Dr. Robert Nam from the University of Toronto presented a clear and concise overview of the use of PSM and its value in studies comparing RP and XRT. Selection bias occurs any time a patient or physician makes a decision about treatment. The magnitude of hidden bias is estimated using a propensity score, which is obtained by performing logistic regression on the available covariates and outcome variables the study is measuring.
There are various ways to incorporate propensity scores, and Dr. Nam highlighted propensity matching in a variety of studies published recently. Interestingly, many of the highlighted studies did not show a difference between regular regression and after investigators applied PSM. This can partly be accounted for by the reduction of power in an analysis after applying PSM. Power reduction occurs because one must find matched pairs to analyze based on selected variables, and cannot analyze the data in its raw form.
The inherent limitation of PSM is its inability to truly account for all unmeasured confounding. Another, perhaps preferred way, to account for this confounding may be to perform appropriate logistic regression followed by sensitivity analysis.
With regard to RP vs. XRT for the treatment of prostate cancer, surgery appears to have better outcomes than radiation over a wide range of observational studies. While PSM studies have been instrumental in evaluating big data, caution should continue to be exercised with this tool’s use. As Dr. Nam concludes, a certain amount of selection bias may be good – we are selecting patients appropriately for the correct treatments.
Presented By: Robert Nam, MD, University of Toronto
Written By: Shreyas Joshi, MD, Fox Chase Cancer Center, Philadelphia, PA
at the 2017 AUA Annual Meeting - May 12 - 16, 2017 – Boston, Massachusetts, USA