Cluster Randomized Trial of ChatGPT-4 Assisted Decision Support in Kenyan Primary Care Facilities

Kenyan Trial Finds AI Integration Fails to Cut Treatment Failures

A pragmatic cluster-randomized trial published in Nature Medicine on 26 June 2026 found that integrating ChatGPT-4o into clinical decision support systems at Kenyan primary care facilities did not significantly reduce 14-day treatment failure rates compared to standard care, according to the study’s authors. The research, conducted across 42 facilities in Kenya, involved numerous patients with common acute illnesses such as pneumonia, malaria, and diarrhea.

42 Clinics, Numerous Patients: The Trial’s Scope

The trial, enrolled 42 primary care facilities between January and December 2025. Facilities were randomly assigned to either a control group receiving usual care or an intervention group where clinicians used ChatGPT-4o to assist in diagnosing and treating patients. The AI tool provided real-time recommendations based on patient symptoms, medical history, and clinical guidelines.

No Significant Gap Between AI and Standard Care

Researchers tracked 14-day treatment failure rates, defined as hospitalization, death, or unscheduled return visits within two weeks of initial care. Data were collected from electronic health records and verified through follow-up interviews with patients and providers. The study’s primary outcome measure was the difference in treatment failure rates between the two groups. The study found no statistically significant difference in 14-day treatment failure rates between the two groups, with both groups showing similar rates. These results remained consistent across subgroups, including patients with varying illness severities and facility types.

AI’s Limits Exposed in Resource-Strapped Health Systems

The lack of significant difference suggests that AI tools may not address systemic challenges in resource-limited healthcare settings.”

In Kenya, where primary care facilities often face staff shortages and limited diagnostic capabilities, the integration of AI tools like ChatGPT-4o was seen as a potential solution to reduce diagnostic errors and improve patient outcomes. However, the study’s results suggest that technological interventions alone may not be sufficient to overcome these challenges.

Reimagining AI: Tailoring Tools for Low-Resource Settings

In settings with high provider workloads and limited resources, AI may not compensate for systemic gaps.”

The researchers emphasized the need for further studies to explore how AI tools could be adapted to better suit the needs of low-resource settings. Potential areas for improvement include tailoring AI recommendations to local clinical guidelines, enhancing user interfaces for healthcare workers with varying levels of digital literacy, and integrating AI with existing diagnostic technologies.

Global AI Trials Show Mixed Results, but Context Matters

The trial aligns with ongoing debates about the scalability of AI in healthcare. Earlier studies in