electricsheepafrica/africa-logistics-senegal

Essential Medicines Stockouts Dataset

Abstract

This dataset provides 30,000 simulated facility-level observations (10,000 per scenario) of essential medicine availability and stockout dynamics across three tiers of healthcare in sub-Saharan Africa. Each record represents a single WHO-recommended tracer drug assessed at a health facility during one quarterly reporting period. The dataset captures 45+ variables spanning facility infrastructure, drug characteristics, day-of-survey availability, stockout duration and frequency, root cause classification, supply chain performance metrics, storage quality indicators, and downstream patient impact. Three facility-level scenarios are modelled: central referral hospital (66% availability), district hospital (50%), and rural health centre (30%) — reflecting the steep gradient documented in multi-country surveys.

This dataset is entirely simulated. It must not be used for procurement decisions, clinical decision-making, or regulatory purposes.

1. Introduction

1.1 The Essential Medicines Crisis in Sub-Saharan Africa

Access to essential medicines is a foundational pillar of Universal Health Coverage, yet remains critically inadequate across much of sub-Saharan Africa (SSA). The World Health Organization estimates that 50–60% of the African population lacks consistent access to essential medicines (Pubmed, 2007). The WHO's Service Availability and Readiness Assessment (SARA) methodology, designed to monitor health facility preparedness, has repeatedly documented that average availability of tracer medicines in public-sector facilities falls well below the 80% target — in some WHO regions as low as 29.4% (Health Systems & Reform, 2019).

1.2 The Scale of Stockouts

Facility-level surveys across the continent paint a consistent picture of chronic shortages. In Ethiopia, a cross-sectional study at Shegaw Motta General Hospital found that while 80% of 15 tracer medicines were available on the day of survey, 60% had been stocked out at least once in the preceding six months, with a mean stockout duration of 38.5 days — and individual drugs reaching 157 days of stockout (PLOS ONE, 2022; doi:10.1371/journal.pone.0274776). At the attached health centre, availability was higher (93.3%) but stockouts still affected 20% of medicines with a mean duration of 11.2 days.

In Malawi, a comprehensive cross-sectional study of 50 essential medicines across 44 facilities found overall public-sector availability of just 48.5%, compared with 71.1% in retail pharmacies and 62.9% in CHAM (faith-based) facilities. Critically, availability varied sharply by facility level: 47% at primary health centres, 56% at district hospitals, and 66% at central hospitals (PLOS ONE, 2019; doi:10.1371/journal.pone.0212125). Pediatric formulations showed even worse availability — just 38.1% in public facilities, with essential antibiotic syrups (amoxicillin, cotrimoxazole, azithromycin) particularly scarce (Frontiers in Pharmacology, 2024; doi:10.3389/fphar.2024.1379250).

Multi-country analyses from Ghana, Kenya, and Uganda documented ACT (artemisinin-based combination therapy) stockout rates of 2–7% and amoxicillin stockouts of 5–21%, with ceftriaxone stockouts exceeding 10% in all three countries (PMC, 2014; PMC4263677).

1.3 Root Causes

The literature identifies a consistent set of interacting root causes for medicine stockouts in SSA:

• Poor quantification and forecasting: Health facilities frequently lack data-driven tools to predict demand. WHO (2017) identifies this as the most common contributor to drug shortages across SSA. A study of the Ethiopian public pharmaceutical supply chain found systemic forecasting challenges including reliance on historical consumption data that itself reflects suppressed demand from prior stockouts (PMC, 2024; PMC11207870).
• Fragmented and manual information systems: Many facilities manage inventory using paper-based records or no logistics management information system (LMIS) at all, leading to data errors and limited visibility (USAID, 2019).
• Inadequate storage and expiry management: Over 20% of Ethiopian health facilities were found to have expired medicines occupying shelf space while essential drugs were simultaneously out of stock (MSH, 2012). Without automated first-expiry-first-out (FEFO) compliance, wastage co-exists with shortage.
• Long and unreliable procurement lead times: Average lead times range from 14 days (central hospitals with direct national medical store access) to 60+ days (rural health centres dependent on irregular push-based distribution).
• Weak supply chain integration: Disconnection between procurement bodies, regional hubs, and health facilities results in irregular deliveries, incomplete orders, and slow replenishment cycles (Tadesse et al., 2021).
• Funding shortfalls: Government health budgets frequently fall short of procurement needs, forcing emergency procurement at higher costs.

1.4 Patient Impact

Medicine stockouts have direct, measurable consequences for patients. When essential drugs are unavailable, patients are turned away, referred to private pharmacies at higher cost, given substitute treatments of uncertain equivalence, or experience treatment delays. Regular stockouts erode public trust in government health facilities and drive patients toward unregulated private-sector outlets, contributing to antimicrobial resistance through incomplete courses and substandard medicines.

1.5 Rationale for This Dataset

Despite the wealth of survey evidence, no open, standardised, multi-variable dataset exists that integrates facility characteristics, drug-level availability, supply chain metrics, root cause classification, and patient impact indicators into a single analytical resource. This simulated dataset fills that gap, enabling researchers, supply chain analysts, and policymakers to develop and benchmark predictive models, explore stockout determinants, and prototype intervention strategies without requiring access to restricted government data.

2. Methodology

2.1 Epidemiological Parameterization

All scenario parameters were derived from peer-reviewed literature and WHO survey methodologies. The table below summarises key parameters and their sources:

Parameter	Central Hospital	District Hospital	Rural HC	Source
Day-of-survey availability	66%	50%	30%	[3] Malawi: 66% central, 56% district, 47% primary
Stocked out ≥1x in 6 months	32%	55%	75%	[2] Ethiopia: 60% hospital, 20% HC
Mean stockout duration (days)	10	21	37	[2] Ethiopia: 38.5d hospital, 11.2d HC
Expired stock on shelf	8%	15%	22%	[6] MSH: >20% in Ethiopia
Order fill rate	75%	55%	35%	[4] WHO regions
LMIS report submitted	90%	60%	25%	[6] USAID end-use verification
Stock card up to date	81%	50%	15%	[6] MedRafa inventory analysis
Pediatric formulation penalty	−15% availability	−15% availability	−15% availability	[1] Malawi: 38% vs 48% overall
NCD medicine penalty	−12% availability	−12% availability	−12% availability	[3] Valproate, captopril very poor

2.2 Tracer Drug Selection

The dataset includes 28 WHO-recommended tracer medicines spanning seven therapeutic categories aligned with the WHO SARA methodology and the 2023 WHO Model List of Essential Medicines (23rd edition):

Category	Tracer Drugs	VEN Class
Antibiotics (oral)	Amoxicillin 500mg, Cotrimoxazole 480mg, Metronidazole 250mg, Ciprofloxacin 500mg	E
Antibiotics (paediatric)	Amoxicillin 125mg/5ml suspension, Cotrimoxazole 240mg/5ml suspension	E
Antibiotics (injectable)	Gentamicin 40mg, Benzylpenicillin, Ceftriaxone 1g	V
Antimalarials	ACT (Artemether-Lumefantrine), Artesunate 60mg injection	V
Diarrhoea management	ORS sachets, Zinc 20mg tablets	V/E
Maternal/reproductive	Oxytocin 10IU, Magnesium sulfate, Ferrous+folic acid, DMPA injection	V/E
NCD medicines	Metformin, Glibenclamide, Amlodipine, Hydrochlorothiazide, Salbutamol, Phenobarbital, Sodium valproate, Insulin	E/V
Emergency	Diazepam 5mg injection	V
Analgesics	Paracetamol 500mg, Ibuprofen 200mg	E/N

2.3 Scenario Design

Three facility-level scenarios model the gradient of pharmaceutical infrastructure documented in the literature:

Scenario A — Central/Regional Referral Hospital Pharmacy Qualified pharmacist on staff, electronic LMIS (e.g., OpenLMIS, DHIS2), functional cold chain, direct procurement from national medical store, monthly resupply with 14-day average lead time, quarterly supervision visits. Analogous to level-3 facilities in Kenya (KEMSA), Tanzania (MSD), Ethiopia (EPSA), and Malawi.

Scenario B — District Hospital Pharmacy Pharmacy technician (no pharmacist), paper-based LMIS, limited cold chain, monthly resupply from regional hub with 30-day average lead time, no regular supervision. Analogous to level-2 facilities in Uganda, Rwanda, DRC.

Scenario C — Rural Health Centre Dispensary Nurse-managed, no LMIS, no cold chain, irregular push-based supply every 2–3 months with 60+ day lead times, no supervision. Analogous to primary facilities in Niger, CAR, rural Mozambique, South Sudan.

2.4 Record Structure

Each of the 10,000 records per scenario represents one observation of one tracer drug at one facility during one quarterly reporting period. The drug assignment cycles through all 28 tracers to ensure balanced representation. Facility IDs are randomly assigned from a pool of 200, simulating repeated observations across facilities.

3. Schema

3.1 Facility Characteristics

Column	Type	Description
facility_level	categorical	central_hospital / district_hospital / rural_health_centre
facility_id	string	Anonymised facility identifier (FAC_0001–FAC_0200)
region_type	categorical	urban / peri_urban / rural
has_pharmacist	binary	Qualified pharmacist on staff
pharmacy_staff_count	int	Number of pharmacy/dispensary staff
lmis_type	categorical	electronic / paper / none
lmis_functional	binary	LMIS operational at time of observation
cold_chain_functional	binary	Cold chain equipment operational
storage_adequate	binary	Storage meets minimum standards (ventilation, shelving, pest control)
supervised_last_quarter	binary	Supervisory visit in preceding 3 months

3.2 Drug Characteristics

Column	Type	Description
drug_name	categorical	28 WHO tracer medicines
drug_category	categorical	Therapeutic category (antibiotic, antimalarial, NCD, maternal, etc.)
formulation	categorical	tablet / capsule / suspension / injection / inhaler / sachet
ven_classification	categorical	V (Vital) / E (Essential) / N (Non-essential) per ABC-VEN analysis
unit_cost_usd	float	Approximate international reference unit cost (USD)

3.3 Availability & Stockout Metrics

Column	Type	Description
year	int	Observation year (2021–2024)
quarter	int	Observation quarter (1–4)
available_on_survey_day	binary	Drug physically present and usable on day of assessment
stocked_out_in_last_6m	binary	Any stockout episode in preceding 6 months
stockout_days_last_6m	int	Total days of stockout in preceding 6 months
stockout_episodes_last_6m	int	Number of discrete stockout episodes
longest_stockout_days	int	Duration of longest single stockout episode
stockout_cause_primary	categorical	Primary root cause (12 categories)
stockout_cause_secondary	categorical	Contributing root cause

3.4 Supply Chain Performance

Column	Type	Description
order_placed_on_time	binary	Routine order submitted within schedule
last_delivery_days_ago	int	Days since last supply delivery
lead_time_days	int	Procurement-to-delivery lead time (days)
delivery_complete	binary	Last delivery contained all ordered items
order_fill_rate_pct	float	Percentage of ordered quantity actually delivered

3.5 Storage & Quality

Column	Type	Description
expired_stock_found	binary	Expired medicines found on shelf
temperature_excursion	binary	Temperature-sensitive product exposed to excursion
damaged_stock_found	binary	Physically damaged stock identified
stock_card_up_to_date	binary	Bin card / stock card current
physical_count_matches_records	binary	Physical count matches LMIS/paper records
FEFO_compliance	binary	First-expiry-first-out dispensing practice observed

3.6 Patient Impact

Column	Type	Description
patients_turned_away	int	Patients turned away due to stockout (estimated daily)
referred_to_private_pharmacy	binary	Patient referred to buy from private sector
treatment_substituted	binary	Alternative medicine substituted
treatment_delayed	binary	Treatment delayed pending resupply

3.7 Reporting & Data Quality

Column	Type	Description
LMIS_report_submitted	binary	Quarterly LMIS report submitted to district/region
LMIS_report_timely	binary	Report submitted within deadline
LMIS_report_accurate	binary	Report data matches physical verification

4. Validation

4.1 Key Validation Results

Metric	Central Hospital	District Hospital	Rural HC	Literature Target
Day-of-survey availability	66%	50%	30%	66% / 56% / 47% [3]
Stocked out in 6 months	32%	55%	75%	60% hospital [2]
Mean stockout duration	10 days	21 days	37 days	38.5d / 11.2d [2]
Expired stock on shelf	8%	15%	22%	>20% [6]
Order fill rate	75%	55%	35%	29–75% range [4]
LMIS report submitted	90%	60%	25%	Literature range
Stock card up to date	81%	50%	15%	Literature range

4.2 Internal Consistency Checks

• Pediatric formulations show 15 percentage points lower availability than adult equivalents across all scenarios, consistent with Frontiers in Pharmacology (2024) finding of 38.1% vs 48.5%.
• NCD medicines (metformin, amlodipine, insulin, valproate) show the lowest availability within each scenario, consistent with PLOS ONE (2019) documenting poor NCD medicine access.
• Vital (V) medicines show marginally higher availability than Essential (E) and Non-essential (N), reflecting priority-based procurement.
• Stockout duration is inversely correlated with LMIS functionality, pharmacist presence, and supervision frequency.
• Patient impact metrics (turned away, referred to private, treatment delayed) are conditional on unavailability, ensuring logical consistency.

5. Usage

from datasets import load_dataset

# Load a specific scenario
dataset = load_dataset(
    "electricsheepafrica/essential-medicines-stockouts",
    "district_hospital"
)
df = dataset["train"].to_pandas()

# Analyse availability by drug category
avail_by_cat = df.groupby('drug_category')['available_on_survey_day'].mean()
print(avail_by_cat.sort_values())

# Identify highest-risk drugs
high_risk = df.groupby('drug_name').agg({
    'available_on_survey_day': 'mean',
    'stockout_days_last_6m': 'mean'
}).sort_values('available_on_survey_day')
print(high_risk.head(10))

5.1 Suggested Analyses

• Stockout prediction: Train classifiers to predict stocked_out_in_last_6m from facility and supply chain features.
• Root cause clustering: Unsupervised analysis of stockout cause patterns across facility types.
• Intervention modelling: Simulate impact of electronic LMIS adoption, pharmacist placement, or increased supervision on availability.
• Equity analysis: Compare availability gradients across facility levels and urban/rural settings.
• Drug-level risk scoring: Identify which tracer drugs are most vulnerable to stockout by category and formulation.

6. Limitations

• Simulated data: Generated from statistical models, not from real facility surveys. Distributions approximate but do not replicate any specific country's supply chain.
• No temporal dynamics: Each record is an independent observation; the dataset does not model longitudinal supply chain behaviour, seasonal demand patterns, or intervention roll-outs over time.
• Simplified root causes: Real stockout causation is multifactorial and context-dependent. The 12-category classification is a pragmatic simplification.
• No cost modelling: Unit costs are approximate international reference prices and do not reflect local procurement costs, markups, or currency fluctuations.
• No geographical specificity: Facility IDs and regions are anonymised and do not correspond to real locations.
• Single drug per record: Each observation assesses one drug at one facility. Cross-drug stockout correlations within a facility are not explicitly modelled.

7. References

1. Khuluza, F. et al. (2024). Availability, pricing, and affordability of essential medicines for pediatric population in Malawi. Frontiers in Pharmacology, 15, 1379250. doi:10.3389/fphar.2024.1379250
2. Endalew, A. et al. (2022). Availability and stock-out duration of essential medicines in Shegaw Motta general hospital and Motta Health Centre, North West Ethiopia. PLOS ONE, 17(9), e0274776. doi:10.1371/journal.pone.0274776
3. Khuluza, F. & Heide, L. (2019). The availability, prices and affordability of essential medicines in Malawi: A cross-sectional study. PLOS ONE, 14(2), e0212125. doi:10.1371/journal.pone.0212125
4. Yadav, P. (2019). Improving supply chain for essential drugs in developing countries. Health Systems & Reform, 5(2). doi:10.1080/23288604.2019.1596050
5. Masters, S.H. et al. (2014). Pharmaceutical availability across levels of care: evidence from facility surveys in Ghana, Kenya, and Uganda. PLOS ONE, 9(12). PMC4263677
6. MedRafa (2025). The persistent problem of medicine stock-outs: inventory management challenges and solutions in developing health systems. medrafa.et
7. PMC (2024). Challenges and the way forward in demand-forecasting practices in Ethiopian public pharmaceutical supply chain. PMC11207870
8. WHO (2023). Model List of Essential Medicines, 23rd list. WHO-MHP-HPS-EML-2023.02
9. WHO (2017). Service Availability and Readiness Assessment (SARA): An annual monitoring system for service delivery.
10. Management Sciences for Health (MSH) (2012). MDS-3: Managing Access to Medicines and Health Technologies.
11. USAID (2019). End Use Verification Survey Reports — Ethiopia.
12. UNICEF (2020). Health Supply Chain Analysis in Ethiopia.
13. Tadesse, A. et al. (2021). Assessment of medicine stock-outs in Ethiopian public hospitals. Ethiopian Journal of Health Development.
14. Ethiopian Ministry of Health (2023). National Health Commodities Management Strategy.

Citation

@dataset{esa_essential_medicines_stockouts_2025,
  title   = {Essential Medicines Stockouts Dataset: Tracer Drug Availability,
             Supply Chain Performance, and Patient Impact Across Three Tiers
             of Healthcare in Sub-Saharan Africa},
  author  = {{Electric Sheep Africa}},
  year    = {2025},
  publisher = {Hugging Face},
  url     = {https://huggingface.co/datasets/electricsheepafrica/essential-medicines-stockouts},
  note    = {Simulated dataset. Not for clinical or procurement use.}
}

License

CC-BY-4.0