Author + information
- Received August 31, 2012
- Revision received October 18, 2012
- Accepted October 23, 2012
- Published online February 19, 2013.
- Seth S. Martin, MD⁎,†,⁎ (, )
- Fang-Shu Ou, MS‡,
- L. Kristin Newby, MD, MHS†,
- Victoria Sutton, CCRP†,
- Patricia Adams, RN, CCRC†,
- G. Michael Felker, MD, MHS† and
- Tracy Y. Wang, MD, MHS, MS†
- ↵⁎Reprint requests and correspondence:
Dr. Seth S. Martin, Division of Cardiology, Johns Hopkins Hospital, 600 North Wolfe Street, Carnegie 565-G, Baltimore, Maryland 21287
Objectives The purpose of this study was to quantitatively examine the association of patient- and trial-specific factors with participation in cardiovascular randomized clinical trials.
Background Randomized clinical trials are central to evidenced-based medicine, but low patient participation rates and potentially modifiable barriers are not well understood.
Methods At a large U.S. academic health system, we examined screening logs from December 1, 2005, to February 28, 2011, from 15 cardiovascular randomized clinical trials. We identified 655 patients who were screened and potentially eligible for participation in at least 1 trial. We used multivariable Poisson regression to quantify the risk of not participating in a trial associated with patient- and trial-specific factors.
Results The median age was 63 years (interquartile range: 54 to 72), 35% were women, and the median Charlson Index was 2 (interquartile range: 1 to 5). Forty-two percent of patients did not participate in a trial. In multivariable regression (C-Index 0.85), trial-specific factors strongly associated with not participating included intensive trial-related testing (relative risk [RR]: 1.89; 95% confidence interval [CI]: 1.63 to 2.20) and anticipated trial participation >6 months (RR: 4.10; 95% CI: 2.30 to 7.29). Patient-specific factors associated with not participating included older age (RR: 1.23; 95% CI: 1.11 to 1.36, per 10-year increase if age ≥65 years), out-of-state residence (RR: 1.26; 95% CI: 1.04 to 1.54), and female sex (RR: 1.17; 95% CI: 1.01 to 1.35). Race was not associated with participation.
Conclusions While patient-specific factors were associated with not participating in cardiovascular trials, longer trial duration and intensive trial-related testing were most strongly associated with risk for patients not participating. Innovative trial designs fostering convenience may most enhance trial participation.
Randomized clinical trials (RCTs) are central to evidence-based medicine. Successful RCTs require efficient recruitment of adequately sized study populations that are representative of contemporary clinical practice. However, patient recruitment is a widespread challenge across RCTs, particularly among cardiovascular RCTs conducted in the United States, and there is increasing reliance on enrollment abroad (1–4). Beyond low overall enrollment, under-representation of certain groups of patients from the community threatens the generalizability of RCTs (5–14).
The developing crisis in RCT participation prompted a recent National Heart, Blood, and Lung Institute workshop on strategies for recruitment, from which several themes emerged, including the role of health care professionals as gatekeepers (15). Complementing these important initiatives, further investigation is needed to identify the most important barriers to patient participation in RCTs that can be potentially modified in future RCT designs and recruitment efforts. We quantitatively examined the association of patient- and trial-specific factors with patient participation in RCTs in a consecutive sample of patients who were screened and potentially eligible for participation in at least 1 cardiovascular RCT at a tertiary medical center.
Study population and data collection
We examined screening logs from December 1, 2005, to February 28, 2011, at Duke University Medical Center (Durham, North Carolina), a high-volume tertiary academic health system in the southeastern United States. Screening logs of patients considered for RCT participation were routinely maintained to monitor for bias in selection of subjects for a RCT by investigators, the institution, and sponsors. Screening logs contained data on patients' sociodemographics, RCT eligibilities, and enrollment outcomes. If a patient considered for participation did not enroll, the primary reason for decline was documented as specifically as possible. Where possible, reasons were determined directly by patient report, while other reasons, including altered mental status, clinical instability, concern for nonadherence or substance abuse, and provider decline were based on discussion with clinical providers, review of the medical chart, and/or interview with the patient or the patient's family.
Screening logs were available for 15 ongoing or completed RCTs (16–30). Table 1 summarizes the designs of each of these trials and displays study features that were typically discussed with potential participants during informed consent. The targeted populations in these mostly phase III and IV RCTs reflected a range of clinical diagnoses, including systolic heart failure, heart failure with preserved ejection fraction, acute coronary syndromes, stable coronary disease, and patients with cardiovascular risk factors.
We identified 667 patients who were considered potentially eligible for participation in at least 1 cardiovascular RCT after an initial screen against protocol-defined medical inclusion and exclusion criteria. Some RCTs explicitly identified an inability to follow the protocol as an exclusion criterion; however, this assessment may entail subjective determination, its importance varies based on trial follow-up requirements, and it is not well understood to what extent this type of exclusion contributes to participation outcomes. Therefore, we specifically captured information on this group, but did not consider this a reason for exclusion from the primary analyses of our study. We did exclude 12 patients with a language barrier because these patients were systematically excluded from RCTs at our institution due to the lack of availability of bilingual trial personnel as well as native language consent forms and other trial materials. Therefore, 655 patients were included in our primary analysis sample. The Duke University institutional review board approved the study and waived patient consent.
We leveraged Duke's Enterprise Data Unified Content Explorer (DEDUCE), a Web-based data query tool, to standardize and streamline electronic medical chart review for the sociodemographic and clinical patient factors displayed in Table 2. Three variables (language, insurance, employment) were not available and were extracted in a similar electronic manner from Duke's Decision Support Repository. After this step, language data were not available for 71 patients; therefore, we performed a manual review of the medical charts and screening logs to assess for evidence of limited English proficiency.
Several variables were derived from patients' home postal addresses. We derived driving distance and driving time by referencing patients' home addresses to the address of Duke University Medical Center using CDX Technologies ZipStream (Randolph, New Jersey) integrated with Microsoft MapPoint and Excel (Redmond, Washington). Crosscheck with Google maps (Mountain View, California) showed similar driving distances and times. Based on patients' home ZIP codes, we assigned population representative values for educational attainment, median income, and rural versus urban residence using U.S. Census Bureau 2000 data (31).
Our main outcome measure, namely, not participating in a RCT, was defined as a patient not enrolling in any of the RCTs for which he or she was screened. Definitions of patient- and trial-specific factors considered in our analyses are included in the Online Methods.
We stratified the primary analysis sample into those who participated and those who did not participate in a RCT. Demographic, clinical, and trial characteristics were compared across groups. Continuous variables are presented as median (interquartile range). Categorical variables are expressed as frequency percentages. Univariable analysis was performed using the Wilcoxon rank-sum test and Fisher's exact test for continuous and categorical variables, respectively.
Our main outcome measure occurred commonly; therefore, we used a modified Poisson regression model with a robust error variance (32) to identify the factors that were independently associated with not participating. Relative risks with 95% confidence intervals were calculated to examine the adjusted associations. Variables were selected for multivariable analysis based on clinically judged a priori importance and based on differences between the groups in univariable comparisons. Patient-specific variables in the model were age, sex, race, out-of-state residence, hospital admission, Charlson Index, and Duke cardiologist. Trial-specific variables were intensive trial-related testing, anticipated trial participation >6 months, and trial sponsor (industry versus government). The relationship between age and not participating was nonlinear; therefore, we used linear spline transformation with 1 knot at age 65 years, and the associations of increases in age with not participating were interpreted before and after 65 years of age. For modeling, Charlson Index scores were categorized into tertiles (0 to 1, 2 to 3, and 4 to 15). No covariate data were missing in the multivariable analysis.
We performed tests for interaction between age and sex on the basis of a prior report of such an interaction in heart failure trial participation (6). We also performed tests for interaction between independent risk factors with chi-square values >10. To facilitate the interpretability of the results, age was considered as a dichotomous variable (≥65 years old vs. <65 years old) in the interaction analyses. We also performed a sensitivity analysis in which we excluded 126 patients who were perceived by the trial team to have barriers to participation after reviewing the medical chart, speaking with the clinical providers, or interviewing the patient or the patient's family. This included patients with altered mental status, clinical instability, a history of or perceived risk for nonadherence, substance abuse, or cases in which the provider declined. In these cases, there was concern that the patients may have been unable to provide informed consent or were unable to follow trial protocols, and decisions regarding participation were largely made by persons other than the patient. We also performed a sensitivity analysis in which we excluded the 139 patients who were considered for multiple trials to assess whether our results were dependent on the coding scheme used for patients who were screened for multiple trials.
Statistical analyses were performed using SAS version 9.2 (SAS Institute, Cary, North Carolina). All p values are 2-sided with an alpha level of 0.05. No adjustments were made for multiple comparisons.
In the primary analysis sample, the median age was 63 years (interquartile range: 54 years to 72 years), 35% were women, and the median Charlson Index was 2 (interquartile range: 1 to 5). Patients who did not participate in a RCT represented 42% of the sample. Compared with RCT participants, those not participating were older and more commonly women, but similar in race and ethnicity (Table 2). Clinically, nonparticipants had a modestly lower comorbidity burden, and similar rates of any broadly defined mental health diagnosis and hospital admissions in the prior year. Participants had more frequent outpatient medical contact (overall and cardiology) at the study institution relative to those who did not participate. There were no significant differences in marital status, religion, insurance status, employment, and ZIP code assigned educational attainment, income levels, or urban living.
Persons who did not participate were more likely to reside out-of-state, but had similar overall driving time and distance compared with persons who participated. Specifically examining the subgroup of patients who did not participate and stated that travel distance was the primary reason for not participating (n = 22), median driving distance was 143 miles (interquartile range: 82 to 241 miles), and median driving time was 196 min (interquartile range: 150 to 277 min).
Table 3 shows the prevalence of trial-specific factors stratified by participation status. More patients who did not participate were considered for trials with intensive trial-related testing, an anticipated participation of >6 months, industry sponsorship, and size of <1,000 participants. A smaller proportion of patients who did not participate were considered for trials that offered <$20 in total compensation or recruited in the outpatient setting.
Factors independently associated with not participating in a RCT
In multivariable regression, intensive trial-related testing, longer trial length, and older age after 65 years were the 3 factors most strongly associated with not participating in a RCT (Fig. 1). Out-of-state residence and female sex were also independent factors. There was no independent association of participation status with industry sponsorship, Charlson Index, number of admissions, Duke cardiologist, race, or increased age before age 65 years. The C-Index for the model was 0.85.
Reasons for not participating
For the 278 patients who did not participate in a RCT, the primary reasons are categorized in Figure 2. The 4 most frequently used individual categories were unspecified reason for decline of participation, clinical instability, a history of or perceived risk for nonadherence, and patient-perceived travel barriers (distance or transportation), accounting for approximately two-thirds of the reasons for not participating. Travel barriers were especially common in patients who were considered for lengthier trials and among the elderly (Online Results).
Sensitivity, interaction, and subgroup analyses (Online Results) were largely consistent with our primary results. In sensitivity analyses (Online Figs. 1 and 2), the 3 factors most strongly associated with not participating in a RCT remained intensive trial-related testing, longer trial length, and older age after 65 years.
Trial-specific factors were more strongly associated with not participating in RCTs than patient-specific factors in our study, which is unique in the granularity of patient- and trial-specific factors simultaneously considered. Two trial-specific factors, intensive trial-related testing and an anticipated participation of >6 months, were most strongly linked with lack of participation. Patient-specific factors that were independently associated with not participating included older age after 65 years, female sex, and out-of-state residence.
Under-participation in RCTs: a longstanding systemic challenge in the United States
Between 1997 and 2009, U.S. federally funded RCTs of coronary artery disease populations required ∼50% enrollment abroad (3). The United States is not alone in its recruitment problems: among 114 trials in the United Kingdom conducted between 1994 and 2002, >2 in 3 failed to recruit the target sample size within the planned time period (2). Slow or incomplete enrollment can undermine trial completion, statistical power, and the timely availability of results to guide clinical practice, and adds expense, limiting funding for other important trials. These challenges also decrease site and sponsor interest in future RCTs.
Beyond impeding trial operations, under-representation of certain community patient groups in RCTs challenges the generalizability of the trial results. We are approaching the 2-decade mark since the National Institutes of Health Revitalization Act called for greater representation of women and minorities in RCTs (5). Under-representation of women has been shown in a number of studies (5–13), with women comprising only ∼27% on average of the participants in cardiovascular RCTs (9,13) and National Institutes of Health initiatives failing thus far to improve on this (9). In our study, female sex was independently associated with not participating in cardiovascular RCTs, yet interestingly, race was not. Although the elderly were not a focus group of the Revitalization Act, or national guidelines on inclusiveness in clinical trials that followed in 1994 (11), we found that older age was a risk factor for not participating in a RCT, independent of comorbidities. Patients with limited English proficiency represent another vulnerable group that was systematically excluded from RCT participation in our study because of the lack of necessary resources for native language communication. There is wide variability across institutions in the availability of such resources, although a case can be made for greater resource allocation to support consistent inclusion of this growing segment of the population in clinical research (14).
Patient factors and provider perceptions as barriers to RCT participation
Among patient-reported reasons for not participating in RCTs, inconvenience and inability to complete study requirements are common themes. In targeted interviews conducted during an earlier study from our institution (33), 27% of patients cited inconvenience as their primary reason for deciding against RCT enrollment. In our study, travel distance and difficulty arranging transportation clustered in patients who were considered for longer trials and in the elderly. The inconvenience of in-person visits likely reaches beyond travel distance and transportation arrangements alone; for example, in women especially, there may be the need to arrange child care to attend a study visit (15).
Many patients in our study did not provide specific reasons for declining RCT participation, suggesting that patients are often not transparent in their attitudes toward RCTs in practice. Patient fear of random assignment to treatment (decision made by “the computer”) and research risks, engrained with elements of mistrust, present formidable challenges in RCT recruitment (33). In women, particularly, the perceived risk of harm from research has been linked to reluctance to participate in RCTs (34,35). Given that public exposure to research may be skewed by sensational media reports of error or duplicity, a continuing mission of the research community is to provide a more balanced perspective by raising awareness regarding the robust mechanisms in place to protect human research subjects and ensuring that they are implemented.
In many cases, trial team and provider perceptions of a patient may play a primary role in the participation outcome, including perceptions regarding adherence, substance use, clinical status, or other factors, which together accounted for 45% of patients who did not participate in a RCT in our sample. Limiting enrollment of such patients, who may be less likely to adhere to the study protocol, is an important consideration in many RCTs. However, it is important to note that barriers to RCT participation perceived by the trial team and providers are often based on overall impressions rather than formal criteria, and their importance varies based on trial follow-up requirements. Ultimately, targeting these issues in future RCTs may promote more balanced representation of patient subgroups and strengthen the efforts of the American College of Cardiology's Coalition to Reduce Disparities in Cardiovascular Care and Outcomes (CREDO) group aimed at achieving equitable care and outcomes for all patients, regardless of race, ethnicity, sex, and age (36).
Implications for future clinical trial design and conduct
Given our results, it is important for the clinical trial community to consider whether a study question could be reliably answered with less intensive testing or through a study of shorter duration. If this is not possible, minimizing inconvenient in-person visits could potentially improve participation. For example, a recently launched longitudinal study is using centralized follow-up telephone interviews instead of multiple patient trips to the research center (37). In the modern era, creative follow-up solutions can be explored; for example, patients may directly interact via Internet and mobile technologies to efficiently and conveniently provide follow-up information.
Practical solutions to trial design and conduct that integrate the research enterprise with clinical care and medical education will likely be most successful. For instance, using trial activities as a venue to provide clinical information to patients, providers, and community members enhances health care quality in real time (36). Informatics systems that integrate trial data with existing electronic medical records could augment the availability of patient information at the point of care while increasing the efficiency of research data collection. Such integration may allow for efficient tracking of patient status and reassessment of trial candidacy after clinical stabilization, as well as more objective identification of likely nonadherent patients (e.g., no-show rate for outpatient appointments). A further advantage of informatics systems that bridge the clinical and research enterprises is the increased ability to expose and engage physicians early in their medical education to clinical research. In appropriate circumstances, research participation might become part of routine care (38) or randomization may be clustered at the site level (39) to alleviate the burden of additional work associated with informed consent.
First, our study setting was a single academic medical center in the southeastern United States with a high volume of cardiovascular RCT research, and we explored only participation in cardiovascular RCTs. Therefore, our results may not be generalizable to other institutions in other locations in the United States or abroad or to participation in RCTs in other therapeutic areas. Second, all patients in our study had access to a RCT; we did not address challenges to enrollment at the provider level (i.e., providers as gatekeepers). Third, this was an observational study, precluding direct assessment of causality, and unmeasured confounders are possible in this observational analysis. Fourth, regarding variables themselves, educational attainment and income were based on population representative levels according to ZIP code as opposed to direct report by patients.
Among patients presenting to a high-volume academic medical center in the southeastern United States who were considered potentially eligible for RCT participation after an initial screen against medical eligibility criteria, we identified multiple potentially actionable barriers to participation. While age, sex, and state of residence were associated with not participating, longer trial duration and intensive testing were the factors most significantly related to RCT participation. One major implication of our study is the need for innovative RCT designs using integrative approaches and emerging technologies to foster convenience and thereby facilitate greater RCT participation.
The authors thank Duke's Cardiovascular Research Study Coordinators, including Carolyne Rotich, Chandra McNeill, and Bernadette Druken, and Duke's Decision Support Repository Team for helping compile the data for this study.
For supplemental Methods, Results, and References section, as well as a figure, please see the online version of this article.
This study was an investigator-initiated, internally-supported project. Dr. Newby has received research funding from Amylin, AstraZeneca, Bristol-Myers Squibb, diaDexus, Eli Lily & Company, GlaxoSmithKline, Merck, MURDOCK Study, National Heart, Lung, and Blood Institute, Regado Biosciences, and F. Hoffmann-La Roche, Ltd.; and compensation for consultancy from Amgen, AstraZeneca, Daiichi Sankyo, Eli Lily & Company, Genentech, Johnson & Johnson, and Novartis. Dr. Felker has received research funding from Amgen, BG Medicine, Cytokinetics, Johnson & Johnson, Roche, and Otsuka; and compensation for consultancy and lectures from Amgen, Cytokinetics, Novartis, Roche, Otsuka, and Medpace. Dr. Wang has received research funding from AstraZeneca, Bristol-Myers Squibb, Heartscape Technologies, Lilly, Sanofi-Aventis, Schering-Plough, and The Medicines Company; and compensation for consultancy and lectures from the American College of Cardiology, AstraZeneca, and Medco. All other authors have reported they have no relationships relevant to the contents of this paper to disclose.
- Abbreviation and Acronym
- randomized clinical trial
- Received August 31, 2012.
- Revision received October 18, 2012.
- Accepted October 23, 2012.
- American College of Cardiology Foundation
- Califf R.M.,
- Harrington R.A.
- Kim E.S.,
- Carrigan T.P.,
- Menon V.
- ↵National Institutes of Health Revitalization Act of 1993. Public Law 103-43. Subtitle B-Clinical Research Equity Regarding Women and Minorities. SEC.131. http://history.nih.gov/research/downloads/PL103-43.pdf. Accessed December 21, 2012.
- Kim E.S.,
- Carrigan T.P.,
- Menon V.
- Melloni C.,
- Berger J.S.,
- Wang T.Y.,
- et al.
- Department of Health and Human Services, National Institutes of Health
- Johnson S.M.,
- Karvonen C.A.,
- Phelps C.L.,
- Nader S.,
- Sanborn B.M.
- Efficacy and Safety of Aliskiren and Aliskiren/Enalapril Combination on Morbi-mortality in Patients With Chronic Heart Failure (ATMOSPHERE). http://clinicaltrials.gov/ct2/show/nct00853658. Accessed June 25, 2012.
- Effectiveness of Ultrafiltration in Treating People With Acute Decompensated Heart Failure and Cardiorenal Syndrome (CARRESS). http://clinicaltrials.gov/ct2/show/nct00608491. Accessed June 25, 2012.
- Using Allopurinol to Relieve Symptoms in Patients With Heart Failure and High Uric Acid Levels (EXACT-HF). http://clinicaltrials.gov/ct2/show/nct00987415. Accessed June 25, 2012.
- Cannon C.P.,
- Giugliano R.P.,
- Blazing M.A.,
- et al.
- Evaluating the Effectiveness of Sildenafil at Improving Health Outcomes and Exercise Ability in People With Diastolic Heart Failure (RELAX). http://clinicaltrials.gov/ct2/show/nct00763867. Accessed June 25, 2012.
- Renal Optimization Strategies Evaluation in Acute Heart Failure (ROSE-AHF). http://clinicaltrials.gov/ct2/show/nct01132846. Accessed June 25, 2012.
- A Study to Evaluate the Safety of 12 Weeks of Dosing With GW856553 and Its Effects on Inflammatory Markers, Infarct Size, and Cardiac Function in Subjects With Myocardial Infarction Without ST-segment Elevation (SOLSTICE). http://clinicaltrials.gov/ct2/show/nct00910962. Accessed June 25, 2012.
- The Stabilization of Atherosclerotic Plaque by Initiation of Darapladib Therapy Trial (STABILITY). http://clinicaltrials.gov/ct2/show/nct00799903. Accessed June 25, 2012.
- Trial to Assess the Ocular Safety of SCH 530348 in Patients With Atherosclerosis (TRA-Ocular). http://clinicaltrials.gov/ct2/show/nct00617123. Accessed June 25, 2012.
- U.S. Census Bureau
- Zou G.
- Sen Biswas M.,
- Newby L.K.,
- Bastian L.A.,
- Peterson E.D.,
- Sugarman J.
- Yancy C.W.,
- Wang T.Y.,
- Ventura H.O.,
- et al.,
- CREDO Advisory Group
- Chin C.T.,
- Wang T.Y.,
- Anstrom K.J.,
- et al.
- Blankenship J.C.,
- Scott T.D.,
- Skelding K.A.,
- et al.
- Alexander K.P.,
- Wang T.Y.,
- Li S.,
- et al.