Impact of selection criteria on recruitment in an interventional stroke trial

Randomized controlled clinical trials are the gold standard for scientific evaluation of clinical diagnostic and treatment concepts. Frequently, recruitment of participants is slower than expected, especially in acute conditions with a short time frame for inclusion. Simple prediction models have been proposed to extrapolate recruitment rates. We hypothesized a significant overestimation of recruitment when ignoring interdependence of selection criteria, leading to an insufficient representation of reality by available models. We proposed that slight modifications to inclusion criteria might augment recruitment without causing selection bias.

We analyzed recruitment in an acute intervention trial of acute ischemic stroke initiated by our facility. Frequencies of selection criteria were recorded and analyzed individually as well as cumulatively. We then amended the trial protocol by moderate modifications to the selection criteria. The main outcome criterion was the rate of recruited over screened patients, with the goal of increasing recruitment fourfold without adding unacceptable selection bias. A previously presented prediction model was applied to our trial and compared with actual recruitment. Data were compared between screening periods at recruitment prior to and after the implementation of the amendments.

The impact of typical as well as novel inclusion criteria such as age limits, imaging-based definition of pathology, time between onset and presentation as well as inability to consent were quantified. Age restriction, definition of index event and late arrival after ictus were identified as the most challenging modifiable selection criteria. Amending those criteria increased recruitment by a factor of 4.1. Inability to consent was a significant exclusion criterion gaining impact with the target population. The selection criteria had a cumulative rather than separate recruitment-limiting impact. A previously presented model did not predict recruitment sufficiently.

We describe frequencies of selection criteria in a typical cohort of patients suffering from acute cerebrovascular events, and their cumulative impact. These data may help to better understand recruitment limitations and allow designing future trials more effectively. Ability to consent especially is a major contributor to trial exclusion, strongly interfering with the targeted trial population of ischemic stroke. Tentative prescreening phases before site or trial initiation should be considered. No predictive statistical models of recruitment have been established so far.