Defining "baseline" using propensity score matching: Application to a clinical trial
Samantha R. Cook, Google, Inc.
Donald Rubin, Harvard University
*Elizabeth A. Stuart, Johns Hopkins Bloomberg School of Public Health
Keywords: observational study, historical data, longitudinal analysis
When comparing exposed and unexposed groups it is often unclear how to define 'baseline' for the unexposed group. We know the date that exposed individuals began receiving the treatment, but we do not know the comparable date for unexposed individuals. In this setting it is helpful to consider an underlying randomized experiment that could have occurred, whereby at some point in time--which varies by individual--each person either did or did not begin receiving the treatment. Propensity score matching techniques can assist in defining this baseline by selecting the time at which each unexposed individual looked the most similar to an exposed individual. We discuss this issue in the context of a clinical trial where patients in the trial were matched to patients in a historical patient registry: for each historical patient we identify the date at which they could have entered the trial.