SIERRA LEONE — The Sierra Leone government has implemented a machine learning-based decision-support system nationwide to improve the allocation of essential medical supplies.

Hamsa Bastani, an operations researcher and statistician at the Wharton School, along with Osbert Bastani and Angel Tsai-Hsuan Chung, collaborated with the Sierra Leone government to develop the system. The machine learning system was initially piloted in five districts across the country.

During the pilot phase, researchers observed a 19% increase in the consumption of allocated medical products in the areas where the system was used. Facilities serving poorer and more remote populations experienced an even greater increase, with medicine consumption rising by 32% with the new tool. The findings from this pilot program were subsequently published in the journal Nature.

The system now supports allocation decisions for over 70 essential products throughout Sierra Leone. It is estimated to reach approximately two million women and children under five. The system incorporates external information, including census data and Google Earth images of vegetation around clinics, to forecast demand for medical supplies. It also uses multitask learning to apply patterns from areas with more extensive data to locations where records are limited. Angel Tsai-Hsuan Chung designed a web application for the system that emulates the agency's existing spreadsheet workflows and conducted personalized training sessions with local officials in Freetown.

Ownership of the allocation tool has been transferred to the Sierra Leone government. Angel Tsai-Hsuan Chung, a Ph.D. candidate, said, "It is designed for a setting where data are sparse, noisy, and often incomplete." The system operates with server costs of $30 per month. Sierra Leone has a maternal mortality rate of 717 deaths per 100,000 live births. Hamsa Bastani said, "Crucially, the system chiefly functions as a 'decision-support' tool wherein local officials always retain final say and can override recommendations."