Semi-Markov Models for Lifetime Data Analysis

A semi-Markov model is applied to a data set of items that change stages over time. Two data sets are considered, in medical studies and in reliability: survival to breast cancer and the lifetime of optical fibres. The lifetimes of the subjects or items are considered, distribution functions are fitted to the staying times, and it is shown that the Weibull and log-normal distributions play an important role in these times. Covariates are incorporated to the model via the probability density functions of the staying times in the medical data set. The likelihood function is built and the parameters estimated. The transition probability functions and the survived functions for different risk groups are estimated from these values. A comparison with the empirical survival functions is made. Original computational programmes have been performed with MATLAB to estimate the transition probability functions, the survival functions, and the graphs of these functions.