Nonparametric estimation of the integrated intensity of an unobservable transition in a Markov illness-death process (Q1085551)

Let \(\tilde X=(\tilde X(t),0\leq t\leq 1)\) be a nonhomogeneous Markov illness-death process with two transient states, 0 and 1, and two absorbing states, 2 and 3. It is assumed that \(\tilde X(0)=0\) and that states 2 and 3 can be directly reached only from states 0 and 1, respectively. One of the main assumption is that the process X is unobservable during the transitions between the transient states. All information for X is collected directly about transitions to the absorbing states. Thus the problem is to estimate the cumulative (integrated) intensity of the unobservable transitions. The author constructs a special process \(X_ d\) called an associated design process as well as another process N which is a point process based on \(X_ d\). Further he suggests nonparametric estimators in terms of N for the unknown integrated intensity and establishes their unbiasedness and strong consistency. Finally, the exact values of the variance of the proposed estimators are found. Other interesting related topics are discussed too.