02] |
| elif scenario == 'district_hospital': |
| cause_p = [0.25, 0.05, 0.18, 0.15, 0.08, 0.08, 0.05, 0.06, 0.04, 0.03, 0.03] |
| else: |
| cause_p = [0.10, 0.20, 0.15, 0.15, 0.05, 0.05, 0.02, 0.15, 0.05, 0.05, 0.03] |
| rec['shortage_cause'] = rng.choice(SHORTAGE_CAUSES, p=cause_p) |
|
|
| |
| rec['year'] = rng.choice([2021, 2022, 2023, 2024], |
| p=[0.15, 0.25, 0.30, 0.30]) |
| rec['month'] = rng.integers(1, 13) |
|
|
| |
| rec['patients_needing_oxygen'] = max(0, int(rng.poisson( |
| 40 if scenario == 'referral_hospital' else |
| (12 if scenario == 'district_hospital' else 3)))) |
| rec['primary_condition'] = rng.choice(PATIENT_CONDITIONS, |
| p=[0.15, 0.12, 0.12, 0.08, 0.10, 0.08, 0.06, 0.08, 0.06, 0.05, 0.05, 0.05]) |
| rec['patients_received_oxygen'] = 0 |
| if rec['oxygen_available_today'] and rec['patients_needing_oxygen'] > 0: |
| coverage = rng.normal( |
| 0.80 if scenario == 'referral_hospital' else |
| (0.45 if scenario == 'district_hospital' else 0.10), 0.15) |
| rec['patients_received_oxygen'] = max(0, min( |
| rec['patients_needing_oxygen'], |
| int(rec['patients_needing_oxygen'] * np.clip(coverage, 0, 1)))) |
| rec['patients_untreated_hypoxemia'] = max(0, |
| rec['patients_needing_oxygen'] - rec['patients_received_oxygen']) |
|
|
| rec['flow_rate_adequate'] = 0 |
| if rec['patients_received_oxygen'] > 0: |
| rec['flow_rate_adequate'] = 1 if rng.random() < ( |
| 0.75 if scenario == 'referral_hospital' else |
| (0.40 if scenario == 'district_hospital' else 0.15)) else 0 |
|
|
| |
| rec['deaths_hypoxemia_related'] = 0 |
| if rec['patients_untreated_hypoxemia'] > 0: |
| mort_rate = sc['mortality_hypoxic_untreated'] |
| rec['deaths_hypoxemia_related'] = max(0, |
| int(rng.binomial(rec['patients_untreated_hypoxemia'], mort_rate))) |
| rec['referred_for_oxygen'] = 0 |
| if not rec['oxygen_available_today'] and rec['patients_needing_oxygen'] > 0: |
| rec['referred_for_oxygen'] = max(0, int( |
| rec['patients_needing_oxygen'] * rng.normal(0.30, 0.15))) |
|
|
| |
| rec['monthly_oxygen_cost_usd'] = max(0, round(rng.normal( |
| 800 if scenario == 'referral_hospital' else |
| (200 if scenario == 'district_hospital' else 30), |
| 150 if scenario == 'referral_hospital' else |
| (80 if scenario == 'district_hospital' else 20)), 0)) |
| rec['distance_to_refill_km'] = max(0, round(rng.exponential( |
| 15 if scenario == 'referral_hospital' else |
| (50 if scenario == 'district_hospital' else 120)), 0)) |
| rec['transport_available_for_cylinders'] = 1 if rng.random() < ( |
| 0.85 if scenario == 'referral_hospital' else |
| (0.40 if scenario == 'district_hospital' else 0.10)) else 0 |
|
|
| |
| rec['power_source'] = rng.choice( |
| ['grid_reliable', 'grid_unreliable', 'generator_only', 'solar', 'none'], |
| p=[0.40, 0.30, 0.15, 0.10, 0.05] if scenario == 'referral_hospital' |
| else ([0.10, 0.35, 0.15, 0.15, 0.25] if scenario == 'district_hospital' |
| else [0.02, 0.10, 0.05, 0.08, 0.75])) |
| rec['power_outage_hours_last_week'] = max(0, int(rng.exponential( |
| 5 if scenario == 'referral_hospital' else |
| (15 if scenario == 'district_hospital' else 60)))) |
|
|
| records.append(rec) |
|
|
| df = pd.DataFrame(records) |
|
|
| print(f"\n{'='*65}") |
| print(f"Medical Oxygen Supply — {scenario} (n={n}, seed={seed})") |
| print(f"{'='*65}") |
| print(f"\n Oxygen available today: {df['oxygen_available_today'].mean()*100:.1f}%") |
| print(f" Sufficient for demand: {df['oxygen_sufficient_for_demand'].mean()*100:.1f}%") |
| print(f" Pulse oximeter functional: {df['pulse_oximeter_functional'].mean()*100:.1f}%") |
| print(f" Patients needing O2 (mean): {df['patients_needing_oxygen'].mean():.1f}") |
| print(f" Patients received O2 (mean): {df['patients_received_oxygen'].mean():.1f}") |
| print(f" Deaths hypoxemia (mean): {df['deaths_hypoxemia_related'].mean():.2f}") |
|
|
| return df |
|
|
|
|
| if __name__ == '__main__': |
| parser = argparse.ArgumentParser( |
| description='Generate medical oxygen supply 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'oxygen_{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'oxygen_{args.scenario}.csv') |
| df.to_csv(out, index=False) |
| print(f" -> Saved to {out}") |
|
|