Modeling studies put it in the millions over a decade, and none of those figures was observed in a patient. One model covers US adults who already have cardiovascular disease. It projects semaglutide 2.4 mg preventing between 496,400 and 1,934,493 major cardiovascular events over 10 years, depending on how eligibility is drawn [2]. Those events are a composite of heart attack, stroke and cardiovascular death. Whether the drug prevents each of those separately is a different question, taken up in whether GLP-1 drugs prevent heart attacks and strokes.
Every model here takes a trial result and multiplies it across survey records. Nobody was treated. The answer therefore depends on three choices: which trial supplies the effect, which people are counted as eligible, and whether they are assumed to take the drug for ten years.
In people who already have heart disease
Nanna and colleagues applied SELECT’s hazard ratio for major cardiovascular events to US census projections and national survey data [2]. Of 6,164,019 adults meeting SELECT’s entry criteria, 2,523,218 (40.9%) were projected to have at least one new event in 10 years without further treatment. Semaglutide 2.4 mg may prevent 496,400 of them, a 16% relative reduction. Widening the pool to the 22,653,158 adults who meet the label for cardiovascular risk reduction raised the projection to 1,934,493 events. The model applied the same 16% relative reduction to deaths, giving 1,231,295 deaths avoided.
In people without heart disease
Wong and colleagues started from the weight-loss trial rather than the heart trial [3]. They applied STEP 1’s eligibility criteria to national survey data and found 93.0 million eligible US adults, 38% of the adult population. Feeding STEP 1’s weight change into a BMI-based Framingham risk score, among those without cardiovascular disease, moved ten-year risk from 10.15% to 8.34%. That is 1.81 points absolute and 17.8% relative, or 1.50 million preventable events over 10 years. Weight is the only route this model allows.
Schrage and colleagues used a wider set of inputs [1]. A risk equation fitted in 610,789 people was applied to 200,012 survey participants twice. The first run used their real values. The second shifted BMI, HbA1c, systolic blood pressure, C-reactive protein and non-HDL cholesterol by the amounts SELECT recorded. Among 21,720 people with a BMI of 27 or above and elevated baseline risk, projected ten-year incidence fell from 13.82% to 10.83%. Absolute reductions were larger in men (3.14 points) than in women (2.70).
In people with type 2 diabetes
Karthikeyan and colleagues used SUSTAIN-6, the semaglutide trial in type 2 diabetes [4]. Of an estimated 33.6 million US adults with type 2 diabetes, 6.9 million fit the trial’s criteria. Applying its event rates, the model projects 75,681 primary-outcome cardiovascular events prevented each year if all of them were treated. The eligible sample was more often female and Black, and less likely to have prior cardiovascular disease, than the trial itself.
What the per-person number looks like
Large totals come from large denominators. Schrage’s 2.99-point drop is roughly three fewer events per hundred people over a decade: a meaningful public health quantity and a modest personal one. The same relative-versus-absolute distinction runs through the semaglutide and tirzepatide comparison and the FLOW numbers needed to treat.
Every total also assumes that everyone eligible starts treatment and stays on it. Schrage’s modeled benefit diminished further under lower compliance assumptions [1]. Persistence is the weak point. The drugs that take off the most weight are also the ones people stop soonest, per the network meta-analysis of nineteen drugs. How many eligible people are actually prescribed one after a heart attack or stroke is its own question, covered in the uptake data.
One assumption could cut the other way. The Schrage model routes the whole benefit through five risk factors. SELECT’s inflammation findings, covered in inflammation fell before the weight, suggest part of the effect may arrive by another route, which that model cannot capture.