AI tool cuts diagnostic errors by 16% in Kenyan clinics, study finds

The study, which analysed 40,000 patient visits, found that clinicians using the "AI Consult" tool—developed by Easy Clinic and deployed via the Easy Clinic EMR platform—made 16% fewer diagnostic errors and 13% fewer treatment errors compared to a control group not using the tool.

The quality improvement study, approved by the AMREF Health Africa Ethical and Scientific Review Committee and Kenya’s Ministry of Health, aimed to address the long-standing "model-implementation gap"—the disconnect between AI’s theoretical promise and its actual clinical impact.

AI Consult functions as an asynchronous safety net, activating at key workflow points such as diagnosis and treatment to provide real-time feedback. To reduce alert fatigue, it uses a tiered traffic-light system to signal the severity of potential issues, helping maintain clinician autonomy while ensuring safety with minimal cognitive load.

“These findings are encouraging,” said Girish Mohata, CEO of Easy Clinic. “We saw a 16% drop in diagnostic errors, a 32% reduction in history-taking omissions, and 75% clinician adoption—without requiring additional workload.”

The study used a randomised design, with clinicians assigned to either the AI-supported group or the control group, across 15 clinics. Independent physicians, blinded to group assignments, evaluated the quality of care based on de-identified clinical records.

Statistically significant reductions in error rates were observed across four key clinical areas: history-taking, investigations, diagnosis, and treatment.

The study also identified a “training effect.” Over time, clinicians in the AI group showed improvements even beyond the tool’s direct feedback. For instance, the proportion of red alerts for unsafe treatment plans declined by 10–15 percentage points, a trend not seen in the control group. This suggests that well-integrated CDS tools may contribute to ongoing clinical learning.

“AI Consult demonstrates that AI-supported care is feasible in everyday clinical settings,” said Dr. Sarah Kiptinness, Medical Director at Penda Health.

The study highlights that successful AI deployment in healthcare requires more than advanced technology—it also depends on integration into existing workflows, alignment with clinical objectives, and the trust and engagement of clinicians.

For healthcare systems globally, these findings provide practical evidence that, when responsibly implemented, AI tools can help improve patient safety and the quality of care, even in resource-limited environments.