Design and implementation of resuscitation research: special challenges and potential solutions

Abstract

Evaluation of the effectiveness of resuscitation interventions is challenging. We describe these challenges, which include design, enrolment and analysis issues. Randomized trials establish if interventions work in predefined populations. "Efficacy" trials determine whether interventions work under ideal conditions. "Effectiveness" trials determine whether interventions work under usual practice conditions. These trials represent a trade-off between internal validity versus external validity. Randomized trials use random allocation of participants to interventions to produce study groups that are similar with respect to known and unknown risk factors, reduce bias in the allocation of participants, and assure that statistical tests have valid significance levels. In the emergency setting, there is a risk that treatment offered to control patients will be contaminated by providers' experiences of applying the intervention to patients receiving the experimental intervention. Frequently there is not time to obtain consent from a patient in an emergency setting. Exception from consent can be applied if certain conditions are met. Enrolment in a research study must be initiated quickly in an emergency setting or the patient will die or become disabled. In any trial, data can be used to explore different aspects of response to treatment: multiple treatments, subgroups, events; and interim analyses. We propose solutions to these challenges to help potential investigators through the myriad of difficulties in initiating trials in a complex environment. Design of simple trials that have adequate power enhances their external validity. Allocating groups of episodes to interventions by randomizing by clusters, rather than by individual patients reduces provider noncompliance. Waiver from consent for emergency research and use of novel technologies could facilitate enrolment despite time constraints. Rigorous statistical methods can be used to analyze multiple data without an excessive increase in the chance of a false-positive result.