Causal Inference
Benchmarking an observational analysis against a randomized trial: an application to the effect of trastuzumab on the risk of disease-free survival in early curative breast cancer Vanessa Voelskow* Vanessa Voelskow Xabier Garcia de Albeniz Anita Berglund Maria Feychting Tobias Kurth Anthony A. Matthews
Most published examples of benchmarking compare an observational study to a trial that estimates the effect of a treatment on the risk of death or another single outcome that is expected to be captured analogously in a trial and a routine clinical practice setting. This limits the scope of how we can use observational data to complement randomized trials.
We designed a target trial with a protocol as similar as possible to the B-31 and N9831 randomized trials, which estimated the effect of adjuvant trastuzumab plus chemotherapy compared with chemotherapy alone on the risk of disease-free survival and death from any cause in individuals with early human epidermal growth factor receptor 2 positive breast cancer in the curative setting. We then emulated the target trial using observational data from Swedish registries to understand if we can undertake an observational analysis that replicates the effect of trastuzumab not only on the risk of death, but also disease free survival. We applied inverse probability of treatment weighting, while cloning individuals who had data compatible with both treatment strategies at baseline and assigning each copy to one arm.
Our target trial emulation included 1578 women. We found a similar effect estimate after five years of follow-up (RR: 0.46, 95% CI: 0.33, 0.64) for the composite endpoint of disease-free survival as the two jointly analyzed B-31 and N9831 trials (HR: 0.48, 95% CI: 0.39, 0.59). The results after five years for death in our target trial emulation (RR: 0.33, 95% CI: 0.22, 0.49) would also lead to the same clinical decision as those from the randomized trials (HR: 0.67, 95% CI: 0.48, 0.93).
The results indicate that, with high-quality data, benchmarking observational analyses against trials can be extended to outcomes less severe than death, such as disease-free survival used in oncological studies with curative treatment intent.