Randomized trial of AI clinical tool in Kenya finds improved note quality but no statistically significant patient outcome benefit

A study published this summer in Nature Medicine tested whether an AI tool called AI Consult could improve care at primary clinics in Kenya operated by Penda Health. The trial enrolled nearly 10,000 patient encounters across 16 sites, with half of clinical officers receiving real-time feedback from GPT-4o — OpenAI's large language model — as they typed notes, and the other half using a computer without the AI.

The tool operates on a traffic-light system: a green indicator signals no issues, yellow prompts the clinician to review a potential gap in care, and red flags a critical concern requiring prompt action. The cost of the intervention was reported at 4 cents per patient encounter.

An independent panel of six Kenyan family physicians reviewed the notes and judged that clinicians using AI Consult produced better diagnoses and treatment plans. One clinical officer, Vyonne Njeri, described a case in which a yellow alert prompted her to check the heart rate of a 4-month-old she had initially assessed as having a cold; she subsequently detected a heart murmur and referred the infant to a specialist who confirmed a congenital heart defect. NPR, the sole outlet covering the story, led with Njeri's account to illustrate the tool's potential real-world benefit.