Measuring cause-and-effect relationships without randomized clinical trials: Quasi-experimental methods for gynecologic oncology research

Clinical research in gynecologic oncology has seen a proliferation of studies that investigate the effectiveness of treatments using existing data sources such as cancer registries, electronic health records, and insurance claims. These observational studies are often feasible when randomized trial may not be, and may be more generalizable than randomized trials, because of greater diversity in the study populations. While statistical methods such as multivariable regression, matching, stratification, and weighting can adjust for the confounding in observational studies, statistical adjustment cannot control for confounders that are unmeasured in the data. Observational studies comparing the effectiveness of treatments for gynecologic malignancies are susceptible to bias from unmeasured confounding because factors like functional status, frailty and disease burden, which influence treatment selection and outcome, are often not reported in existing data sources. Like randomized trials, quasi-experimental designs attempt to account for both measured and unmeasured confounding by exploiting natural experiments arising in the real world. These methods are underutilized in gynecologic oncology research and are particularly relevant to studies that use large datasets to study the effectiveness of treatments. In this review, we consider methodological challenges that arise in the analysis of non-randomized studies, and describe how application of quasi-experimental methodology can estimate unbiased treatment effects even in the presence of unmeasured confounders. (C) 2018 Elsevier Inc. All rights reserved.