Subscribe
Share
Share
Embed
Stents or bypass for stable heart disease add no survival at 11 years
Stents or bypass for stable heart disease add no survival at 11 years
The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.
Routine angiography and revascularization for stable coronary disease with moderate or severe ischemia did not extend survival over 11 years compared with medicines alone, according to ISCHEMIA follow-up data JAMA published Saturday.
The same day at the American Heart Association's Scientific Sessions in Chicago, full results showed that pelacarsen, which cut lipoprotein(a) by 78%, and the interleukin-6 antibody ziltivekimab, which lowered CRP, did not reduce major cardiovascular events. A randomized trial of AI-screened ECGs found a statistically fragile gain in atrial fibrillation detection, and a Stanford study using a language model found lipid therapy was left unchanged at 89% of visits where LDL was above goal.
Weill Cornell Medicine disclosed that patient imaging details were entered into AI tools without authorization for a radiology research study in June.
In this episode
How technological innovation is changing medicine for patients and the people who practice on the front lines.
Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Sunday, November 8th, 2026.
Voice 1: Eleven years out, stents or bypass for stable heart disease add no survival at all over medicines alone. That's the headline from the ISCHEMIA follow-up, published Saturday in JAMA.
Voice 2: Eleven years is a long follow-up for a heart trial. What were the actual numbers?
Voice 1: All-cause mortality was about 26% with angiography plus revascularization, versus about 27% for patients started on guideline-directed medical therapy alone. The adjusted hazard ratio was 0.97, essentially a wash.
Voice 2: So the invasive approach isn't buying people more years. Did cardiovascular deaths split any differently?
Voice 1: A little. Cardiovascular death ran roughly 16% with the invasive strategy against about 17% with medicines alone, but noncardiovascular deaths were close to even too. Dr. Judith Hochman of NYU Grossman School of Medicine led the investigators on this.
Voice 2: This was a big trial, right? How many patients are we talking about?
Voice 1: 5,179 patients, all with moderate or severe ischemia, no left main disease and no recent acute coronary syndrome, with median follow-up reaching 9 years.
Voice 2: And the statisticians ran their own gut check on all this?
Voice 1: They did. A bayesian analysis put the odds that the invasive strategy improves survival by more than a single percentage point at just 18%. The 11-year survival difference between the two groups came down to 14 days.
Voice 2: Fourteen days, over eleven years. That's about as close to nothing as a trial gets.
Voice 1: It's the kind of number that reshapes a conversation with a stable patient weighing a procedure against staying on pills.
Voice 2: Two heart drugs had their own rough Saturday at the same meeting in Chicago.
Voice 1: Pelacarsen cut lipoprotein(a) by 78% in more than 8,000 patients with established cardiovascular disease. And it still didn't prevent heart attacks or strokes.
Voice 2: Cutting a risk marker that much and getting nothing for it. What did the event rates actually look like?
Voice 1: About 12% of patients had a primary event on pelacarsen, versus about 12% on placebo, over a median of roughly 5 years. Discontinuation for side effects was far higher on the drug too, about 21% versus 8% on placebo.
Voice 2: Was there any subgroup where it helped, even a little?
Voice 1: No. Extended analyses found no benefit in any subgroup or by how much Lp(a) actually came down. Dr. Steven Nissen of Cleveland Clinic, the senior author, said the results were, quote, “completely neutral across multiple analyses.”
Voice 2: And the other drug, the one going after inflammation instead?
Voice 1: Ziltivekimab, an antibody against interleukin-6, lowered CRP in more than 6,000 patients with atherosclerotic disease, kidney disease and inflammation. Three-point major cardiovascular events still landed at about 18% in both the drug and placebo groups, over roughly 33 months. Dr. Paul Ridker led that report.
Voice 2: Two completely different mechanisms, same result. That's a rough day for the inflammation and Lp(a) theories both.
Voice 1: A smaller study at the same meeting tried something different: letting an AI model flag who needs heart rhythm monitoring in the first place.
Voice 2: How'd that work?
Voice 1: Researchers at National Defense Medical University in Taipei had an AI model screen old sinus-rhythm ECGs, then offered Holter monitoring to whoever it flagged as high risk. New atrial fibrillation, flutter, or pre-AF findings turned up in about three-quarters of one percent of the AI group within 90 days, against under half a percent with usual care.
Voice 2: That sounds like a real gain. Is there a catch?
Voice 1: A big one. The authors themselves called the result “fragile statistically.” Reclassifying just a single event would have pushed the result out of statistical significance entirely. It was a single-center, open-label trial of 14,665 adults, and detection of AF or flutter alone didn't differ between groups, nor did starting antiarrhythmic therapy.
Voice 2: So the headline number barely survives contact with one reclassified patient.
Voice 1: That's the authors' own word for it, and it's a fair one.
Voice 2: Stanford had a less flashy but maybe more useful finding: they used a language model just to read what doctors actually wrote in the chart.
Voice 1: What did it find?
Voice 2: Among more than 13,000 patients with atherosclerotic disease and LDL cholesterol above goal, lipid therapy went un-escalated at about 89% of the most recent visits. Only about 40% of those notes even gave a reason why.
Voice 1: What were the reasons doctors did write down?
Voice 2: Patient-reported side effects led, at about 13%. After that: LDL close to goal, patient preference to avoid therapy, preference for lifestyle changes first, and waiting on repeat labs, all in the mid single digits each. A treat-to-target strategy showed up in about 28% of notes.
Voice 1: Can we trust a language model's read on all that, though?
Voice 2: The researchers checked. The model's agreement with human clinical annotators didn't differ statistically from how much the human annotators agreed with each other, with F1 scores between 0.80 and about 0.98. It drew on a large health system in Northern California, from 2019 through May of this year.
Voice 1: So the model isn't the weak link here. The gap between LDL above goal and anyone actually doing something about it is.
Voice 2: Weill Cornell Medicine has an AI story of a different kind this week.
Voice 1: They disclosed that patient imaging details, medical record numbers, diagnoses, procedure histories and medications, were entered into AI tools without authorization, for a radiology research study in June.
Voice 2: Did anyone outside the research team actually see any of it?
Voice 1: The organization says it found no evidence anyone unauthorized viewed the information, and that names, addresses, Social Security numbers and financial details weren't involved. It learned of the incident in September, has suspended the research activity, and asked for the data to be deleted.
Voice 2: Do we know which AI tools, or how many patients were caught up in this?
Voice 1: The notice didn't say either of those things.
Voice 2: A research study moving faster than its own authorization paperwork is a familiar shape for one of these.
Voice 1: It is becoming one.
Voice 2: Between the ISCHEMIA numbers and two drugs that moved their targets but not outcomes, it's a day that argues for sticking with what's already proven rather than chasing the next mechanism.
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.