This paper is the second of two dealing with newly developed statistical techniques for relating risk factors to heart disease and mortality for use with data from prospective epidemiological studies. The first paper introduced an exponential-Weibull model, which is contrasted with the multiple logistic function using the relationship of systolic blood pressure, serum cholesterol, and cigarette smoking to death from all cases in 1233 white males from the Chicago Peoples Gas Company Study. This model utilizes individual survival time as the primary variable, with the risk factors treated as covariates. The present paper introduced two competing risk models for use with cause specific mortality. These two models are the logistic discrimination model, an extension of the multiple logistic function to the problem of multiple endpoints, and an exponential-Weibull model which is the competing risk analog of the model given in . These two models are compared with one another and with the multiple logistic function, utilizing the relationships of systolic blood pressure, serum cholesterol, and cigarette smoking to 14-yr mortality from both cardiovascular disease and coronary heart disease in this cohort of 1233 men. The two competing risk models are shown to provide stronger relationships between the risk factors and cause specific mortality than are obtainable with the multiple logistic function, although the estimated probabilities calculated from the multiple logistic function do not differ markedly from those of the other two models until one reaches the upper end of the distribution of risk. The exponential-Weibull model and the logistic discrimination model are comparable in their assessment of the strength of the relationships of the variables of the two causes of death, and in the probabilities calculated from the estimated coefficients. However, the exponential-Weibull model can be used to estimate the probability of death for any time period, while the logistic discrimination model can be used only for the period for which the coefficients are estimated.
ASJC Scopus subject areas