SCIENCE UNDER AUTOMATION
Generative AI decision support in primary care
A cluster trial at 16 Kenyan facilities enrolled 9,691 patients. The primary 14-day treatment-failure outcome did not differ significantly between assisted and control groups.
What this result supports
Clinical officers delivered care in both groups; the trial evaluated assistance, not autonomous clinical practice.
Evidence record: Peer-reviewed pragmatic randomized trial. Checked: 2026-10-21.
Where this evidence is used
Clinical medicine
- Integrate examinations and clinical context: Virtual cases do not supply autonomous physical examinations; real clinical officers retain responsibility in the trial.
- Verify the plan before acting on it: Guideline grounding supports recommendations, while clinicians execute and check patient care.
- Execute treatment safely in a real patient: Ex-vivo robotic steps are demonstrated, while the real-patient trial retains clinical officers for treatment.
- Recognize complications and adapt care: Multivisit reasoning has been simulated; responsibility for real complications remains with clinicians.
- Verify sustained clinical benefit: The randomized trial measured outcomes under clinical supervision and found no significant primary-outcome benefit.
Population health
- Define a causal comparison and recruit participants: The field trial used a researcher-designed randomization and enrolled patients through clinical facilities.
- Implement the intervention across settings: People and institutions delivered the intervention; forecasting systems supply predictions.
- Interpret outcomes against bias and competing explanations: Statistical and predictive tools assist analysis, but forecasts alone do not identify intervention effects.