Every few days, we will be publishing layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.
The article featured today is from Statistics in Medicine and the full article, published in issue 38.23, is available to read online here.
, , . Analysis of linear transformation models with covariate measurement error and interval censoring. Statistics in Medicine. 2019;; 38: 4642– 4655. https://doi.org/10.1002/sim.8323
In medical studies, the actual time of disease onset is often unobserved. Subjects are usually scheduled for regular doctor visits. Thus, only the two time points between which the event occurs are observed. For subjects who fail to show up at these scheduled visits, the observed data consist of irregularly spaced time intervals and for those who drop out of the study, the observed data are just the last recorded doctor visits signifying that the events occurred after that time. The primary goal of these studies is to understand the association between the time-to-event and the recorded subject characteristics (covariates). Besides the issue of not observing the time-to-event, often some of these covariates are measured with error. In this paper, the authors propose a two-step approach to study this association when neither the time-to-event nor the covariates are truly observed. First, they use a flexible model to impute the values of the covariates that are measured with error. Next, they use these values to impute the time-to-event from the knowledge that they either lie in between two known doctor visits or if the subjects drop out these events occur after the last recorded doctor visit. By repeating these two steps, the authors generate such imputed datasets multiple times. To study the association, they use a highly versatile linear transformation model that contains the popular models of survival analysis such as Cox proportional hazards and proportional odds models as special cases. They estimate the model parameters from each imputed dataset. The final results are based on the average over all these estimates. The authors demonstrate their method in various settings through simulation and then analyze a real data set from an AIDS clinical trial. An R package named ICEMELT that implements this methodology will be available in CRAN soon.