Here is a number that should unsettle the tidy story you have been told about semaglutide. In the FLOW trial, funded like the other two trials in this piece by Novo Nordisk, the manufacturer of semaglutide, the people taking the drug lost an average of 5.5 kilograms (about 12 pounds) over 3.4 years, against 1.45 kilograms (about 3 pounds) on placebo. That is a modest difference, nothing like the near-15 percent body weight loss reported in the drug's obesity trials. And yet their risk of a major kidney event, kidney failure, a sustained halving of kidney function, or death from cardiovascular or kidney causes, fell by 24 percent relative to placebo (hazard ratio 0.76, 95 percent confidence interval 0.66 to 0.88), according to the trial's own report in The New England Journal of Medicine. In a mediation analysis presented at last year's American Diabetes Association meeting and reported by Healio, investigators asked how much of that kidney benefit the weight loss actually explained; the answer came back at roughly zero, an estimated negative 0.5 percent mediation effect, which is to say none to speak of. A FLOW steering-committee investigator, Johannes Mann, made the same point in remarks Healio reported from that meeting: gauging whether the drug is working does not require tracking a patient's weight, because the benefit does not depend on it.
That finding matters more than the covers of most magazines let on, because the plausible, comfortable explanation for everything semaglutide seems to do, lower blood pressure a little, ease strain on the heart, protect the kidneys, has always been that it makes people thinner, and thinner people do better. FLOW says that story does not hold, at least for the kidney.
The number needed to unpack first, though, is the 24 percent, because a relative risk reduction tells you shape, not size. The primary kidney outcome occurred in 331 of 1,767 people on semaglutide (18.7 percent) against 410 of 1,766 on placebo (23.2 percent), in adults with type 2 diabetes and chronic kidney disease who were required, as a condition of enrollment, to already be taking a renin-angiotensin-system inhibitor, a blood-pressure drug that is standard care for diabetic kidney disease with protein in the urine at any stage, not only advanced disease, and that itself has a modest kidney-protective effect on its own. Set side by side like that, the absolute difference is 4.5 percentage points, which means roughly 22 people would need to take semaglutide for the trial's median 3.4 years to keep one of them off dialysis or spare one death. All-cause death was 20 percent lower (hazard ratio 0.80, 0.67 to 0.95), cardiovascular death 29 percent lower (hazard ratio 0.71, 0.56 to 0.89), and death from causes nobody could pin down fell 38 percent (hazard ratio 0.62, 0.42 to 0.91), in a later cause-specific breakdown presented at the American Society of Nephrology's Kidney Week 2024. These are real, hard-to-fake outcomes: dialysis avoided, deaths postponed, not a scale reading.
The cardiovascular data followed a similar shape, twice over. SOUL, published last year and also funded by Novo Nordisk, put 9,650 adults with type 2 diabetes and established heart or kidney disease on an oral form of semaglutide, the pill rather than the weekly injection, at a dose capped for glycemic control rather than weight loss. Major cardiovascular events occurred in 12.0 percent of the drug group against 13.8 percent on placebo, a 14 percent relative reduction (hazard ratio 0.86, 95 percent confidence interval 0.77 to 0.96) and a 1.8 percentage point absolute one, over a median follow-up of about four years. That absolute gap puts the number needed to treat at roughly 56 people, treated for about four years, to prevent one cardiovascular event. Mean weight loss on the pill was 4.22 kilograms (about 9 pounds), barely more than FLOW's, which again leaves little room for weight to be doing the heavy lifting.
SELECT, the largest of the three trials, is the trial that keeps this from being a tidy story. According to the trial's own NEJM report, it enrolled 17,604 people with obesity and existing cardiovascular disease but no diabetes, a population that skewed toward men (72 percent) and White participants (84 percent), narrower than the population that will actually end up prescribed this drug, and important to keep in mind since SELECT does the most work below. Everyone was on the full 2.4 milligram weight-management dose, and here the weight loss was substantial: 9.4 percent of body weight against 0.9 percent on placebo. Cardiovascular events fell from 8.0 percent to 6.5 percent, a 20 percent relative reduction (hazard ratio 0.80, 95 percent confidence interval 0.72 to 0.90). What that 1.5 percentage point absolute difference means for how many people need treating depends, awkwardly, on how you count. A simple calculation from the trial's own raw event rates puts it around 67 people treated for the trial's mean 3.3 years to prevent one event, while a reanalysis by Ferreira and colleagues, reported by TCTMD, using annualized event rates instead, arrived at roughly 200. The same TCTMD reporting notes that SELECT's own investigators declined to state a number needed to treat at all, calling the exercise arbitrary depending on which time frame one chooses, and both of those competing figures are defensible, which is exactly the trouble with the measure. When the trial's own investigators went looking for what mediated the cardiovascular benefit, they did not find weight loss itself. In a prespecified analysis published in The Lancet in October 2025, they found that the drop in waist circumference, not the number on the scale, accounted for an estimated 33 percent of the effect, while body weight loss itself showed no straight-line relationship to who benefited. Something else, still unnamed, is doing the other two-thirds of the work.
It is worth being precise about what "still contested" means here: it is not a hedge reached for out of caution, it is the honest state of the evidence. FLOW's kidney benefit and SELECT's cardiovascular benefit are not the same benefit, in the same organ, in the same population, mediated by the same missing variable. What exists is two separate trials, each showing that the crude story, weight down, risk down, does not fit its own data, without agreeing on what does.
The newest piece extends the pattern to people with type 1 diabetes, who have been excluded from every randomized trial above on hypoglycemia concerns. It is not a fourth trial. It is a March Nature Medicine analysis, funded by the National Institutes of Health rather than by any drug maker, with its authors stating they have no competing interests, built from health records for 174,678 such people, using a technique called target trial emulation that tries to approximate a randomized comparison from data that already exists by splitting each person's records into repeated monthly comparisons, 6,092,537 of them in total, of which 14,488 showed a GLP-1 drug being started that month; that count is of monthly comparisons, not unique people, since one person's records could contribute to many of them. After that statistical adjustment, five-year cardiovascular event risk was 4.3 percent among GLP-1 drug users against 5.0 percent among those who did not use one, a 14 to 15 percent relative reduction (hazard ratio 0.85, 95 percent confidence interval 0.77 to 0.95), and five-year risk of end-stage kidney disease was 1.6 percent against 1.9 percent, a 19 percent relative reduction (hazard ratio 0.81, 0.69 to 0.95). A separate, secondary figure in the same analysis, an 18 percent lower rate of hospitalization for heart failure, gets conflated with the topline cardiovascular result in some coverage; it measures a narrower outcome and should be read on its own. None of this came with more hospitalizations for dangerously low blood sugar. What this is, and is not, deserves to be said plainly. It is not a randomized trial. It is a statistical approximation of one, built from prescriptions and outcomes that already happened, and its own authors say a real trial is still needed before anyone should treat it as settled. The numbers are consistent with what the type 2 diabetes trials found. Consistent is not proven.
What a reader managing kidney disease or a family managing a child's type 1 diagnosis should take from this is narrower than the headlines suggest: a drug built to treat blood sugar and appetite, and sold by a company with an obvious stake in that being true, appears to be protecting organs through some route besides the scale, in at least two different ways, and nobody has yet named either one.
