evidoria

← Back to browse

Good practice Imported

Sierra Leone's Decision-Aware Machine Learning Tool Lifts Essential-Medicine Consumption 19% Nationwide for $30 a Month

Sierra Leone · Freetown · See the Sierra Leone profile · See the Freetown profile

Evidence: Quasi-experimental Top 6% 87/100 · Ask Evidence Copilot about this practice

Sierra Leone's National Medical Supplies Agency and University of Pennsylvania researchers built a decision-aware ML tool that raised essential-medicine consumption 19% (p<0.01) in a five-district pilot, then scaled nationwide for $30/month, reaching 2 million women and children.

19 %
Increase in consumption of allocated products in treated districts (p<0.01) (May 2023 pilot)
32 %
Increase in consumption among the poorest, most remote facilities (May 2023 pilot)
18-21 %
Effect range across alternative estimators
1,058
Health facilities analysed
~2 million women and children under five
Beneficiaries reached after national scale-up (from Q3 2023)
$30 per month
Monthly running cost after national scale-up
Sierra Leone's Decision-Aware Machine Learning Tool Lifts Essential-Medicine Consumption 19% Nationwide for $30 a Month

Details

Maturity
Established
Promoter
Sierra Leone National Medical Supplies Agency (NMSA); University of Pennsylvania (Wharton School & School of Engineering and Applied Science)
Period
2023-2026
Keywords
health, public administration, machine learning, supply chain, social protection

Context

Sierra Leone's National Medical Supplies Agency (NMSA) procures and distributes essential medicines - more than 70 products used in maternal, newborn and child health - to public health facilities nationwide. Like many low-income countries, NMSA historically allocated fixed stock to facilities using rough rules of thumb, leading to chronic stockouts at the poorest, most remote clinics even while other facilities held surplus.

Objectives

Build a low-cost, decision-aware machine-learning system to forecast facility-level demand and compute how to distribute limited national stock to maximise coverage, while leaving officials the final say and the ability to override any recommendation.

Activities

Researchers Hamsa Bastani and Osbert Bastani (University of Pennsylvania) and PhD candidate Angel Tsai-Hsuan Chung worked with NMSA and Sierra Leone's Ministry of Health and Sanitation to combine multitask learning across products and facilities, catalytic priors to handle sparse data, and external covariates (census and satellite-derived population data). The system was piloted from May 2023 across five of Sierra Leone's sixteen districts (Tonkolili, Falaba, Karene, Kono and Pujehun), with the remaining eleven districts serving as controls.

Results

A synthetic difference-in-differences analysis across 1,058 facilities found a 19% increase in consumption of allocated products in treated districts (p<0.01) - a proxy for improved access - rising to 32% among facilities serving the poorest, most remote populations that had suffered chronic stockouts. Alternative estimators (standard difference-in-differences, geographic matching, alternative control products) corroborated the effect, ranging from 18% to 21%. Based on the pilot results, Sierra Leone's government scaled the tool nationwide from the third quarter of 2023; it now manages allocation of more than 70 essential products, reaching an estimated 2 million women and children under five, at a running cost of about $30 a month in server fees with no additional staff.

Conclusions

Ownership of the tool has been fully transferred to the Sierra Leonean government, and the results were published in Nature in April 2026.

Implementation
Indicative cost
Low (< €50k)
Time to results
Medium (1–3 years)
Staffing & skills
University of Pennsylvania research team (Hamsa Bastani, Osbert Bastani, Angel Tsai-Hsuan Chung), Sierra Leone National Medical Supplies Agency (NMSA) as implementing and now owning institution, Sierra Leone Ministry of Health and Sanitation as government partner

Conditions for success

  • Officials retain final say and can override any model recommendation
  • Use of external covariates (census and satellite-derived population data) to handle sparse facility-level data
  • Full ownership transfer from the university research team to the national government agency for sustainability
  • Very low ongoing running cost (~$30/month in server fees) with no additional staffing needs

Commonly funded by

Horizon Europe Philanthropic / foundation funding National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Replication kit

Reusable artefacts from this practice — as published by their sources.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful