My current research is focused on the development of statistical methods for analyzing randomized studies in which subjects selectively non-comply with the protocol through either premature withdrawal or discontinuation of assigned therapy. When evaluating the efficacy of treatments, standard methods for handling such data can yield biased results. These methods rely on strong assumptions, which typically cannot be emprically validated and may be scientifically unreasonable. My colleagues and I are advocating the use of non/semi-parametric sensitivity analyses to evaluate how robust the standard analyses are to more plausible assumptions about the non-compliance mechanism.
The sensitivity analysis approach relies heavily on expert opinions about plausible ranges for non-identifiable selection bias parameters. While the methodology is useful in assessing the sensitivity of treatment comparisons to standard assumptions, it may be dissatisfying to some researchers/decision makers because a single answer is not provided. In contrast, a Bayesian analysis allows the investigator to draw a ``single'' inference by formally incorporating prior beliefs about model parameters. My colleagues and I are developing a flexible Bayesian approach, which provides minimum sensitivity of the posterior inference to model or prior specifications for well identified parameters.
Some researchers may feel uncomfortable with the sensitivity and Bayesian approaches because of their relienace on quantifcation of prior beliefs. To address this concern, we are developing a methodology for constructing informative bounds on treatment effects that rely on soft, yet plausible, belief statements, such as "Sicker subject at baseline are more likely to non-comlpy" or "Subjects whose health worsens are more likely to non-comply."
We are applying our methodolgies to the reanalysis
of a number of AIDS and mental health clinical trials.
PAPERS (Postscript and PDF)
Scharfstein DO and Irizarry RA (2002), "Generalized Additive Selection Models for the Analysis of Studies with Potentially Non-ignorable Missing Data," Revised for Biometrics. [Postscript][PDF]
Scharfstein DO, Manski CF, and Anthony JC (2002), "On the Construction of Bounds in Prospective Studies with Missing Ordinal Outcomes: Application to the Good Behavior Game Trial," Submitted to Biometrics. [Postscript][PDF]
Scharfstein DO, Daniels M, and Robins, JM (2002), "Incorporating Prior Beliefs about Selection Bias into the Analysis of Randomized Trials with Missing Outcomes," To Appear Biostatistics. [Postscript][PDF]
Scharfstein DO and Robins JM (2002,) "Estimation of the Failure Time Distribution in the Presence of Informative Censoring," Biometrika, 89, 617-634. [Postscript][PDF]
Rotnitzky A, Scharfstein DO, Su TL, and Robins JM (2001), "Methods for Conducting Sensitivity Analysis of Trials with Potentially Non-ignorable Competing Causes of Censoring," Biometrics, 57, 103-112. [Postscript][PDF]
Robins JM, Rotnitzky A, and Scharfstein DO (2000), "Sensitivity Analysis for Selection Bias and Unmeasured Confounding in Missing Data and Causal Inference Models," in Statistical Models in Epidemiology, the Environment, and Clinical Trials, Editors E. Halloran and D Berry, Springer-Verlag, pgs. 1-95. [PDF]
Scharfstein DO, Rotnitzky A, and Robins JM (1999), "Adjusting for Non-ignorable Drop-out Using Semiparametric Non-response Models," Special Invited Paper for the Theory and Methods Section of Journal of the American Statistical Association, 94, 1096-1146. [Postscript][PDF]
Rotnitzky A, Robins, JM, and Scharfstein DO (1998), "Semiparametric Regression for Repeated Measures Outcomes with Non-ignorable Non-response," Journal of the American Statistical Association, 93, 1321-1339.