gms | German Medical Science

The use of an internal pilot study for blinded sample size reestimation allows to reduce uncertainty on the appropriate sample size compared to conventional fixed sample size designs. Recently blinded sample size reestimation procedures for recurrent event data were proposed and investigated [1, 2, 3]. These approaches assume treatment-specific constant event rates which might not always be appropriate as found in relapsing multiple sclerosis (RMS) [4]. Based on a proportional intensity frailty model we propose methods for blinded sample size reestimation in situations where a time trend of the event rates is present. For the sample size planning and the final analysis standard negative binomial methods can be used, as long as the patient follow-up time is approximately equal in the treatment groups. To reestimate the sample size at interim, however, a full likelihood analysis is necessary. Operating characteristics such as rejection probabilities and sample size distribution are evaluated in a simulation study motivated by a systematic review in RMS [4]. In a second step the model is extended by considering time-dependent treatment effects. The proposed procedures for blinded sample size reestimation control the type I error rate and maintains the desired power against misspecifications of the nuisance parameters.