ABSTRACT
Fitting models to incomplete categorical data requires more care than fitting models to the complete data counterparts, not only in the setting of missing data that are non-randomly missing, but even in the familiar missing at random setting. Various aspects of this point of view have been considered in the literature. We review it using data from a multi-centre trial on the relief of psychiatric symptoms. First, it is shown how the usual expected information matrix (referred to as naive information) is biased even under a missing at random mechanism. Second, issues that arise under non-random missingness assumptions are illustrated. It is argued that at least some of these problems can be avoided using contextual information.
Subject(s)
Data Interpretation, Statistical , Fluvoxamine/therapeutic use , Mental Disorders/drug therapy , Models, Biological , Selective Serotonin Reuptake Inhibitors/therapeutic use , Fluvoxamine/adverse effects , Humans , Likelihood Functions , Multicenter Studies as Topic , Patient Dropouts , Selective Serotonin Reuptake Inhibitors/adverse effects , Treatment OutcomeABSTRACT
We discuss pragmatic clinical trials with survival endpoints in which subjects commonly change treatment during follow-up. Suppose that an intention-to-treat (ITT) analysis shows a significant difference between the randomized groups. We may want to ask questions about the reason for such a difference in outcome between randomized groups: for example, was the difference due to different policies for change to a third more beneficial regime? We address such questions using the semi-parametric accelerated life models of Robins, which exploit the randomization assumption fully and avoid direct comparisons of possibly differently selected subgroups. No assumption is made about the relationship of treatment actually prescribed to prognosis. A sensitivity analysis, using a range of plausible values for the causal effect of a covariate, estimates the contrasts between randomized groups that would have been observed if the covariate had universally been 0. The main technical problem is in dealing with censoring, for the method requires different degrees of recensoring for different values of the causal effect, and this can lead to estimates of low precision. The methods are applied to a randomized comparison of two anti-hypertensive treatments in which approximately half the subjects changed treatment during follow-up. Various time-dependent covariates, representing patterns of side-effects and treatments, are used in the model. We find that the observed difference in cardiovascular deaths between the randomized groups cannot be explained in this way by their different covariate patterns.
Subject(s)
Clinical Protocols , Data Interpretation, Statistical , Drug Therapy , Models, Statistical , Randomized Controlled Trials as Topic , Adrenergic beta-Antagonists/therapeutic use , Age Factors , Aged , Antihypertensive Agents/therapeutic use , Diuretics/therapeutic use , Female , Follow-Up Studies , Humans , Hypertension/drug therapy , Hypertension/mortality , Male , Placebos , Time FactorsABSTRACT
Assuming that a drug is active and different from placebo, a patient's gain or loss is likely to depend on how much of the drug is taken and when. The present paper motivates the ethical imperative of statistical compliance analysis by considering important clinical questions, the advent of more precise compliance measuring instruments and new statistical efforts towards well understood analyses that seek to preserve scientific integrity. An extension of Efron and Feldman's approach is developed which exploits the randomization assumption in combination with structural models. It also generates more promising designs and alternative statistical approaches.
