Non-parametric methods for analysing recurrent complications of varying severity

Non-parametric methods have recently been proposed to take into account the recurrent nature of complications in clinical trials, by estimating the prevalence Q(t), defined as the probability of relapse-free patients being in a complication state at time t. Recoveries are also considered. These methods have been adapted in this paper to take into consideration not only the recurrent nature of complications but also their severity through the interpretation of average weight functions. Results of a randomized trial comparing late complications following two alternative doses of pre-operative internal brachytherapy in patients with localized stage cervix carcinoma are used to illustrate the methods.