An oral weight-loss pill, orforglipron, lowered study participants' 10-year predicted risk of type 2 diabetes by roughly half and nudged down their predicted heart-disease risk over 72 weeks. Predicted is the word doing the work. Nobody in the analysis actually developed diabetes or had a heart attack that got counted. The numbers come out of three risk calculators run on the trial's own data.
The post hoc analysis ↗, published September 17 in Diabetes, Obesity and Metabolism, reworked results from ATTAIN-1, a Phase 3 trial of orforglipron in 3,127 adults who were overweight or obese but did not have diabetes. Orforglipron is Eli Lilly's once-daily oral GLP-1 drug, a small molecule that hits the same receptor as the injected peptides semaglutide ↗ and tirzepatide ↗. In ATTAIN-1 the top dose produced about 12 percent weight loss, roughly 27 pounds, at 72 weeks, the headline Lilly reported this year ↗.
This new paper asks a narrower question. If you feed those 72 weeks of change into validated risk models, how far does the projected long-term risk move? The authors used three engines. Cardiometabolic Disease Staging estimated diabetes risk. Framingham and PREVENT, the American Heart Association's newer calculator, estimated cardiovascular risk. Each takes measured inputs like weight, blood sugar, blood pressure, and cholesterol and returns a 10-year probability.
The projected diabetes risk fell 48.6 to 59.4 percent across the three orforglipron doses, against a 10.5 percent drop on placebo (p less than 0.0001). Predicted cardiovascular risk fell too, though by far less: Framingham total-CVD risk dropped 0.6 to 1.0 percentage points while placebo rose 0.6. PREVENT showed lower projected risk of heart failure and atherosclerotic disease on the drug.
Here is the catch built into the method. A risk engine is a formula. Orforglipron moves weight, glucose, blood pressure, and lipids, and those are exactly the numbers the formulas read. A drug that improves the inputs will, almost by construction, lower the output. The analysis confirms the models behave as designed. It does not show that a single person was spared a diagnosis. A predicted 59 percent risk reduction and an observed one are different claims, and only the second is an outcome.
Two more things temper the read. This is a post hoc analysis, planned after the trial, not the endpoint ATTAIN-1 was powered to test. And the work was run largely by the manufacturer: most of the authors are Eli Lilly employees, analyzing a Lilly drug with models the company selected rather than built. None of that makes the arithmetic wrong. It makes the framing a marketing-adjacent one, that the drug prevents disease, when what the data show is that it improves the surrogates disease models are built from.
What would settle it is a hard-outcome trial that counts real diabetes diagnoses and real cardiovascular events over years, not weeks. Those take longer and cost more, and for orforglipron they are not yet in hand. Until then, "cuts predicted risk" is a fair thing to say and "cuts risk" is not.
The distinction reaches past one pill. Peptidemodel hosts cards for the injectable GLP-1 receptor ↗ agonists orforglipron is chasing, and much of the enthusiasm around the whole class rests on surrogate endpoints, weight and A1c, that stand in for outcomes still being measured. A modeled risk score is one more surrogate. Useful for sizing a bet, not for closing it.