All Times EDT
Keywords: Real World Data
Following the 20th century cures act, there has been increased interest in using real world data (RWD) to enhance decision making. While there is great potential for RWD, there are many challenges and concerns such as confounding and data comparability. For example, compared to RWD, clinical trial patients generally represent a more stringent population that must satisfy the pre-specified eligibility criteria. Intuitively, the eligibility criteria represent knowledge on the data generating mechanism of the clinical trial patients and accounting for this aspect can reduce bias between the data-sources. A simple approach is to simply select RWD patients that satisfy all available eligibility criteria. This approach has two key problems: (1) It is unlikely all criteria can be applied, (2) Strict filtering can result in small populations if there are numerous criteria and/or missing criteria data (ex: ECOG). In general, filtering tends to discard much of the available RWD. An alternative approach is to model the eligibility criteria and weight patients by their estimated probability of satisfying all criteria. The advantage here is that more RWD patients can be used in the analysis. To illustrate the utility of this method, we examine common clinical eligibility criteria for Flatiron non-small lung cancer pembrolizumab data and assess overall survival differences across different eligibility adjustment methods. Overall, compared to strict eligibility filtering, our analysis indicates that the weighted approach increases efficiency without compromising the results.