Can Response-Adaptive Randomization Increase Participation in Acute Stroke Trials?
Background and Purpose—A response-adaptive randomization (RAR) trial design actively adjusts the ratio of participants assigned to each trial arm, favoring the better performing treatment by using outcome data from participants already in the trial. Compared with a standard clinical trial, an RAR study design has the potential to improve patient participation in acute stroke trials.
Methods—This cross-sectional randomized survey included adult emergency department patients, age ≥18, without symptoms of stroke or other critical illness. A standardized protocol was used, and subjects were randomized to either an RAR or standard hypothetical acute stroke trial. After viewing the video describing the hypothetical trial (http://youtu.be/cKIWduCaPZc), reviewing the consent form, and having questions answered, subjects indicated whether they would consent to the trial. A multivariable logistic regression model was fitted to estimate the impact of RAR while controlling for demographic factors and patient understanding of the design.
Results—A total of 418 subjects (210 standard and 208 RAR) were enrolled. All baseline characteristics were balanced between groups. There was significantly higher participation in the RAR trial (67.3%) versus the standard trial (54.5%), absolute increase: 12.8% (95% confidence interval, 3.7–22.2). The RAR group had a higher odds ratio of agreeing to research (odds ratio, 1.89; 95% confidence interval, 1.2–2.9) while adjusting for patient level factors. Trial designs were generally well understood by the participants.
Conclusions—The hypothetical RAR trial attracted more research participation than standard randomization. RAR has the potential to increase recruitment and offer benefit to future trial participants.
In time-sensitive emergency conditions, participation in research is limited. Medical care should aim to provide the best possible care for that individual; however, it is important to balance this with research goals of gathering unbiased data regarding the effect of treatment. In standard 2-arm clinical trial designs, each participant has an equal but random chance of receiving either treatment. Response-adaptive randomization (RAR) is one way to address the tension between the medical and research aims.1 In a trial using RAR, the ratio of participants assigned to each study group is adjusted based on accumulating data while the study is ongoing, using a predetermined defined set of rules. This works to collectively favor the patients within the trial in situations when one treatment is ultimately better than the other.
There is limited knowledge on the extent of use and effectiveness of RAR study designs in the emergency setting. However, some studies have assessed willingness to join research studies in emergency conditions such as ischemic stroke or subarachnoid hemorrhage. These studies have shown that just more than half of participants or their proxies consent to research.2,3 Hesitation to join emergency research studies was attributed to the perceived risk of such trials and pre-existing negative attitudes toward research.3
Participation in emergency care research may be unattractive to a significant proportion of patients. The influence of trial design on research participation in emergency care has not been studied previously. Therefore, we hypothesized that a hypothetical acute stroke trial that included RAR would be more agreeable to participants than a trial using fixed 1:1 randomization, with all other aspects of the trial design presented exactly the same.
A more detailed description of the methods is available in the online-only Data Supplement. Briefly, we performed a cross-sectional study of noncritically ill emergency department adult patients, without presenting symptoms consistent with stroke, altered mental status, or alcohol intoxication. Participants were introduced to the study, gave consent, and were randomly allocated to see 1 of 2 videos. They also answered questions about demographics and stroke symptom knowledge. The video was the same across both groups, with the exception of the explanation of the hypothetical study: either described as a standard clinical trial or an RAR study. The RAR video can be viewed at http://youtu.be/cKIWduCaPZc; the standard trial video can be viewed at http://youtu.be/SrI4FdCTZ-A. All participants in both groups were informed that the trial had recruited approximately one half of the total planned enrollment and then were asked if they would participate in the stroke trial (primary outcome). Statistical analysis was performed using SPSS version 19. The entire protocol was placed online prior to the analysis or visualization of the data (http://bit.ly/11gTfLU). All hypotheses and main analyses were prespecified prior to any visualization of participant responses. We conducted stratified analyses based on participant understanding based on post hoc review of the results.
This study was determined to be exempt by University of Michigan Institutional Review Board under 45 CFR 46.101(b). Participation was voluntary, and verbal consent was obtained from all participants.
Four hundred eighteen participants were enrolled in the study, 208 to the RAR group and 210 to the standard group by randomization. Age, sex, history of stroke, hypertension, diabetes mellitus, atrial fibrillation, heart attack, education, ethnicity, and previous knowledge of stroke were comparable across groups (Table I in the online-only Data Supplement).
When patients were presented with the hypothetical acute stroke study, 140 of the 208 (67.3%) in the RAR group chose to participate in the study, versus 114 of the 210 (54.3%) of those in the standard group; an absolute difference of 13% (95% confidence interval, 3.7–22.1).
Self-reported understanding between the standard and RAR groups was not significantly different; however, significantly fewer in the RAR group actually correctly identified the method of trial allocation (Table II in the online-only Data Supplement).
In the multivariable logistic regression model, the RAR group had a higher odds of agreeing to research (odds ratio, 1.89; 95% confidence interval, 1.2–2.9) while controlling for age, sex, ethnicity, education, self-reported understanding of protocol, ability to identify allocation technique correctly, and stroke awareness (Table III in the online-only Data Supplement). In stratified analyses, those indicating complete understanding of the protocol and who correctly identified the allocation method were more likely to agree to research if in the RAR group (Tables IV–VI in the online-only Data Supplement).
In this study, we found significant higher participation in the hypothetical acute stroke trial when an RAR design was used compared with the standard clinical trial design. In certain respects, this could be expected as the potential for better outcome is posed as greater (although still subject to random chance) in the RAR scenario. This suggests that the design and potential benefits of the RAR feature were able to attract and recruit more participants to join the research study in an acute stroke trial scenario. Importantly, the participation rate of the standard group was comparable to that of other studies,2,3 indicating that the conditions of our hypothetical scenario were likely to be comparable to actual experiences in acute stroke trials.
Previous work has demonstrated that most clinical research participants are uncomfortable with treatment allocation determined by random chance.4 A trial design with RAR has the potential to be a viable alternative because it preserves the random element and a participant’s chance of getting an ultimately better treatment is increased. Other trial goals may also be incorporated via similar designs, such as balance of prognostic factors using covariate adjusted response-adaptive allocation.5,6
Our study has several limitations. First, the participants in the study had a mean age of 43. Although the presentation of strokes is typically at an older age, it has been shown that the proportion of patients declining acute care research (either by patient or proxy) was similar in the adult (18–64) and geriatric age (>65) groups, supporting the generalizability of our results with regard to age.7 Second, the fact that the scenarios presented to the patients were hypothetical is another limitation. In the hypothetical acute stroke scenario used in our study, the patient would not be able to answer for themselves because the symptoms of the stroke would have left the patient cognitively impaired. However, given that our study showed a significant number were in favor of RAR over the standard randomization, it is likely that this preference would extend to the surrogates as well under the 2 likely situations: (1) the surrogate is expected to select the treatment that the patient would have wanted or (2) existing knowledge that surrogates tend to make decisions for care that they themselves would choose.3,8 Finally, the generalizability of the study has its limitations. The study population was from a single suburban academic center, and thus, the results may not be generalizable to other regions, countries, or hospital types. Also, we only explored 1 type of RAR design, the play-the-winner design, and it is unknown whether our findings would be applicable to other adaptive designs incorporating RAR such as multiple arm dose finding trials.
In summary, our results show that the RAR trial design attracted a higher research participation rate than standard randomization for a hypothetical acute stroke trial. Implementing the RAR trial design could increase recruitment and also offer the overall trial population added benefit compared with a fixed randomization trial. Future studies aimed at learning more about why RAR was preferred, or what specific components of RAR were viewed favorable, would be valuable in gaining a better understanding of RAR and allow for optimizing its implementation. In addition, improving the rapid communication of adaptive clinical trial designs in the emergency setting with potential research subjects is an important area for future work.
Dr Meurer had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Sources of Funding
J.S. Tehranisa and Dr Meurer were supported by grants from the University of Michigan Medical School Research Program and National Institutes of Health (5T35HL007690-30; principal investigator: Benjamin Margolis). The National Institutes of Health had no role in the preparation, review, or approval of the article.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.114.005418/-/DC1.
- Received March 12, 2014.
- Revision received May 7, 2014.
- Accepted May 12, 2014.
- © 2014 American Heart Association, Inc.
- Bandyopadhyay U,
- Bhattacharya R
- Glickman SW,
- Anstrom KJ,
- Lin L,
- Chandra A,
- Laskowitz DT,
- Woods CW,
- et al