Version 1 of this summary, as approved on . It weighed 96 gains, 98 problems and 87 Traces. A newer version has replaced it. See the topic as it stands today
Most of the evidence comes from chatbots answering written test questions and from small studies at a single site; trials in real patients are rare. TRV-2026-0355TRV-2026-1613 A scoping review of reports on AI decision tools for clinicians found that only 14.4% evaluated implementation or impact. TRV-2026-1613
A randomized crossover trial shows clinical trial staff using AI spent 2.71 fewer minutes per item reviewing charts for immunotherapy side effects, though severity grading was much weaker than event detection. TRV-2026-2318 In randomized trials with over 90,000 US patients, messages about personal high flu risk raised vaccination 1.7-3.5 percentage points over no message. TRV-2026-1852 A randomized preclinical trial of 77 surgeons found AI assistance cut the time to find a bleeding site in laparoscopic video to 2.53 seconds from 3.23, with no test on live operations. TRV-2026-2032 An observational study of 12,468 chronic disease patients in three Chinese provinces linked AI-enabled care to hypertension control targets met rising from 68% to 82%. TRV-2026-0328
A randomized trial of 7294 patients at Mount Sinai found an AI-triggered alert did not reduce delirium (35.1% vs 32.7%), though only 17.2% had a documented delirium assessment. TRV-2026-1664 In a 960-response test with clinician-written cases, ChatGPT Health undertriaged 52% of true emergencies, sending diabetic ketoacidosis and impending respiratory failure to 24-48 hour evaluation. TRV-2026-0181 Researchers testing 21 chatbots on standardized vignettes found all of them failed differential diagnosis at rates above 0.80, even though final-diagnosis accuracy was high. TRV-2026-0145 Tests of eight speech-to-text systems found clinically significant errors in 30% to 68% of 800 transcripts; AI-simplified children's discharge notes carried 12 safety-critical errors in a 15-document test. TRV-2026-1687TRV-2026-1916
A scan of AI studies built on US health record networks found only a small subset was tested prospectively or used in clinical workflows. TRV-2026-0187 A scoping review found only seven eligible studies of agentic AI in healthcare, with just one trial involving patients. TRV-2026-0166 A review of 310 publications on medical education found no study measured AI's effect on critical thinking or clinical reasoning. TRV-2026-0458 A review of FDA-cleared surgical AI devices found only 5.1% had higher-quality prospective evidence. TRV-2026-1693
On AI scribes and burnout, an observational study of 263 clinicians in six US health systems found self-reported burnout fell from 51.9% to 38.8% after 30 days. TRV-2026-0478 A systematic review of six real-world studies found burnout on standardized scales was unaffected, and clinicians' notes got longer. TRV-2026-0330 At one cancer center, only 190 of 958 clinicians used the AI scribe, and after-hours time fell just from 86.9 to 81.4 minutes a week. TRV-2026-1819
Specialists mostly rated chatbot answers accurate in test questions, yet readability stayed above recommended levels and every chatbot in one nurse-rated test gave some unsafe answers. TRV-2026-1069TRV-2026-1746
ChatGPT-5 triage matched emergency experts more often than nurse triage did (83.5% vs 73.3%) in a retrospective study, yet ChatGPT Health undertriaged 52% of true emergencies in a separate test. TRV-2026-1421TRV-2026-0181
Clinicians reported less documentation time with AI scribes, but a review found burnout on standardized scales unchanged and one cancer center saw only 19.8% of clinicians adopt the tool. TRV-2026-0330TRV-2026-1819
A randomized trial and a matched comparison tie AI-based practice to higher student scores, while a literature review warns doctors' exam and diagnosis skills may erode.
Surgical and imaging tools were mostly tested on small retrospective samples or in simulation; a review of FDA-cleared surgical AI devices found only 5.1% had prospective evidence. TRV-2026-1693TRV-2026-2109TRV-2026-1623
A survey found 43.0% of US adults uncomfortable with their provider using AI, and freedom of information responses show trusts listed as benefiting from Palantir's platform say it is not live or unused. TRV-2026-1843TRV-2026-2335
Each record sits in one thread at most. A record with both a gain and a problem counts on both sides. 34 records fit none of these threads.
ace