KenyaHealthcare and life sciences2025 – ongoingPenda Health
AI Consult clinical decision support
Clinicians using an LLM safety net in Nairobi clinics made 16% fewer diagnostic errors, though a later trial found no clear patient benefit
Problem
Penda Health runs primary care clinics in Nairobi where clinical officers see high volumes of patients with limited time per visit. Missed history questions, wrong diagnoses and inappropriate treatments are common risks in this setting.
Approach
Penda worked with OpenAI on AI Consult, a GPT-4o-based tool delivered through the Easy Clinic medical records platform. It reads clinicians' notes in the electronic record in the background and flags possible problems in a traffic-light format, with red alerts requiring review. In 2025 Penda and OpenAI compared 39,849 visits across 15 clinics, with clinicians split into groups with and without the tool and independent physicians rating care quality blind to the group. A separate randomised trial sponsored by the non-profit PATH was published in Nature Medicine in 2026.
Results
Takeaway
The PATH-sponsored trial found better diagnoses and treatment plans at about 4 US cents per patient. It found no statistically clear improvement in patient outcomes because serious failures were too rare to measure.
Categories
- Industry
- Healthcare and life sciences
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Generative AI
- Adoption
- Co-developed with a partner
- Goal
- Quality and risk
- Evidence
- Company's own claim
Sources
Source summaries
Penda Health and OpenAI staff compare 39,849 visits across 15 clinics with and without AI Consult. They report 16% fewer diagnostic errors and 13% fewer treatment errors in the AI group, and project about 22,000 diagnostic errors averted per year at Penda. All surveyed clinicians said the tool improved care, and 75% called the effect substantial.
The Kenyan outlet reports the study's 16% and 13% reductions in diagnostic and treatment errors across about 40,000 visits. It says the tool was delivered through the Easy Clinic medical records platform and that the study had approval from an AMREF ethics committee and Kenya's Ministry of Health. It quotes a 32% drop in history-taking omissions and 75% clinician adoption.
NPR covers a randomised trial of nearly 10,000 encounters in 16 Penda clinics, sponsored by PATH, funded by the Gates Foundation and published in Nature Medicine. A panel of Kenyan family physicians judged diagnoses and treatment plans better with the tool, which cost about 4 cents per patient. Treatment failures fell 23%, but the difference was not statistically significant, and the authors said a far larger trial would be needed to show patient benefit.
Full details are in the original sources.
