Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Friday, November 13th, 2026. Voice 1: The American Cancer Society just tore up 16 years of shared decision-making on prostate cancer. As of Thursday, the recommendation is a flat one: a PSA test every two years for average-risk men 50 to 69. Voice 2: No more conversation at all, or just a different kind of one? Voice 1: They still want men to understand the benefits, limits and harms first. But once they do, the guideline tells doctors to screen them every two years, not just offer the option. Voice 2: What about men at higher risk, Black men, people with a family history? Voice 1: Black men, and men with a first-degree relative diagnosed before 65 or several affected relatives, get screened every year starting at 45. One relative diagnosed later than that, it's every two years from 45. Voice 2: And once a PSA actually comes back high, what happens next? Voice 1: An elevated result is 3 ng/mL or higher from 45 to 69, 6.5 ng/mL or higher from 70 to 74. The guideline calls for a repeat test two weeks to three months later, then an MRI or a validated biomarker before anyone gets a biopsy. Low-risk cancer gets active surveillance as the first option. Voice 2: Did they say why now, after 16 years of leaving it to the patient? Voice 1: The society points to stronger evidence that screening lowers prostate cancer deaths and advanced-stage diagnoses. The guideline was led by Dr. Richard Hoffman and Dr. Robert Smith, in CA: A Cancer Journal for Clinicians. Voice 2: A different kind of diagnosis shift showed up in a physician survey this week. Half of U.S. physicians now say they use AI every day at work. Voice 1: That's up from what? Voice 2: Up from 37% a year ago, in a survey of more than 3,300 physicians. And 85% now use it at least weekly, up from 51%. Voice 1: But are they catching it making mistakes? Voice 2: About three in four reported at least one factual inaccuracy from AI in the past year. Two-thirds say they verify AI-generated clinical information nearly every time. Voice 1: Two-thirds checking it, after it's wrong that often, still leaves a gap. How many of their organizations actually have rules for this? Voice 2: Just 28% said their organization has clear policies and approvals for AI use. Voice 1: Which specialty's leaning on it hardest? Voice 2: Oncology. More than 93% use AI at least weekly, the highest of any specialty, and 46% of everyone surveyed said it's cut their after-hours administrative work. Voice 1: And patients are showing up with their own AI opinions too? Voice 2: About three in four physicians said patients are more likely than six months ago to arrive with a self-diagnosis or treatment plan from a generative AI tool. This is Doximity's own survey, and the company sells its own AI tools. Healthcare Dive reported separately that use for clinical decision support jumped to 63% from 37%. Voice 1: A blood pressure trial out this week undercuts a popular idea: that pushing systolic way down protects the aging brain too. Voice 2: Pushing it down to what, exactly? Voice 1: Under 120 mm Hg instead of under 140, in a secondary analysis of the BPROAD trial, in adults with type 2 diabetes. After a year, the intensive group averaged about 122, the standard group about 133. Voice 2: So tighter control, but no payoff for dementia? Voice 1: None. Over nearly four years, the combined rate of dementia or mild cognitive impairment came out nearly identical between the groups, a hazard ratio of 0.96. Voice 2: Any upside at all buried in there? Voice 1: In an MRI substudy of 862 participants, the intensive group did accrue less white matter lesion volume. It just didn't translate into less dementia or impairment in the main analysis. Voice 2: Twelve thousand four hundred and thirty four people, and the brain just didn't follow the blood pressure number down. Voice 1: On a smaller scale, MDCalc tried something different this week: a filter that strips out patient details before a question ever reaches an AI chatbot. Voice 2: So it works with whatever chatbot a doctor's already using? Voice 1: Right. CrossCover removes patient, provider and other identifying information from the query itself, and it doesn't require registering with a national ID number or signing an agreement to share protected health information. Voice 2: Why build that, with tools like OpenEvidence or Epic's AI already in the mix? Voice 1: Those tools are expanding the patient, provider and payer information they use, not pulling back from it. MDCalc's head of AI, Dr. Graham Walker, said, “We can take away some of the challenges and risks and problems with the technology.” Voice 2: A privacy layer instead of a new chatbot is a pretty different bet on where the risk actually lives. Voice 1: Before we go, something to watch. The Senate health committee votes November 18th on Heidi Overton, the nominee for FDA commissioner. Voice 2: Does a committee vote actually put her in the job? Voice 1: No. Approval there just sends the nomination to the full Senate for a final vote. Voice 2: A committee yes is still just the first step toward whoever ends up running the agency. Voice 1: Sources and corrections are at physicianintheloop.org, where you can sign up for the free email briefing. This briefing was read by AI voices.