This is when you do a retrospective study, select cohorts according to exposure, but measure outcomes — usually death, or hospitalization, or something else bad — in a way that guarantees one or more of the cohorts have a period of time when that outcome couldn’t have happened. That’s how you get “immortal”, or “guaranteed” time.
Three classic examples: Courtesy of Bing.
- A study that evaluated the hypothesis that menopause occurring before age 40 years is associated with the development of cardiovascular disease (CVD). The study counted women who had premature menopause as exposed to CVD risk from their date of birth, ignoring the fact that they could not have developed CVD before menopause.
- A study that assessed the effect of statins on diabetes progression and insulin use. The study classified patients as statin users based on their first prescription after hospital discharge, but included the time between discharge and prescription as exposed time, even though they were not taking statins during that period.
- A study that examined the association between beta-blocker use and mortality in patients with heart failure. The study defined beta-blocker exposure based on pharmacy claims after hospitalization, but did not account for the time lag between hospital discharge and drug initiation, which could have biased the results in favor of beta-blockers.
To these three classics we can now add two more, one highly publicized, the other less so, both surprising considering the journals and the supposed peer review they must have gone through:
- A study that compared hospitalization rates of those who had a positive SARS-CoV-2 test and did not take the Covid-19 drug Paxlovid to those who took Paxlovid regardless of test status. The study counted the day of the positive test as “Day 0” for the untreated cohort; however, for patients who started taking Paxlovid and did not have records of a positive test “Day 0” was one day before treatment start. See this Twitter thread for relevant excerpts and a more detailed explanation. As the study excluded patients who were already hospitalized when they started Paxlovid, but included untreated patients who had a positive test and were hospitalized on the same day, this guaranteed 1 day of “immortal time” for the Paxlovid cohort.
- A study that used “real world data” to compare standard blood thinners to a procedure called left atrial appendage occlusion (LAAO) for prevention of death and stroke in patients with atrial fibrillation. However, patients who underwent LAAO had to live long enough with atrial fibrillation without getting a stroke to “graduate into” the procedure. Not only is immortal time here so glaring, the study is eerily similar to the original heart transplantation study in which this bias was first identified.
Cardiologist John Mandrola explains in depth why the LAAO paper, and the way it was spun, is particularly egregious.
Note that this is only a problem in retrospective — or, how they now like to be rebranded, “real-world” — studies. As the most recent cases show, these are not only worthless for informing anyone’s real-world decision, but also contribute to the noise, the chaos, and the general fear-uncertainty-doubt of medicine. A voluntary moratorium would not be out of line.