Prepared end to end by AI. The editorial is Dr. Hemant L. Patel. This is a newspaper-style weekly, not a numbered insight series. The PDF is the issue.
The strongest items this week cut in opposite directions. A generalist diagnostic AI posted its best benchmark yet. The week’s best oversight evidence found the odds of a correct clinical decision about ten times higher when automated advice was correct than when it was wrong.
In this issue
- Microsoft’s MAI-DxO, in Science: 83% of NEJM clinicopathological cases against 36% for experienced physicians — written puzzles, not patients.
- npj Digital Medicine: odds of a correct clinician decision about ten times higher with correct than incorrect automated advice (OR 10.19); 26.6% of erroneous ECG interpretations never corrected.
- FDA generative-AI device docket closed 19 October with more than 500 comments; draft guidance promised within months.
- Google’s AI-guided gestational-age assessment from novice “blind sweep” ultrasounds held up across new cohorts (JAMA Network Open).
- Johns Hopkins Bayview alerted 3,182 patients after a staffer entered their data into an AI tool outside governance.
- India: a public register of AI health systems is still a proposal; ICMR picked seven institutions with ₹105 crore for indigenous devices.
“The safeguard we cite most is the one we measure least.”
AI Health Weekly · Vol. I · No. 3 · 5 pages · Dr. Hemant L. Patel
If the viewer is blank on this device, download the PDF.
Sunday, 1 November 2026
The week patients met the model directly
The front door of medicine has moved. The rulebook has not.
Read No. 2 →The week outcomes stopped being a report and became the deal
The WHO documented a CSR-backed eye-care programme as a national model, a state tranche releases only when 600 rural youth land jobs, and a skilling centre announced offers at its inauguration.
Read Impact Weekly →Authored data series live on Insights.