A retrospective statistical study has been performed in order to identify the biological, clinical and behavioural variables that could potentially predict the survival status of irradiated non-human primates (NHP) and to refine the future use of humane end points (HEP). The available data come from experiments that were initially designed and implemented to evaluate the efficacy of various treatments of the radiation-induced hematopoietic syndrome. A group of 15 NHP was selected on the basis of similar irradiation and treatment protocols. Among those 15 NHP, 3 animals died during the study despite efforts developed to prevent high suffering and death. Among individual variables collected during the study, the ones retained for the present analysis were Flt3-ligand concentration in plasma, body temperature, alimentary behaviour and general behaviour score. A Bayesian predictive analysis showed that, although some of these variables were of potential interest, none of them alone proved valuable in predicting the death of the three animals. A refined analysis showed that a rapid decrease in body temperature during 3 consecutive days of more than 2.5°C was predictive of the death of two out of three animals. Moreover, a combination of several variables including decrease in body temperature and general behaviour score, improved the prediction of death. These results confirm that a HEP cannot only rely on the use of a single variable, strongly suggesting using a combination of variables, including behaviour, as a composite HEP in NHP experiments using high doses of irradiation.