Trials for LLM-supported clinical decisions in African primary healthcare

Primary healthcare systems in sub-Saharan Africa face considerable challenges, particularly constrained clinical capacity¹ and variable quality-of-care delivery². Large language models (LLMs) represent a potentially transformative solution for clinical decision-making. Some LLMs have shown that they can recall clinical knowledge at expert level³ and can diagnose clinical vignettes (in simulated settings) more accurately than clinicians⁴. However, despite the fact that LLMs have shown promise, there is a striking lack of real-world evidence for their safe and effective use in clinical settings in Africa. Thus far, only handful of randomized trials of artificial intelligence (AI) for health (and none for generative AI tools) have been conducted on the continent⁵.

In Kenya, a pragmatic randomized controlled trial (RCT) has been embedded into routine care delivery undertaken at 16 of Penda Health’s clinics⁶, a Nairobi-based social enterprise that provides primary care services across all socioeconomic groups, from state social insurance recipients to privately insured and out-of-pocket payers. The trial aims to enroll 9,000 patients. The solution being trialed is an LLM-based ‘co-pilot’ feature that has been integrated directly into Penda’s electronic medical record system. Clinicians randomly assigned to the intervention arm receive automated, real-time suggestions for diagnosis, treatment planning and lab interpretation during patient consultations.