Statistical Methods for Panel Data from a Semi-Markov Process, with
Applications to HPV

Minhee Kang
Harvard University
Department of Biostatistics
655 Huntington Avenue
Boston, MA  02115

A common difficulty in studies of disease history is that complete
individual histories are not observed.  Inference methods for such
panel data are simplified under a time-continuous Markov model
(Kalbfleisch and Lawless, 1985), because the resulting sequence of
observed states forms a Markov chain.  However, there has been little
done for panel data from non-Markov models.  We propose
likelihood-based methods for the analysis of panel data when the
underlying process is a continuous-time, semi-Markov process with
discrete states, with at least one state associated with constant
hazards.  To illustrate the methods, a model of human papillomavirus
(HPV) infections and development of cervical cancer precursor is
considered.