Estimation of the treatment effect in a clinical trial when recurrent events define the endpoint.

Abstract

Recurrent events are frequently encountered in clinical trials when individuals may experience an event more than once. Examples are common in medical research, including infectious episodes, myocardial infarctions and hospital admissions. However, when only the first event is considered, there is a loss of information. Furthermore, the analysis of recurrent events is complicated by the dependence of the related failure times of each subject, that is, the occurrence of an event influences the risk of other events. Thus, naive statistical methods that consider recurrent events as independent observations will produce misleading conclusions. We recommend the use of a marginal hazards model that is derived from the multivariate generalization of the Cox proportional hazards model. This marginal model allows estimation of the relative risk of recurrence in clinical trials, taking into account the dependence between the recurring events of a same individual without explicit modelling. Two applications are used to compare the results of the marginal model analysis with those of usual methods: (i) a placebo-controlled randomized clinical trial performed to evaluate the efficacy of an immunostimulant in the prevention of recurrences of infectious rhinitis in adults; (ii) a randomized clinical trial comparing transfusions of plasma rich in anti-HIV1 versus transfusions of seronegative plasma in the prevention of opportunist infections. To analyse recurrent events, usual methods are often irrelevant and the marginal model allows use of all the available information to accurately estimate the relative risk of recurrences. Moreover, it enables the estimation of the relative risk for each rank of recurrence.