AI decision support sharpens primary care notes, treatment plans

Clinical takeaway: AI support can improve the quality of clinical documentation and treatment planning without disrupting workflow or patient trust, but don't count on any short-term gains in patient outcomes.

Most evidence that clinical AI improves care comes from simulated cases, not patients seen in real clinics. This trial tested a generative AI tool where it counts: embedded in routine primary care visits across 16 primary care clinics, with patient outcomes as the primary endpoint. The tool effectively aided clinician notes and treatment decisions, but patient outcomes didn't follow.

Clinicians using the tool recorded better documentation across the board. In a blinded review of 2,000 visits, they were about 70% more likely to log an appropriate diagnosis, a comprehensive note, and a sound treatment plan. All three gains were statistically significant. Safety held up too: among 1,000 high-severity alerts that the tool generated, more than 90% were judged safe and appropriate on expert review, and no serious adverse event was linked to the tool.

Patient outcomes didn't budge, though. Treatment failure within 14 days was 2.2% in the AI-supported group and 2.0% in usual care, a difference that wasn't significant. Hospitalizations and deaths were similar in both arms.

The model ran in the background during a patient visit, reading each note as it was typed and flagging any diagnostic or treatment concerns against national guidelines in real time. Clinicians chose what to adopt and the tool was never visible to patients. On blinded review, that translated into more complete notes, more appropriate diagnoses, and sounder treatment plans.

This pragmatic, cluster-randomized trial enrolled 9,691 patients across 16 Penda Health primary care clinics in Nairobi and Kiambu, Kenya, from April to July 2025. The 103 clinical officers, mid-level practitioners who deliver much of Kenya's primary care, were randomized to use an electronic medical record with or without an embedded LLM-based tool that read each note in the background and flagged diagnostic or treatment concerns with a green-yellow-red alert. The primary outcome was an expert-adjudicated composite of treatment failure within 14 days.

The open question is not whether AI can sharpen frontline decisions, but whether that can positively impact patients and how to measure it. Detecting modest outcome gains in a setting like this would take more than 100,000 patients, so future trials may need composite or process endpoints rather than clinical events.

"The technology appears safe and clearly improves aspects of clinical decision-making, but translating those gains into measurable patient benefit is much more challenging, particularly in everyday primary care," said Bilal Mateen, MD, chief AI officer at PATH and honorary professor of machine learning for health at the University of Birmingham.

Source: Agweyu A, et al. Nat Med. 2026 Jun 26. Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial