Comparison of state occupation, entry, exit and waiting times in two or more groups based on current status data in a multistate model

Lan and Datta have a paper in Statistics in Medicine. This complements their previous work on non-parametric estimation of current status data multi-state models. Specifically, they develop a K-sample test to compare state occupation probabilities, entry time distribution or state waiting time distributions between groups. The test statistic is based on the integrated L1 distance between the quantities of interest. A p-value for the test is found by bootstrap resampling. This is computationally feasible because of the relative simplicity of the estimator requiring only pool-adjacent-violator algorithm and kernel smoothing. While no Markov assumption is required for the state occupation probabilities, entry and waiting time distributions require a Markov assumption.

The method is demonstrated using a 5-state progressive model on data relating to pubertal development in children. They confirm that there is a significant difference in pubertal development between girls and boys, with boys reaching stages of development later.