The effects of different methods of accounting for observations from euthanized animals in survival analysis

The issue of euthanasia is unique to veterinary clinical studies evaluating survival time. The decision to euthanize an animal is based on several factors including the health of the animal but also age and cost of treatment. The literature shows inconsistent methods used to account for observations from euthanized animals. Also, over 50% and up to 100% of animals in many studies have been euthanized. Our study illustrates the effects of different methods of accounting for observations from euthanized animals in survival analysis. Three data sets with different proportions of outcomes (alive, lost-to-follow-up, dead due to disease of interest, dead due to other disease, euthanized due to disease of interest, euthanized due to other disease) were used. Each data set was stratified according to treatment or a group characteristic (e.g. tumor type). Our methods for accounting for observations from euthanized animals were established from methods used in the literature and included right-censoring. Kaplan-Meier product-limit survival-function estimation was performed on each data set. Different methods resulted in inconsistent conclusions of significant differences between strata. At times, the ranking of the estimates of median survival time for strata was reversed. Right-censoring and use of Kaplan-Meier methods is inappropriate to evaluate observations from euthanized animals because censoring of such observations is informative. The current methods used by clinical investigators are inadequate to measure survival time reliably.