Comparing the incomparable: hemodialysis versus peritoneal dialysis in observational studies

A randomized trial comparing survival in hemodialysis and peritoneal dialysis remains a utopian aspiration. Dialysis is still relatively rare on a population basis, and a natural tension exists between desirability and feasibility in terms of quality of evidence. In practice, it is very difficult to perform prospective comparisons with large groups of contemporary representative subjects, and much of the literature comes from retrospective national registries. This article considers several questions to address when trying to compare the outcomes of peritoneal dialysis and hemodialysis. Prognostic similarity at baseline is a fundamental issue. Traditionally, adjustment for known prognostic factors has been used in an attempt to minimize the bias caused by nonrandom treatment assignment. Propensity scores have been suggested to be superior, and matched-case analysis may also be a useful method for comparison. Other questions include, when, in relation to starting dialysis, to start the observation clock; the definition and handling of switches of dialysis therapy; and the decision to censor at transplantation. Finally, comparisons are complicated by hazards ratios that vary over time, and time-segmented analysis is obligatory. Many types of analytical approaches are needed to begin to appreciate outcome disparities between dialysis therapies.