Using Propensity Scores Subclassification to Estimate Effects of Longitudinal Treatments: An Example Using a New Diabetes Medication

Background:When using observational data to compare the effectiveness of medications, it is essential to account parsimoniously for patients’ longitudinal characteristics that lead to changes in treatments over time. Objectives:We developed a method of estimating effects of longitudinal treatments that uses subclassification on a longitudinal propensity score to compare outcomes between a new drug (exenatide) and established drugs (insulin and oral medications) assuming knowledge of the variables influencing the treatment assignment. Research Design/Subjects:We assembled a retrospective cohort of patients with diabetes mellitus from among a population of employed persons and their dependents. Methods:The data, from i3Innovus, includes claims for utilization of medications and inpatient and outpatient services. We estimated a model for the longitudinal propensity score process of receiving a medication of interest. We used our methods to estimate the effect of the new versus established drugs on total health care charges and hospitalization. Results:We had data from 131,714 patients with diabetes filling prescriptions from June through December 2005. Within propensity score quintiles, the explanatory covariates were well-balanced. We estimated that the total health care charges per month that would have occurred if all patients had been continually on exenatide compared with if the same patients had been on insulin were minimally higher, with a mean monthly difference of $397 [95% confidence interval (CI), $218–$1054]. The odds of hospitalization were also comparable (relative odds, 1.02; 95% CI, 0.33–1.98). Conclusions:We used subclassification of a longitudinal propensity score for reducing the multidimensionality of observational data, including treatments changing over time. In our example, evaluating a new diabetes drug, there were no demonstrable differences in outcomes relative to existing therapies.

[1]  J. Robins.  A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. , 1987, Journal of chronic diseases.

[2]  D. Rubin,et al.  Reducing Bias in Observational Studies Using Subclassification on the Propensity Score , 1984 .

[3]  J. Robins,et al.  Marginal Structural Models and Causal Inference in Epidemiology , 2000, Epidemiology.

[4]  J. Lunceford,et al.  Stratification and weighting via the propensity score in estimation of causal treatment effects: a comparative study , 2004, Statistics in medicine.

[5]  Dennis D. Kim,et al.  Effects of exenatide (exendin-4) on glycemic control over 30 weeks in sulfonylurea-treated patients with type 2 diabetes. , 2004, Diabetes care.

[6]  Kristin Taylor,et al.  Exenatide improves glycemic control and reduces body weight in subjects with type 2 diabetes: a dose-ranging study. , 2005, Diabetes technology & therapeutics.

[7]  Dennis D. Kim,et al.  Effects of exenatide (exendin-4) on glycemic control over 30 weeks in patients with type 2 diabetes treated with metformin and a sulfonylurea. , 2005, Diabetes care.

[8]  J. Robins,et al.  Estimation of Regression Coefficients When Some Regressors are not Always Observed , 1994 .

[9]  D. Rubin.  Estimating causal effects of treatments in randomized and nonrandomized studies. , 1974 .

[10]  Donald B. Rubin,et al.  Bayesian Inference for Causal Effects: The Role of Randomization , 1978 .

[11]  B. Yawn,et al.  Identifying Persons with Diabetes Using Medicare Claims Data , 1999, American journal of medical quality : the official journal of the American College of Medical Quality.

[12]  Dennis D. Kim,et al.  Effectiveness of progressive dose‐escalation of exenatide (exendin‐4) in reducing dose‐limiting side effects in subjects with type 2 diabetes , 2004, Diabetes/metabolism research and reviews.

[13]  Dennis D. Kim,et al.  Effects of exenatide (exendin-4) on glycemic control and weight over 30 weeks in metformin-treated patients with type 2 diabetes. , 2005, Diabetes care.

[14]  J. Leahy,et al.  Exenatide Versus Insulin Glargine in Patients With Suboptimally Controlled Type 2 Diabetes: A Randomized TrialHeine RJ, for the GWAA Study Group (VU Univ, Amsterdam; et al) Ann Intern Med 143:559–569, 2005§ , 2006 .

[15]  Donald R. Miller,et al.  Who has diabetes? Best estimates of diabetes prevalence in the Department of Veterans Affairs based on computerized patient data. , 2004, Diabetes care.

[16]  Dennis D. Kim,et al.  Effect on glycemic control of exenatide (synthetic exendin-4) additive to existing metformin and/or sulfonylurea treatment in patients with type 2 diabetes. , 2003, Diabetes care.

[17]  D. Rubin,et al.  The central role of the propensity score in observational studies for causal effects , 1983 .

[18]  D. Rubin.  [On the Application of Probability Theory to Agricultural Experiments. Essay on Principles. Section 9.] Comment: Neyman (1923) and Causal Inference in Experiments and Observational Studies , 1990 .