A Compartment Model Approach to the Estimation of Tumor Incidence and Growth: Investigation of a Model of Cancer Latency

Publisher Website
Google Scholar
Abstract
The problems involved in determining the effects of a chronic disease process, such as stomach cancer, on the observed mortality of the U.S. population are considered. Since the time of initiation of tumor growth is unknown and the tumor becomes clinically manifest only after reaching considerable size, the early rate and pattern of tumor growth is unobserved. As a possible solution to the analysis of such problems, it is proposed to use stochastic compartment modelling techniques which deal with the problems of estimating the transition probabilities of a partially observed stochastic process. Implementation of the stochastic compartment techniques in this case depends on the selection of certain mathemathical expressions from theories of carcinogenesis, epidemiologic studies and animal studies which allow the calculation of transition probabilities to unobserved states by making them explicit functions of time or age. Though the selection of the specific functions might be subject to debate, the general strategy of explicitly selecting such functions, and thereby exposing them for review in terms of biologic reasonableness and consistency with the data, seems to be a valid and useful methodology. Various ways of viewing the model results (from its internal behavior, e.g., from implied distributions of waiting times in various disease states) yield different insights into the various factors in carcinogenesis. The model, with parameters representing tumor incidence, time to tumor death given onset, genetic susceptibility to tumor growth and the effects of competing forces of mortality, is fitted to data on deaths due to stomach cancer for male U.S. residents age 25 and over in 1969. Two basic forms of the model, one with a waiting time distribution for occupants of the latent state and another with a single latency time, achieved excellent fits to the data. Examination of parameter estimates and compartment waiting time distributions are consistent with theoretical expectations and intuition. Such strategies, involving the integration of clinical, experimental and vital statistics data into a comprehensive model of population carcinogenesis, are potentially powerful tools for investigation of the temporal dimensions of disease development in a human population.