Regression Analysis of Current-Status Data: An Application to Breast-Feeding

Abstract

Abstract Current-status data are often collected for use in survival analyses, because this type of data generally is considered to be more reliable than retrospective reports of when an event occurred. Although techniques for calculating mean survival time from current-status data are well known, their use in multiple regression models is somewhat troublesome. Using data on current breast-feeding behavior, this article considers a number of techniques that have been suggested in the literature, including parametric, nonparametric, and semiparametric models as well as the application of standard schedules. Models are tested in both proportional-odds and proportional-hazards frameworks. Although the choice of models does not strongly affect the conclusions that would be drawn, I recommend that a logistic regression in which the baseline log-odds of breast-feeding by child's age are represented by a natural cubic spline should be the preferred methodology. This methodology offers a reasonable compromise between the parsimony of parametric models and the flexibility and good fit of nonparametric models. The same methodology may be well suited to other applications in which there are no theoretical reasons to use a particular parametric form.