Emerging Themes in Epidemiology

Purpose: To control for confounding bias from non-random treatment assignment in observational data, both traditional multivariable models and more recently propensity score approaches have been applied. Our aim was to compare a propensity score-stratified model with a traditional multivariable-adjusted model, specifically in estimating survival of hemodialysis (HD) versus peritoneal dialysis (PD) patients. Methods: Using the Dutch End-Stage Renal Disease Registry, we constructed a propensity score, predicting PD assignment from age, gender, primary renal disease, center of dialysis, and year of first renal replacement therapy. We developed two Cox proportional hazards regression models to estimate survival on PD relative to HD, a propensity score-stratified model stratifying on the propensity score and a multivariable-adjusted model, and tested several interaction terms in both models. Results: The propensity score performed well: it showed a reasonable fit, had a good c-statistic, calibrated well and balanced the covariates. The main-effects multivariable-adjusted model and the propensity score-stratified univariable Cox model resulted in similar relative mortality risk estimates of PD compared with HD (0.99 and 0.97, respectively) with fewer significant covariates in the propensity model. After introducing the missing interaction variables for effect modification in both models, the mortality risk estimates for both main effects and interactions remained comparable, but the propensity score model had nearly as many covariates because of the additional interaction variables. Conclusion: Although the propensity score performed well, it did not alter the treatment effect in the outcome model and lost its advantage of parsimony in the presence of effect modification. Introduction Using observational data to compare outcomes associated with different treatments may result in biased estimates because of non-random treatment assignment. To correct for variables that may confound an association, the traditional approach is to apply multivariableadjusted modeling, but in recent years, use of propensity scores has become increasingly popular [1]. The concept of a multivariate confounder score was first introduced by Miettinen in 1976 [2], but the formal concept of propensity scores to estimate causal effects in observational studies was first described by Rosenbaum and Rubin [3]. A propensity score is a conditional probability of assignment to a particular treatment given a vector of baseline covariates. Except for unmeasured potential confounding factors, two patients having the same propensity score but assigned to different treatments are considered to be equivalent to a random assignment of treatment. Thus, adjustment for the propensity score in the outcome model can balance the observed and included covariates and remove bias that may arise due to these confounders. This adjustment can be accomplished by either 1) selecting matched pairs of patients each on a different treat* Correspondence: [email protected] 1 Program for the Assessment of Radiological Technology (ART Program), Department of Epidemiology & Biostatistics and the Department of Radiology, Erasmus University Medical Center Rotterdam, Dr. Molewaterplein 50, 3015 GE Rotterdam, the Netherlands Full list of author information is available at the end of the article © 2010 Liem et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Liem et al. Emerging Themes in Epidemiology 2010, 7:1 http://www.ete-online.com/content/7/1/1 Page 2 of 8 ment arm, but with similar propensity scores, 2) stratifying the sample on the propensity score, calculating the treatment effect within strata and then pooling the strata-specific treatment effect estimates, or 3) including the propensity score itself as a covariate in the outcome model. Several advantages of propensity score-stratified versus traditional multivariable-adjusted modeling have been suggested. The propensity model does not need to be parsimonious and easy to understand because it is not the focus of the study [4]. Furthermore, the propensity score enables a direct estimation of comparability of the treatment groups by assessing the covariate balance between groups. Inability to balance confounders alerts investigators that the treatment groups are not sufficiently overlapping with respect to these confounders and that selection bias may not be resolvable [4]. Traditional multivariable regression modeling will not detect this directly. Patients with end stage renal disease (ESRD) require renal replacement therapy (RRT). Of all therapeutic options, renal transplantation is generally associated with the highest survival and quality of life. However, due to the shortage of organs, the majority of ESRD patients are treated with renal dialysis. Two main forms of renal dialysis can be distinguished: hemodialysis (HD) and peritoneal dialysis (PD). Many factors influence dialysis treatment assignment: not only the clinical characteristics of a patient, but also patient and physician preference, cultural factors and reimbursement policy decisions may play a role. Therefore, comparison of patient survival on HD and PD is complicated. Because the one randomized controlled trial that has been undertaken to assess survival differences had to be stopped prematurely because of low inclusion rates [5], observational studies have to be relied upon to compare survival on HD versus PD. Our aim was to compare a propensity score-stratified model with a traditional multivariable-adjusted model, specifically in estimating survival of hemodialysis (HD) versus peritoneal dialysis (PD) patients to assess the possible advantages of using a propensity score.