Researchers evaluated an artificial intelligence system called AI Consult in a trial involving nearly 10,000 patient encounters across 16 primary care clinics in Kenya operated by Penda Health. The tool, powered by OpenAI’s GPT-4o language model, reviewed clinical notes taken by health workers and provided feedback using a traffic light system: green for satisfactory notes, yellow for minor issues with guidance, and red for critical concerns requiring prompt attention.
Half of the clinical officers in the trial used the AI tool while taking computer-based notes, while the control group used note-taking software without AI assistance. An independent panel of six Kenyan family physicians evaluated the quality of clinical documentation and treatment plans. These expert reviewers determined that clinicians using the AI tool produced better diagnoses and treatment recommendations. The system operated at minimal cost, functioning at approximately 4 cents per patient encounter.
Despite improvements in clinical documentation quality, the trial did not demonstrate statistically significant improvements in patient outcomes. The study found a 23 percent reduction in treatment failures such as death or unresolved symptoms, but this difference was not statistically significant due to the low baseline frequency of such cases. Researchers estimated that a trial with approximately 139,000 participants would be needed to detect meaningful outcome differences.
Clinical officers reported finding the AI feedback valuable, with one health worker describing it as a helpful “second pair of eyes” that reminded them of established protocols and current guidelines. In settings where clinicians manage five to six patients per hour without specialist backup, the tool was viewed as adding practical value. The expert panel determined that most AI recommendations were safe and appropriate.
Researchers and medical professionals see potential for broader applications, including the possibility that AI could generate draft notes for clinician review, potentially freeing time to see more patients. Concerns remain about oversight and the risks posed by AI systems that may make errors or provide inappropriate guidance, according to healthcare AI researchers.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI