Estimation of treatment effect adjusting for dependent censoring using the IPCW method: an application to a large primary prevention study for coronary events (MEGA study)

Background The MEGA study is a randomized controlled trial conducted in Japan to evaluate the primary preventive effect of pravastatin against coronary heart disease (CHD), in which 8214 subjects are randomized to diet or diet plus pravastatin. Pravastatin reduces the incidence of CHD (hazard ratio = 0.67; 95%CI: 0.49—0.91). In the MEGA study, in addition to the usual loss to follow-up cases, there is another problem of drop-outs due to the refusal of further follow-up at 5 years. Purpose To estimate the treatment effect adjusting for some types of dependent censorings observed in the MEGA study and to assess the sensitivity of standard analysis results for these censoring cases. Methods The proposed method is a straightforward extension of the inverse probability of censoring weighted (IPCW) method for settings with more than one reason for censoring, where the propensities for drop-outs are modeled separately for each reason. Simulation studies are also conducted to compare the properties of the IPCW estimate with the standard analysis assuming independent censorings. Results Simulation studies show that the IPCW estimate can correct for selection bias due to dependent censoring that can be explained by measured factors, while the standard analysis is biased. Applying the proposed method to the MEGA study data, several prognostic factors are associated with the censoring processes, and after adjusting for these dependent censorings, slightly larger treatment effects for pravastatin are observed for both CHD (primary endpoint) and stroke (secondary endpoint) events. Limitations The method developed is based on the fundamental assumption of sequentially ignorable censoring. Conclusions Our proposed method provides a valuable approach for estimating treatment effect adjusting for several types of dependent censorings. Dependent censorings observed in the MEGA study did not cause a severe selection bias attributable to the covariates and the results from the standard analysis were robust in relation to the censorings. Clinical Trials 2007; 4: 318—328. http:// ctj.sagepub.com

[1]  Yasuo Ohashi,et al.  Primary prevention of cardiovascular disease with pravastatin in Japan (MEGA Study): a prospective randomised controlled trial , 2006, The Lancet.

[2]  D. Heitjan,et al.  Nonignorable censoring in randomized clinical trials , 2005, Clinical trials.

[3]  Design and baseline characteristics of a study of primary prevention of coronary events with pravastatin among Japanese with mildly elevated cholesterol levels. , 2004, Circulation journal : official journal of the Japanese Circulation Society.

[4]  O. Borgan.  The Statistical Analysis of Failure Time Data (2nd Ed.). John D. Kalbfleisch and Ross L. Prentice , 2003 .

[5]  J. Kalbfleisch,et al.  The Statistical Analysis of Failure Time Data: Kalbfleisch/The Statistical , 2002 .

[6]  James M. Robins,et al.  Estimation of the failure time distribution in the presence of informative censoring , 2002 .

[7]  Somnath Datta,et al.  The Kaplan–Meier Estimator as an Inverse-Probability-of-Censoring Weighted Average , 2001, The American statistician.

[8]  J. Robins,et al.  Correcting for Noncompliance and Dependent Censoring in an AIDS Clinical Trial with Inverse Probability of Censoring Weighted (IPCW) Log‐Rank Tests , 2000, Biometrics.

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

[10]  J. Robins,et al.  Estimating the effect of zidovudine on Kaposi's sarcoma from observational data using a rank preserving structural failure-time model. , 1998, Statistics in medicine.

[11]  James M. Robins,et al.  Causal Inference from Complex Longitudinal Data , 1997 .

[12]  J. Robins,et al.  Analysis of semiparametric regression models for repeated outcomes in the presence of missing data , 1995 .

[13]  L. J. Wei,et al.  The Robust Inference for the Cox Proportional Hazards Model , 1989 .

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

[15]  J. Kalbfleisch,et al.  The Statistical Analysis of Failure Time Data , 1980 .

[16]  D. Rubin.  INFERENCE AND MISSING DATA , 1975 .

[17]  D. Horvitz,et al.  A Generalization of Sampling Without Replacement from a Finite Universe , 1952 .