Aspects of patient recruitment in clinical trials with continuous outcomes: A simulation study
Description
Recent attention has focused on the performance of sample selection techniques employed in clinical research. In studies having a specified, yet arbitrary, target population, one objective of recruiting is to limit enrollment to members of this group. However, when eligibility is determined by observations of a continuous response variable, ineligible patients may be enrolled due to the misclassification resulting from truncated sampling. It is believed that these screening failures will have a negative impact on a study although the severity of the effect is unknown. This study examined the potential effects of truncated sampling under a variety of different situations Adopting a clinical model provided by cardiovascular research, computer simulations examined the potential effects of five experimental factors on clinical trials. Outcomes of interest included the true sample mean, probability of misclassification, cost of recruiting and power of ANOVA to detect a time by treatment interaction (which is known to exist). Simulation results demonstrated that sample characteristics were dependent upon the population variance structure, selection rule, and truncation limit. In addition, power was influenced by the truncation limit employed and size of the study sample. These latter effects were compared to evaluate the feasibility of employing the truncation limit as a design variable It was concluded that the conventional sample selection rule employing multiple qualifying observations performs best in terms of power. However, if safety and/or cost considerations are important, a modified plan requiring fewer screening observations provides a reasonable alternative. Based on the results of these simulations, the specification of truncation limits as a means of controlling power could not be recommended