Easy Clinic has successfully rolled out an AI-powered copilot that helps doctors make better clinical decisions in real time, which is a major breakthrough for African health-tech.
Tested across 15 Penda Health clinics and nearly 40,000 patient visits, the AI tool reduced diagnostic errors by 16%.
Integrated directly into the electronic health record (EHR) used in its clinics, AI Consult operates in real time to flag potential clinical issues before they affect patient care.
Unlike earlier versions that required manual input from clinicians, the latest AI Consult runs automatically in the background during every patient visit. As clinicians document the visit, non-identifiable patient information is sent to OpenAI’s API at specific points.
AI Consult analyzes the interaction and returns one of three feedback signals:
- Green: No concerns. Displayed as a green checkmark.
- Yellow: Moderate concerns. Displayed as a bell; the clinician may review.
- Red: Safety-critical issues. Displayed as a mandatory pop-up that must be acknowledged before proceeding.
This alert system is designed to surface only relevant issues, reducing the cognitive burden on clinicians while maintaining safety.
Clinician Still Retains Full Decision Authority
AI Consult does not act autonomously. It identifies possible issues for the clinician to review, but final decisions remain with the clinician. The tool is designed as a support system, not a replacement.
Clinicians may choose to accept, ignore, or investigate flagged feedback. This structure ensures clinical judgment is preserved.
To ensure relevance and accuracy, AI Consult has been customized to align with:
- Kenya-specific epidemiology
- Local clinical treatment guidelines
- Standard procedures followed in Penda clinics
This localized approach increases accuracy and relevance, avoiding mismatches that may occur in systems trained on global data alone.
An internal audit reviewed 100 real outputs from patient encounters using AI Consult. Some of the findings were:
- In many cases, the AI’s output was clinically useful.
- No instances of harmful or misleading feedback were identified.
Earlier versions of AI Consult required clinicians to actively request assistance. Despite accurate suggestions, uptake was low due to workflow interruptions. The current passive model, embedded into daily operations, has improved adoption.
Penda Health leverages OpenAI’s large language models (LLMs) to power AI Consult. By tapping into the capabilities of LLMs, AI Consult can support a broader range of conditions and potential errors compared to traditional rule-based systems.
