Time-to-event studies: how robust are inferences when the assumption of independent censoring is in doubt?
نویسندگان
چکیده
The Cox proportional hazards model is often used to analyse time-to-event data. In many patients, the event time is unknown, either due to dropout or study end. This is known as censoring. When using the Cox proportional hazards model, it is assumed that censoring is independent of event time. However, this is an untestable, often implausible assumption; it may be that censoring is associated with a change in the hazard of an event. Our interest is in examining departures from this assumption, as a sensitivity analysis. One recentlyproposed approach involves treating event time data of censored patients as missing, adjusting the hazards of those patients, and using multiple imputation to impute the missing event times. How the hazard changes after censoring is determined by a user-defined sensitivity parameter. A positive value indicates an increase in hazard (harmful effect), while a negative value indicates a decrease (protective effect). The most basic change in hazard is a step change applied to all censored patients. However, the change may be time-dependent, or applied to a subset of censored patients based on their covariates or strata. By examining how parameter estimates change as this sensitivity parameter is varied, we see how robust our analysis is to the assumption of independent censoring. We apply this approach for the first time to example data containing stratified data, in an area for which there is prior evidence that censoring is associated with event hazard. We relax the independent censoring assumption, and show how the parameter estimates change.
منابع مشابه
Modeling and Inferential Thoughts for Consecutive Gap Times Observed with Death and Censoring
In the perspective of biomedical applications, consider a re- current event situation with a relatively low degree of recurrence. In this setting, the focus is placed on successive inter-event gap times which are observed in the presence of both a terminal event like death and inde- pendent censoring. The terminal event is potentially related to recurrent events while the censoring process is a...
متن کاملAn EM Algorithm for Estimating the Parameters of the Generalized Exponential Distribution under Unified Hybrid Censored Data
The unified hybrid censoring is a mixture of generalized Type-I and Type-II hybrid censoring schemes. This article presents the statistical inferences on Generalized Exponential Distribution parameters when the data are obtained from the unified hybrid censoring scheme. It is observed that the maximum likelihood estimators can not be derived in closed form. The EM algorithm for computing the ma...
متن کاملA Bayesian model for time-to-event data with informative censoring.
Randomized trials with dropouts or censored data and discrete time-to-event type outcomes are frequently analyzed using the Kaplan-Meier or product limit (PL) estimation method. However, the PL method assumes that the censoring mechanism is noninformative and when this assumption is violated, the inferences may not be valid. We propose an expanded PL method using a Bayesian framework to incorpo...
متن کاملRelaxing the independent censoring assumption in the Cox proportional hazards model using multiple imputation
The Cox proportional hazards model is frequently used in medical statistics. The standard methods for fitting this model rely on the assumption of independent censoring. Although this is sometimes plausible, we often wish to explore how robust our inferences are as this untestable assumption is relaxed. We describe how this can be carried out in a way that makes the assumptions accessible to al...
متن کاملStrong Convergence Rates of the Product-limit Estimator for Left Truncated and Right Censored Data under Association
Non-parametric estimation of a survival function from left truncated data subject to right censoring has been extensively studied in the literature. It is commonly assumed in such studies that the lifetime variables are a sample of independent and identically distributed random variables from the target population. This assumption is often prone to failure in practical studies. For instance, wh...
متن کامل