Model-based analysis of survival experiments with zero and non-zero final survivors.

Abstract

Modified logistic, Weibull and Gompertz survival models are described which are suitable for the model-based analysis of survival experiments with both zero and non-zero final survivors. They are applied to 3 examples: the influence of food restriction and body fat on the life span of mice and the effect of an immunosuppressant and of an antiviral agent on experimentally virus-infected mice. It is shown that besides the median survival time and the fraction of animals finally surviving the hazard rate at the median survival time, the ratio of maximum to minimum survival time and the area under the survival curve are useful parameters for a more detailed analysis of survival experiments. Hazard plots display differences between experiments more clearly than survival curves. The models described provide a powerful tool which is appropriate for a great variety of different survival experiments. The model-based analysis of survival data is also a prerequisite for the development of computerized data bases on the survival behaviour of laboratory animals. The approach presented can be easily generalized to cases for which spontaneous mortality and mortality due to an experimental challenge interfere.