Family Medicine

Systematic map: LLMs show promise for psychological outcomes but variable diagnostic accuracy

This evidence map synthesized 55 studies (Jan 2022–Jun 2025) of LLM deployments in clinical settings. Most interventions were human–AI collaborative (65.5%) and LLM-assisted care showed positive psychological health effects. Diagnostic accuracy in randomized trials was variable with reported ranges of 0.65–0.88, and nonrandomized studies typically reported ≥0.80. Reporting quality was suboptimal with mean CONSORT‑AI adherence 78.8%, indicating gaps in describing data quality and error handling. 1

RCT: Stroke Riskometer app did not improve Life's Simple 7 at 6 months

In a two‑arm RCT of 862 adults aged 35–75 with ≥2 stroke risk factors, app access did not change Life's Simple 7 at 6 months (mean difference 0.03; 95% CI -0.19 to 0.25; P=0.788). Per‑protocol analyses were similar and no individual Life's Simple 7 components improved meaningfully. There was a nonsignificant increase in physical activity (MET‑minutes/week difference ~313; P=0.052). In primary prevention, giving app access alone did not change global risk-factor burden over 6 months. 2

CDSMP meta-analysis: small sustained reductions in pain, depression, fatigue, and sleep problems

This meta‑analysis included 20 studies with ≥6‑month follow‑up assessing the Chronic Disease Self‑Management Program (CDSMP). Mean improvements at 6 months were modest: pain 0.53, depression 1.47 (PHQ‑9 units), fatigue 0.46, and sleep problems 0.57 on reported scales. Benefits were consistent across countries and among participants with multimorbidity. Implementation challenges included funding, recruitment, and retention, limiting broad scalability. 3