Causal Inference
Transporting the effects of a randomized trial to a target population: an example using the TASTE trial Conor MacDonald* Conor MacDonald Anthony Matthews Sarah Robertson Ole Frobert Stefan James David Erlinge Bertil Lindahl Tomas Jernberg Maria Feychting Issa Dahabreh Anita Berglund Miguel A. Hernan
Background
When the distribution of effect-modifiers varies between participants of a randomized trial and individuals in a target population, the effect of treatment in the trial population may not equal that in the target population. We illustrate how to extend treatment effects estimated from a registry-based randomized trial (the index trial) to a target population of individuals from the corresponding registry. We describe methods that can be used for time-to-event outcomes in the presence of non-adherence to the trial protocol.
Methods
The TASTE trial, nested in the SWEDEHEART registry that collects baseline characteristics of patients undergoing percutaneous coronary intervention (PCI) in Sweden, compared PCI plus thrombus aspiration to PCI alone on the risk of death after myocardial infarction. To estimate the risks in the target population under full adherence to the trial protocol, we used (separately) standardization and inverse probability (IP) weighting to (i) estimate the risks in the trial under full adherence to the trial protocol, and (ii) extend the estimates to trial-eligible individuals in the SWEDEHEART registry. We compared the risks via differences and ratio. 95 % confidence intervals obtained by bootstrapping with 500 samples.
Results
Under both IP weighting and standardization, the risks of death were higher in the target population than in the trial, but the risk differences and ratios were comparable to those in the trial (TASTE trial, death: IP weighted risk difference -0.1 (95% CI: -1.0, 0.7), standardized risk difference -0.1 (95% CI: -1.0, 0.7); target population, death: IP weighted risk difference 0.6 (95% CI: -1.0, 2.4), standardized risk difference -0.1 (95% CI: -1.1, 1.6)).
Conclusion
Registry data from Sweden allowed identification of enrolled and non-enrolled populations eligible for the TASTE trial. Extending the results from the trial to the target population found no difference on death for PCI alone or PCI with thrombus aspiration.