How to Read a Composite Endpoint, Starting With Edwards’ TAVR Trial

Edwards Lifesciences’ PROGRESS trial met its main goal in San Diego on Monday and drew what J.P. Morgan analysts called a “very strong round of applause.” Yet at two years, 8.4% of the patients who got a new heart valve early had died, against 7.3% of those who waited. Both facts can be true because the main goal was a composite endpoint, the share of patients who had any one of several bad outcomes. A founder planning a pivotal study will likely have to choose one, and should know how it comes apart.

PROGRESS, funded by Edwards, enrolled 744 people, average age 78, with moderate narrowing of the aortic valve and at least one at-risk feature, such as symptoms or signs of strain on the heart. Half got transcatheter aortic valve replacement, or TAVR, in which a new valve goes in by catheter. They got Edwards’ Sapien 3, which is still investigational for moderate disease. The other half got what guidelines call for: a heart ultrasound every one to two years and a valve when one was indicated.

The composite counted a death from any cause or a heart failure event by two years. It came in at 16.7% for TAVR and 23.9% for surveillance, with a P value of 0.02, under the usual 0.05 line for calling a result unlikely to be chance. Designers combine events this way because a few hundred patients may not produce enough of any one event to reach an answer. The cost is that one number can hide which event moved, so I read a composite in three passes.

Split the composite into its parts

Heart failure events ran 10.8% against 19%, a gap of 8.2 points, wider than the composite’s 7.2. Deaths ran the other way, so heart failure supplied the win. Edwards’ release calls mortality “similar,” and I agree: 1.1 points in an arm of about 370 people is roughly four patients, too few to carry a conclusion. Still, the trial hasn’t shown that patients live longer. The investigators plan to follow participants for 10 years to find out whether fewer heart failure events become longer survival.

Ask who decides that each event happened

A death needs no judgment. A heart failure event does, and this composite counted them with or without a hospital stay. PROGRESS was open-label, so patients and doctors knew who had a new valve. A doctor watching a 78-year-old whose narrowed valve hasn’t been replaced may well be quicker to call new breathlessness heart failure. I’ve read the abstract and the meeting summary, not the full methods, so I can’t tell you how the trial defined these events or whether a blinded committee judged them. In your own protocol, write down what counts as an event, and have people who don’t know the treatment arm judge it.

Hospitalizations for heart failure alone ran 7.4% against 14.9%, with a P value of 0.001, so the gap survives when a hospital bed is required. Put death back in and add stroke, as the trial’s secondary endpoint did, and the result narrows to 17.7% against 23.3%, with a P value of 0.07, over the 0.05 line.

Check what the control arm did

Edwards reports that 57% of the surveillance group had a valve replaced within two years, and about 80% of those patients had progressed to severe disease first. They still count as surveillance, and their later events count against surveillance, because an intention-to-treat analysis keeps patients in the arm they were assigned. So the 16.7% and 23.9% measure a valve now against a valve when indicated. Count how many of your own control patients get the treatment anyway. The more who do, the more your result speaks to timing and the less it says about treatment against no treatment.

Timing also changes how a market adds up. An Edwards executive told MedTech Dive that one to two times as many people have moderate disease as severe, and that about half of them have at-risk features. In a trial with scheduled ultrasounds, more than half of the patients told to wait reached a valve within two years. Real-world numbers will differ, but some of that apparent new market is the same patients treated sooner. A founder pitching an earlier-treatment story should count new patients and earlier patients separately.

The same reading works outside medicine: at a robot maker, someone decides whether a remote operator’s rescue counts as downtime. Before signing off on a pivotal endpoint or a board metric, I’d have the team draft the results table before any data comes in. Each component gets its own line, the least flattering goes on top, and next to each line sits the role of whoever makes the judgment call and what they can see when they make it.

Dave Saunders is the founder of Base Reality Group and a Fractional CPO for product companies. He was a founder and operator at Galen Robotics, where the surgical-robotics platform earned FDA De Novo authorization in 2023, and he managed a 35-patent portfolio licensed from Johns Hopkins. He wrote Founders Who Finish and publishes The Build. More about Dave →