In one of the first real-world tests of artificial intelligence in health care, an AI tool deployed at primary care clinics in Nairobi helped clinicians cut diagnostic errors by 16% and treatment errors by 13%, according to a study by OpenAI and Penda Health.
The study, based on 20,000 patient visits across 16 Penda Health clinics, offers rare, real-world evidence that AI can improve health outcomes—not just in simulated environments, but during actual patient interactions.
“It’s like having an expert in the room,” said Dr. Robert Korom, chief medical officer at Penda Health. “AI Consult doesn’t dictate decisions, but it acts as a safety net—flagging questionable steps like antibiotic overuse.”
The tool, AI Consult, was designed to shadow clinicians silently, intervening only when their decisions diverged from best-practice guidelines. The AI taps into a vast database of local and national clinical protocols and provides real-time alerts using a green-yellow-red system.
Built for Local Context
Unlike many global AI solutions, AI Consult is fine-tuned for Kenya’s health realities—like malaria variants and common symptom patterns. That specificity made it a valuable assistant to Penda’s mid-level clinicians, who often operate with limited time and broad case variety.
Dr. Sarah Kiptinness, head of medical services at Penda, said the tool has “redefined our standard of care.” Clinicians now rely on AI Consult not just for quality assurance, but also for continuous professional learning. “It builds confidence and speeds up mastery,” she said.
Beyond Error Reduction
OpenAI’s Health AI lead, Karan Singhal, said he was surprised to see the tool double as an educational aid. “It wasn’t just about catching mistakes—it expanded clinicians’ understanding of care pathways,” he said.
The findings were validated by external experts, including Harvard’s Dr. Isaac Kohane and Stanford’s Dr. Ethan Goh, who praised the study as a rare, prospective look at AI’s tangible impact in medicine. Goh noted that as more such studies emerge, AI’s role may shift “from mistake correction to care plan optimization.”
What’s Next
AI Consult is now embedded in Penda’s operations, and further iterations will offer personalized training recommendations based on clinician performance patterns.
Penda’s long-term vision? Use AI to extend quality care beyond clinicians—to social workers, community health volunteers, and remote areas with limited medical personnel.
“This isn’t theoretical anymore,” Korom said. “We wouldn’t want our clinicians practicing without this tool.”
As global health systems strain to balance quality and scale, Penda’s pilot offers a promising blueprint for AI-assisted care—especially in resource-constrained settings.
