Bayesian Methods for Comparative Effectiveness Research
*Francesca Dominici, Harvard School of Public Health
Treatment strategies in cancer research are becoming increasingly complex, and the need to compare these strategies on patient outcomes cannot be greater. Often, it is infeasible to compare these complex treatments in randomized clinical trials (RCTs). This is because RCTs cannot include a large enough sample of the population, are too costly and time consuming, and the treatments are too complex to implement a randomization. More recently, large administrative databases have been increasingly used for CER. The use of these data in CER is attractive: they include large populations, they allow us to compare complex treatment strategies, and the data are relatively inexpensive. However, these large databases have many challenges that must be overcome to make reliable inference. First, these are observational data and therefore methods for confounding adjustment must be developed to assure that a large set of potential confounders are balanced between treatment groups. Second, there is often a misclassification in the treatment assignment---that is the incorrect diagnostic or procedural code is used. Third, often more than one administrative database is available to make the comparison of interest, but these databases include different sets of measured potential confounders, yet undoubtedly careful integration of these databases will improve estimation.