Back

Extending treatment effects from a randomized trial using observational data: an application to coronary thrombus aspiration

Matthews, A. A.; Dahabreh, I. J.; Frobert, O.; Lindahl, B.; James, S.; Feychting, M.; Jernberg, T.; Berglund, A.; Hernan, M.

2021-09-08 epidemiology
10.1101/2021.09.06.21263156 medRxiv
Show abstract

To increase confidence in the use of observational analyses when addressing effectiveness questions beyond those addressed by randomized trials, one can first benchmark the observational analyses against existing trial results. We use Swedish registry data to emulate a target trial similar to the TASTE randomized trial, which found no difference in the risk of death or myocardial infarction by 1 year with or without thrombus aspiration among individuals with ST-elevation myocardial infarction. We benchmark the emulation against the trial at 1 year, then extend the emulations follow up to 3 years and estimate effects in subpopulations underrepresented in the trial. As in the TASTE trial, the observational analysis found no differences in risk of outcomes by 1 year between groups (risk difference 0.7 (-0.7,2.0) and -0.2 (-1.3,1.0) for death and myocardial infarction respectively), so benchmarking was considered successful. We additionally show no difference in risk of death or myocardial infarction by 3 years, or within subpopulations by 1 year. Benchmarking against an index trial before using observational analyses to answer questions beyond those the trial could address allowed us to explore whether the observational data can be trusted to deliver valid estimates of treatment effects.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.