ESTIMATING PARAMETRIC SURVIVAL MODEL PARAMETERS IN GERONTOLOGICAL AGING STUDIES - METHODOLOGICAL PROBLEMS AND INSIGHTS

Studies of the biology of aging (both experimental and evolutionary) f requently involve the estimation of parameters arising in various mult i-parameter survival models such as the Gompertz or weibull distributi on. Standard parameter estimation methodologies, such as maximum likel ihood estimation (MLE) or nonlinear regression (NLR), require knowledg e of the actual life spans or their explicit algebraic equivalents in order to provide reliable parameter estimates. Many fundamental biolog ical discussions and conclusions are highly dependent upon accurate es timates of these survival parameters (this has historically been the c ase in the study of genetic and environmental effects on longevity and the evolutionary biology of aging). In this article, we examine some of the issues arising in the estimation of gerontologic survival model parameters. We not only address issues of accuracy when the original life-span data are unknown, we consider the accuracy of the estimates even when the exact life spans are known. We examine these issues as a pplied to known experimental data on diet restriction and we fit the f requently used, two-parameter Gompertzian survival distribution to the se experimental data. Consequences of methodological misuse are demons trated and subsequently related to the values of the final parameter e stimates and their associated errors. These results generalize to othe r multiparametric distributions such as the Weibull, Makeham, and logi stic survival distributions.