Application of multistage Markov modeling to malignant melanoma progression

A fundamental research goal in clinical melanoma studies is to understand the natural history of melanoma and its relationship with prognostic factors. The current understanding of melanoma progression and the relationship of risk factors is based on two-stage modeling.

The authors propose a multistage Markov model for melanoma progression. This model is applied to a data set consisting of approximately 3900 follow-up staging visits of patients with melanoma. The approach takes into account the heavy censoring encountered in this data set and all chronic, subclinically progressive disease. The Markov transition parameters are expressed as Cox regression functions of the relevant prognostic variables. Parameter estimation is achieved by using a missing-data approach.

Tumor thickness, level, and site and patient gender and age at diagnosis are independent risk factors in the transition from local to nodal disease. Tumor thickness and level of invasion and patient age are factors in the transition from local disease to dissemination (without intervening involvement of the nodes). Tumor thickness, patient age, primary site, and the number of involved lymph nodes are factors in the transition from nodal disease to dissemination. Transition from dissemination to death is affected by primary thickness, patient gender, number of nodes involved, and site of metastases.

This multistage analysis contributes to a more accurate understanding of the progress of melanoma and is likely to be applicable to the study of other progressive diseases. Graphic goodness-of-fit results suggest satisfactory predictability of the model to other data sets.