Difficulties with study recruitment have been linked to study failure (1,2), and it is estimated that half of all studies fail to reach recruitment goals (3,4). Improving the understanding of how study characteristics influence recruitment and retention may allow researchers to design studies that maximize recruitment and retention. By recruiting and retaining more participants, studies will have larger sample sizes for the statistical analyses, which results in larger statistical power (5). Improving study recruitment and retention can also reduce the study budget and make research findings more generalizable because study samples are more likely to be representative of the larger population when more participants are recruited and retained in the study (4,6).
Although recruiting for studies involving patients with breast cancer or physical activity are each difficult, recruiting patients with breast cancer for physical activity studies may be even more difficult. Researchers have generally experienced low success rates when recruiting patients with breast cancer (7–9), with some estimates that as low as 3% of patients with breast cancer enroll to participate in research (7). In a systematic review of exercise interventions for patients with advanced cancer, Sheill et al. (10) found slightly more optimistic estimates with mean recruitment being 49% with study-level recruitment ranging from 28% to 75%. In a physical activity study for women with metastatic breast cancer, Yee et al. also reported promising results by successfully recruiting 93% of participants assessed for eligibility. However, Yee et al. did recommend interpreting their recruitment results with caution, as other studies have reported successfully recruiting only 5%, 30%, and 36% of participants assessed for eligibility (11–14).
Although Sheill et al. (10) and Yee et al. (14) reported promising recruitment results, their methodologies limit the generalizability of their findings. Notably, Sheill et al. (10) and Yee et al. (14) defined recruitment to be the number of participants recruited from the number of participants screened for eligibility. Thus, these reported recruitment percentages are not entirely reflective of the researchers’ entire efforts to recruit participants, as the researchers undoubtedly contacted potential participants who were not screened for eligibility. Therefore, the findings from Sheill et al. (10) and Yee et al. (14) may be considerably higher than the recruitment percentages found in practice, which speaks to the need to better understand how to improve recruitment for the breast cancer population especially in light of other findings that indicate historically poor recruitment (7–9). Outside of individual studies, reviews have examined the accrual rates for patients with breast cancer participating in research (10,15), but little research has focused on how to optimize recruitment for the breast cancer population.
Cancer stage and response to treatment in the breast cancer population may be one reason for low enrollment in studies. In one study, 61% of patients with breast cancer were not offered participation in a clinical trial, and a statistically significant association was found between cancer stage and whether patients were offered participation in a clinical trial (8). For those not offered participation in a clinical trial, the primary reason for 53% of those not offered participation was that no studies were available for the cancer stage, and a secondary reason for 17% of those not offered participation was that these patients were already receiving treatment (8).
Study participation is also affected by several practical factors (e.g., work schedules, transportation, parking, and childcare) that must be addressed to enable successful recruitment (9,16). This is particularly true in studies involving exercise and/or physical activity interventions, which can entail considerable expenditure of time and energy. This may be a significant barrier to participation for individuals with breast cancer, many of whom are actively undergoing treatments that significantly reduce energy levels (17–19).
Once participants are recruited into a study, retention is critical. We conceptualized retention as an extension of recruitment in that the factors influencing recruitment are likely to continue influencing retention. Thus, retention involves collaboration with participants to address issues related to motivation, study requirements, and/or logistical burdens so that these issues do not threaten the participant’s continued participation. However, retention may be vulnerable to external influences beyond those affecting recruitment, such as the death of a family member or a severe case of an acute illness.
In a longitudinal study of adult female survivors of childhood cancer, the authors reported attrition rates of 39% in a longitudinal study (20). In a systematic review of studies using exercise interventions for adults with advanced cancer, Sheill et al. (10) found mean attrition to be 24% with study-level attrition ranging from 10% to 41%, and advancing disease was the most commonly reported reason for attrition.
This review aimed to extend the previous reviews of recruitment by estimating the recruitment rate (RR), recruitment efficiency (RE), and dropout for the breast cancer population. In addition, this review aimed to address the gap in the literature of what factors are associated with recruitment and retention for breast cancer studies involving exercise or physical activity interventions. To do this, we examined the associations between methodological, intervention, participant, and study characteristics and RR, RE, and dropout.
Protocol and Registration
The results of this review are presented in adherence to the PRISMA guidelines for presenting systematic review and meta-analyses (see Supplemental Content 1, http://links.lww.com/TJACSM/A124). The review was registered with the Prospective Register of Systematic Reviews (PROSPERO) with the registration number CRD42017057284. As the protocol has been published (21), a summary of the systematic review methods is provided, except where post hoc amendments were made to the published protocol.
Table 1 provides the eligibility criteria for systematic review. The decision to include only studies that had patients with breast cancer composing more than 50% of the sample was made post hoc. Another post hoc decision was to exclude studies with sample sizes less than 10 from meta-analysis. Recruitment-related studies were defined to mean studies that explicitly address recruitment and/or report recruitment variables as primary outcomes. Exercise-related studies were defined to mean studies that require participants to engage in some sort of exercise or physical activity intervention. The full definitions for these criteria can be found in the published protocol (21).
TABLE 1 -
Inclusion and Exclusion Criteria.
|1. The study was written in English
||1. The study used a cross-sectional design
|2. The study used humans as the research subjects
||2. The publication was a review article
|3. The study recruited adult subjects
|4. The study explicitly addressed recruitment
|5. The study implemented an exercise-related intervention
|6. The study was peer reviewed
|7. >50% of the study sample were current or previous patients with breast cancer
PubMed, Cumulative Index to Nursing and Allied Health Literature (CINAHL), and Online Resource for Recruitment research in Clinical Trials (ORRCA) databases were originally searched from inception through February 10, 2017, and this search was updated to include studies published between February 11, 2017, and March 15, 2019.
The search string for CINAHL was “Exercise AND Recruitment NOT ‘musc* recruit*,’” where * indicates truncation, and this was representative of the search strings for PubMed and ORRCA. The full search strategies can be found in the online supplementary file of the published protocol (21).
The authors abstract screened 100 random selected studies from the search results and resolved any disagreements to establish a reliable abstract screening procedure. Once the reliable abstract screening procedure was established, the remainder of the search results was screened according to the systematic review eligibility criteria presented above. Screening for meta-analysis eligibility was performed during full-text data extraction. The full study selection processes can be found in the published protocol (21).
Data Collection Process
To ensure accuracy of the extracted data, two authors independently extracted the data described below from each study eligible for systematic review, and the extracted data were compared. The two authors resolved any discrepancies between the two extracted data sets. The first author emailed studies’ corresponding authors if primary outcome data were not reported in the included studies. The full data collection processes can be found in the published protocol (21).
Study variables included variables describing study-level factors that support how a study was conducted and what resources are available to the researchers and participants. Dichotomous indicators of whether the study was funded, had multiple sites, or was a feasibility or pilot study were collected. In addition, the amount of compensation for the treatment group and control group (if applicable) was collected.
Intervention variables included variables describing the exercise or physical activity intervention used in the study. The duration of the intervention in weeks, the number of required visits, the hours of intervention received by the treatment group and control group (if applicable), and the type of exercise (e.g., aerobic, resistance, multiple, other) and intervention (e.g., exercise based, behaviorally based [motivational interviewing, exercise counseling], lifestyle based [combination of exercise, education, and nutrition]) were collected. In addition, dichotomous indicators of whether the intervention was home based or supervised were collected.
Participant variables included variables describing the participants of the study. Average age, percentage male, percentage Caucasian, and percentage who completed 12th grade or less of formal education (or comparable for studies outside of the United States) were included.
Design variables included variables describing the research design of the study. Dichotomous indicators of whether studies used random assignment, blinded researchers, blinded participants, monitoring protocols to ensure fidelity to the protocol, use of a true control group, and use of an intention-to-treat design were collected.
RR was calculated by dividing the number of enrolled participants by the number of weeks spent recruiting. RE was calculated by dividing the number of enrolled participants by the number of participants contacted to participate. Dropout was added as a post hoc outcome measure and was calculated by dividing the number of participants who withdrew from the study by the total number of enrolled participants.
Additional recruitment variables included the number of study personnel recruiting, the professional discipline of recruiters, the training provided to recruiters, the hours per week dedicated to recruiting, the recruitment methods, and the financial investment into recruiting. These additional recruitment variables were presented narratively rather than being summarized through meta-analysis.
Risk of Bias in Individual Studies
The Cochrane Collaboration assessment tool was chosen post hoc to assess the within-study and between-study risk of bias for the included studies (22).
Study Quality in Included Studies
We made a post hoc decision to not assess the study quality of included studies. The Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) quality assessment is natural choice for assessing study quality, although the GRADE is focused on specific study outcomes. Because of the nature of our recruitment and retention outcomes, the GRADE is less suitable than risk of bias assessments for this review.
Synthesis of Results
The analyses for synthesizing the results of this review were not planned a priori, given the difficulty of calculating the outcome measures. Consequently, the analyses for synthesizing the results were chosen post hoc.
Meta-analysis of RR was not possible because study-level variance could not be calculated. Therefore, we estimated RR using weighted averages that weight RR based on sample size. The interquartile range was used to examine heterogeneity in RR.
Random-effects meta-analyses with traditional inverse-variance weighting (i.e., V + T2) were used to analyze the reported proportions for RE and dropout (23). Although double arcsine transformations could have been used to prevent skewing of the variance estimates of proportions less than 0.2 or greater than 0.8 (24), we opted against these transformations to preserve the interpretability of the resulting estimates. Because effect sizes can be misleading with sample sizes less than 10, a post hoc decision was made to exclude studies with less than 10 participants from meta-analysis (25). The I2 and T statistics were used to estimate heterogeneity in RE and dropout. The summary statistics were presented as a pooled effect size estimate and a 95% confidence interval along with the I2 and T heterogeneity statistics. The Egger regression test and the trim-fill method were used to assess publication bias in the meta-analysis results (26,27). Although this approach to assessing publication bias could be perceived as overly simplistic, recruitment and retention are unlikely to directly influence by publication bias. Conversely, it is more likely that poor recruitment and/or excessive attrition directly influence publication bias. Thus, we are using a relatively simple approach to assess publication bias, given the somewhat unusual relationship between the outcome variables and the publication bias.
Additional analyses were planned post hoc to explore relationships between the collected data and the outcome measures. Pearson correlations were used to explore relationships between RR and study, intervention, participant, and design variables. Meta-regression analyses were used to explore how study, intervention, participant, and design variables affected RE and dropout. For the meta-regression analyses, each continuous variable was standardized using z-scores, and standardized composite indices representing the study, intervention, participant, and design domains were first used independently to identify impactful variables (see Supplemental Content 2, http://links.lww.com/TJACSM/A128; Supplemental Content 3, http://links.lww.com/TJACSM/A125). In the second step of the meta-regression analyses, all variables from domains with P < 0.15 were entered as univariate predictors into the meta-regression analysis. In the third step of the meta-regression analyses, all variables with P < 0.15 were entered as multivariate predictors into the meta-regression analysis. Missing RE and dropout data were imputed using multiple imputation for the meta-regression analyses when these data could not be obtained by querying authors.
It is possible that whether a study is randomized will have differing levels of recruitment and retention (28). Consequently, a dichotomous variable for whether a study was randomized or not was included as a predictor in the meta-regression analyses to determine whether randomized studies had differing levels of recruitment or retention.
The database search results were managed in EndNote (29). SPSS, version 25, and the tidyverse, metafor, and mice packages in the R software package were used for statistical computations (30–34).
A total of 2,836 studies were identified from database searches. Figure 1 provides the PRISMA chart. Twenty-eight studies with a combined sample of 2,685 participants met inclusion criteria for systematic review (18,35–61), and 27 met inclusion criteria for meta-analysis (18,35–50,52–61).
Studies generally did not provide much information about their recruitment (e.g., how many research team members were recruiting, the training provided to recruiters, how much time per week was dedicated to recruitment activities), except for the method of recruitment, which was reported by 26 studies (see Supplemental Content 4, http://links.lww.com/TJACSM/A127). Six studies reported how many research team members were recruiting participants, with numbers ranging from 1 to 6 (44,46,53,55,59,61). Eleven studies reported the discipline of recruiters, which were primarily nurses, physicians, or nurse practitioners (18,42–44,46,51,53,58–61). No studies reported the training given to recruiters. Only one study reported the number of hours dedicated to recruiting per week, and this study reported dedicating 208 hours per week (59). Five studies reported treatment group compensation, with compensation ranging from $30 to $145 (35,50,52–54).
A full summary breakdown of the study-level information and recruitment variables for all variables was provided in the supplemental content (see Supplemental Content 5, http://links.lww.com/TJACSM/A129). Broadly, the included studies included middle-aged adults, who were largely Caucasian and highly educated. The mean age of participants across all studies was 54.4 ± 8.84 yr. Over 70% of the participants in 11 studies had achieved more than a high school diploma. The study samples were more than 90% Caucasian in 12 studies. In terms of research design, 11 studies implemented a monitoring procedure to ensure intervention protocol fidelity, which was defined as a yes/no indicator of whether studies made concerted efforts to ensure that the exercise or physical activity interventions were administered as intended. Similarly, 10 studies implemented a true control group, which required a randomized design in addition to a comparison group that received no treatment.
Risk of Bias in Included Studies
The included studies generally demonstrated moderate levels of risk of bias across studies. The risk of bias ratings for each included study is included in Figure 2. The proportions of studies in each risk category for the assessed sources of bias are presented as a supplementary file (see Supplemental Content 6, Study risk of bias proportions). The rationale for each risk of bias rating is also presented as a supplementary file (see Supplemental Content 7, http://links.lww.com/TJACSM/A123).
Results of Individual Studies
The RR for each study is provided in Figure 3. Figure 4 provides the forest plot for the RE meta-analysis results, and Figure 5 provides the forest plot for the dropout meta-analysis results. Because of the mathematical properties of RR, variances could not be calculated. Because RE and dropout are proportions, the variance estimates were calculated using the sample sizes (see Supplemental Content 5, http://links.lww.com/TJACSM/A128), and Barendregt et al. (24) described how to calculate the variance of proportions.
Synthesis of Results
The weighted average of RR was 2.6 participants per week. The interquartile range for RR was 1.6.
We obtained the necessary information to calculate RE for 23 of the 27 studies (85%). A random-effects meta-analysis indicated a pooled RE of 0.30 with a 95% confidence interval of [0.24–0.36], which is statistically significant (P < 0.001). Heterogeneity was large with I2 = 94.55, which indicates that approximately 95% of the heterogeneity for RE was true between-study heterogeneity, and T = 0.16, which indicates large absolute heterogeneity as this statistic functions similarly to a standard deviation for the pooled RE. The funnel plot for RE is presented in the supplemental material (see Supplemental Content 8, http://links.lww.com/TJACSM/A121). Both the Egger regression test (z = 1.01, P = 0.31) and the trim-fill method indicated that publication bias was likely not present.
We obtained the necessary information to calculate dropout for 24 (89%) of the 27 included studies. A random-effects meta-analysis indicated a pooled dropout of 0.16 with a 95% confidence interval of [0.11–0.22], which is statistically significant (P < 0.001). Heterogeneity was large with I2 = 95.65, which indicates that approximately 96% of the heterogeneity for dropout was true between-study heterogeneity, and T = 0.13, which indicates large absolute heterogeneity. The funnel plot for dropout is presented in the supplemental material (see Supplemental Content 9, http://links.lww.com/TJACSM/A120).
The Egger regression test indicated that publication bias may be present (z = 3.23, P < 0.01). The trim-fill method noted that two studies with higher dropout are possibly missing. The trim-fill method suggested that the adjusted pooled dropout was 0.18 with a 95% confidence interval of [0.13–0.23] and P < 0.0001, with I2 = 96.66 and T = 0.13.
Correlation analyses indicated that only dropout (r = 0.64, P = 0.003) and race (r = −0.54, P = 0.024) had statistically significant associations with RR. The correlation between dropout and RR indicates that studies with higher dropout tended to recruit more participants per week. The correlation between race and RR indicated that studies with more Caucasians tended to recruit fewer participants per week. The correlations between RR and the other variables are provided in the supplemental material (see Supplemental Content 10, http://links.lww.com/TJACSM/A126).
In the first stage of the analysis, only treatment group compensation was statistically significant (b = −0.06, SE = 0.03, P = 0.048), and using a true control group (b = −0.12, SE = 0.06, P = 0.063), control group compensation (b = −0.07, SE = 0.04, P = 0.060), and study monitoring (b = −0.11, SE = 0.06, P = 0.094) were used as predictors because they had P values less than 0.15 although they were not statistically significant. The correlation between treatment group compensation and control group compensation was large (r = 0.91), so control group compensation was not included as a moderator.
The meta-regression using true control group, treatment group compensation, and monitoring status as moderators was statistically significant in predicting RE (Qm  = 13.66, P = 0.003; Qe  = 372.95, P < 0.001). These three moderators explained approximately 33% of the between-study heterogeneity in the RE. Treatment group compensation (b = −0.07, SE = 0.03, P = 0.013) and monitoring status (b = −0.13, SE = 0.06, P = 0.023) were statistically significant. Using a true control group (b = −0.10, SE = 0.04, P = 0.071) was not statistically significant. Consequently, the results suggest that each $39 increase in treatment group compensation predicted a 7% decrease in RE. Implementing a monitoring procedure predicted a 13% decrease in RE.
In the first stage of the analysis, only age was statistically significant (b = −0.05, SE = 0.03, P = 0.049). However, education (b = −0.05, SE = 0.03, P = 0.12) explained a moderate amount of heterogeneity (R2 = 0.11) along with a P value <0.15, so education was also included in the final meta-regression analysis.
The meta-regression using age and education as moderators was statistically significant in predicting dropout (Qm  = 12.68, P = 0.002; Qe  = 54.35, P < 0.001). These two moderators explained approximately 53% of the between-study heterogeneity in dropout. Age (b = −0.07, SE = 0.02, P = 0.003) and education (b = −0.06, SE = 0.03, P = 0.024) were both statistically significant. Consequently, the results suggest that each 5-yr increase in mean age predicted a 7% decrease in dropout. Each 19% increase in the percentage of the sample attaining a high school education or less predicted a 6% decrease in dropout.
Researchers conducting exercise or physical activity studies for patients with breast cancer can expect to successfully recruit approximately two to three participants per week and approximately 30% of the individuals they contact. Although it is difficult to compare the findings of this review and other studies in the extant literature because of differences in the definitions of successful recruitment (e.g., 10,14), the findings of this study appear to be in line with previous studies (10–12). Because the included studies primarily had samples with a majority of highly educated, female Caucasians, it is possible that these estimates for RR and efficiency will not generalize to other populations.
Despite lower prevalence of breast cancer in African American, Hispanic, and Asian women in the United States, women in these groups generally experience a lower survival rate (62). Given that African American, Hispanic, and Asian women are generally underrepresented in clinical trials for patients with breast cancer (63), it is possible that the RR and efficiency for these populations may be lower than the results of this review. In addition, given that males constitute less than 1% of breast cancer diagnoses worldwide and often do not make up enough of study samples to even examine whether there are differences in recruitment across sexes (63,64), it is possible that the RR and efficiency for males may be lower than the results of this study.
For the studies included in this review, recruitment was primarily influenced by study-related characteristics, which is intuitive because individuals must decide whether they find the study commitments acceptable or not. Consequently, the presentation of the study information is critical for individuals in deciding whether to participate in a study. In presenting study information, researchers should consider who they contact and the reasons that participants might participate as well as participants’ perception of the study requirements. The studies included in this review indicated that there was a strong correlation between race and RR. Successful recruitment involves recruiting from the largest pool of participants possible. Consequently, studies with racially homogenous samples may recruit at a slower rate because they are not recruiting from the largest pool of potentially eligible participants. Successful recruitment also involves presenting study information so that participants’ motivations to participate are not undermined and participants are not overwhelmed. The studies included in this review indicated that there was a tendency for compensating participants and monitoring studies for protocol fidelity to reduce RE. Although the notion that compensating participants would reduce RE is counterintuitive, it is possible that monetary incentives undercut participants’ sense of altruism (65). As such, researchers should consider the effect of monetary incentives and how monetary incentives are presented to participants so that participants’ motivations for participating are not undermined. The relationship between monitoring studies for protocol fidelity and RE is more puzzling because they should be unrelated. It is possible that the added complexity required to monitor the intervention was inadvertently conveyed to potential participants (66).
When presenting information about the study to potential participants, it is important that research teams provide the information necessary to obtain genuine informed consent without overwhelming potential participants. As studies have become more complex (66), there is often a tremendous amount of information that could be presented to potential participants to help them decide whether to participate in a study or not. Effective recruitment likely entails relaying the information necessary to obtain informed consent (e.g., potential risks and benefits of the study, resources in the event of an adverse event), the information that the potential participant specifically requests (e.g., “How many times per week will I attend the intervention?”), and the information that is judged to be valued by the potential participant (e.g., a description of ways that flexible scheduling can be implemented if the participant mentions that he or she has a busy schedule).
It is also worth noting that monetary incentives are guided by institutional review boards (IRBs). However, IRBs typically restrict overcompensating individuals so that individuals are not coerced into participating. Our findings suggest that decreasing the amount of monetary incentives may improve recruitment, which would not contradict the guidelines set forth by most IRBs. This finding is limited, though, in that this finding is based on the reported compensation of 26 studies (93%), where only 6 studies (21%) compensated their participants.
Researchers should also consider how participants are being contacted. It is possible that potentially eligible participants may be more receptive to joining the study if they hear about it through a trusted source such as friends or family members who are participating in the study or respected community members (e.g., community health workers, church leaders). It is also possible that research teams may not be aware of the best methods for contacting potential participants about the study. Several of the included studies used community-based recruitment methods, such as speaking events (39,59), word of mouth (35), community events (54), community outreach (60), and personal referrals (44).
In terms of generally improving recruitment, studies in the extant literature involving patients with different forms of cancer have noted several factors that may influence recruitment. These factors include a lack of time (19), transportation (67–70), and a lack of interest (19,68,71). Although some of these are participant-specific factors (e.g., a lack of interest), others involve the intersection of study characteristics (e.g., the number of hours of intervention, the number of weekly intervention) and participants’ lifestyles (e.g., busy schedules, ability to travel to and from the intervention and/or study appointments). Consequently, researchers may consider interpreting the results of the current review to mean that recruitment is influenced by how well study requirements match potential participants’ interest and ability to participate. Thus, researchers should consider how acceptable potential participants’ find the study requirements to optimize recruitment.
Researchers conducting exercise or physical activity studies for patients with breast cancer can expect approximately 16%–18% dropout depending on the extent of publication bias present, which is in line with other studies in the extant literature (10,14). Heterogeneity in dropout was primarily predicted by participant characteristics. This is also intuitive because retaining participants involves the research team collaborating with participants to address personal factors that may threaten continued participation.
Although participant characteristics are likely not the causal factors for dropout, they do suggest participants’ lifestyle demands that researchers should consider when designing their study. For example, this review found that studies with older participants had lower dropout, which may suggest that age-specific factors influence dropout. These findings must, of course, be interpreted with regard to the included studies. Most of the included studies had a sample with an average age in the 50s. Although this review found that dropout tended to decrease as age increase, it is likely these results cannot be generalized beyond the average ages of samples in the included studies. For example, patients with breast cancer in their 70s or 80s may face the increased comorbidities, decreased functioning, sarcopenia, or frailty that might increase their risk of dropping out of a study.
In terms of minimizing dropout, researchers should be prepared to creatively problem solve issues that may threaten a participant’s continued participation. For example, Cheville et al. (19) reported lack of time as a common difficulty with recruitment and retention in studies. Consequently, it is possible that flexible scheduling (e.g., allowing participants to vary their intervention appointments from week to week, allowing participants to schedule intervention appointments on weekends), incorporating home-based intervention activities, or reducing the length of in-person intervention appointments could minimize dropout. Cantrell et al. supported the notion of flexible scheduling (20), which originally recommended by Young and Dombrowski (72). By minimizing dropout, participants will have a better opportunity to benefit from the intervention, and researchers will obtain improved generalizability and statistical power by retaining participants in the study.
The random-effects meta-analysis results for RE and dropout indicated large heterogeneity, both the absolute and the proportion of variability above chance. On one hand, this indicated that there was substantial variation across the included studies regarding the observed recruitment and retention. On the other hand, we expected large heterogeneity because recruitment and retention can be influenced by a variety of factors such as the efficacy of different recruitment methods and procedures, cultural norms toward trusting scientists and researchers, resources available to participants (e.g., childcare, transportation, flexible scheduling), and resources available to studies to recruit and retain participants (e.g., funding to hire a research assistant who recruits participants full-time, money budgeted for advertisements). Further, large heterogeneity allows a greater opportunity to identify factors that influence recruitment.
Reporting Recruitment Strategies
The included studies inconsistently reported their recruitment activities. Inconsistent reporting prevents other research teams from learn about effective (or ineffective) methods for recruiting. We hope researchers will make an effort to more fully report their recruitment methods so that researchers can better learn from one another, which could lead to improved recruitment and thus higher quality study findings.
Ideally, future work should be done to publish standardized guidelines for reporting recruitment strategies, similar to other existing publishing guidelines (e.g., PRISMA). Until that point in time, however, researchers would benefit by reporting descriptions of the study team members who are recruiting (e.g., how many study team members are recruiting, what are the disciplines of study team members who are recruiting, what training did recruiting study team members receive, and how much time do recruiting study team members dedicate to recruiting) because interactions between the study team and the potential participants may play an influential role in recruitment success. In addition, researchers would likely benefit by reporting the level of success with each recruitment method that was used. Almost all the included studies reported which recruitment methods they used, although it was sometimes unclear how successful each method was. Finally, research could benefit by reporting the financial investment into recruiting, which would allow for empirically based evidence regarding the money spent to recruit each patient. For example, Irwin et al. (41) reported spending approximately $4894 to recruit 75 participants, which equates to spending roughly $65 dollars per recruited participant. If this information were reported in almost all studies, researchers would likely have much better intuition regarding how much money to budget for recruitment purposes.
The findings and conclusions of this review are dependent on the data that was obtained through the included studies. Hence, studies with poor internal validity were a possible limitation (22). In addition, this review did not include any unpublished or gray literature studies, which could have further influenced the findings.
Regarding the collected data, reviews are limited to data that can be reasonably collected from publications. As an example, this review only collected data for monetary compensation, not services provided free of charge to the participants (e.g., clinic visits, exercise sessions). These services are clearly valuable, although it is extremely difficult to monetize the value of these services across a variety of populations, settings, and contexts.
Given the diversity of the recruitment methods, it was not possible to further examine the effectiveness of individual recruitment methods, as many of the recruitment methods were only used in a few studies. We expect there may be differences in the efficacy of recruitment methods, especially for community-based recruitment methods. Future work examining the efficacy of specific recruitment methods would help to improve recruitment.
Because of practical limitations, we were only able to include studies written in English. It is possible that these inclusion criteria introduced a language bias. Further related to the identified studies, we only used three databases to identify studies. It is possible that the use of other larger databases (e.g., EMBASE) would improve the quality of this review.
Finally, recruitment and retention are complex topics. It is possible that our conceptualization of recruitment and retention is erroneous. As such, there may be alternative conceptualizations for recruitment and retention that allow for better examination of what influences recruitment and retention.
For recruitment, we found dropout rates and race to be strongly correlated with RR, and we found study characteristics, specifically increases in monetary compensation and monitoring the intervention for fidelity, predicted decreases in RE. For retention, we found participant characteristics, specifically increases in mean age and the percentage of participants obtaining the equivalent of a high school education or less, predicted decreased dropout. This review may help identify characteristics associated with low recruitment and retention, as well as areas for future research, although this review cannot identify causal factors. It is our hope that these findings may encourage researchers to improve reporting of recruitment factors and assist researchers in optimizing recruitment and retention.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
The authors have stated that they have no conflict of interest. The results of the present study do not constitute endorsement by the American College of Sports Medicine.
Author contributions: JCH, AMA, MMA, SA, JLR, and PMK were involved in conceptualizing this project, creating the search strings, and defining the inclusion and exclusion criteria. JCH conducted the databases searches. JCH, AMA, MMA, SA, and BAA were involved in screening the identified studies. JCH, AMA, MMA, SA, BAA, CS, and AY extracted data from eligible studies. JCH, AMA, MMA, and BAA rated the eligible studies for risk of bias. JCH performed the data analyses. All authors were involved in producing the final manuscript and provided approval for the final version of this manuscript.
1. Damen L, van Agt F, de Boo T, Huysmans F. Terminating clinical trials without sufficient subjects. J Med Ethics
2. Foy RJ, Parry J, Duggan A, et al. How evidence based are recruitment strategies to randomized controlled trials in primary care? Experience from seven studies. Fam Pract
3. Puffer S, Torgerson DJ. Recruitment difficulties in randomised controlled trials. In: Controlled Clinical Trials
. New York (NY): Elsevier Science Inc.; 2013, pp. 214S–5.
4. Sully BG, Julious SA, Nicholl J. A reinvestigation of recruitment to randomised, controlled, multicenter trials: a review of trials funded by two UK funding agencies. Trials
5. Biau DJ, Kerneis S, Porcher R. Statistics in brief: the importance of sample size in the planning and interpretation of medical research. Clin Orthop
6. Patel MX, Doku V, Tennakoon L. Challenges in recruitment of research participants. Adv Psychiatr Treat
7. Johansen MA, Mayer DK, Hoover HC. Obstacles to implementing cancer clinical trials. Semin Oncol Nurs
8. Simon MS, Brown DR, Du W, LoRusso P, Kellogg CM. Accrual to breast cancer clinical trials at a university-affiliated hospital in metropolitan Detroit. Am J Clin Oncol
9. Lara PN Jr., Higdon R, Lim N, et al. Prospective evaluation of cancer clinical trial accrual patterns: identifying potential barriers to enrollment. J Clin Oncol
10. Sheill G, Guinan E, Brady L, Hevey D, Hussey J. Exercise interventions for patients with advanced cancer: a systematic review of recruitment, attrition, and exercise adherence rates. Palliat Support Care
11. Schmitz KH, Ahmed RL, Troxel AB, et al. Weight lifting for women at risk for breast cancer–related lymphedema: a randomized trial. JAMA
12. Schmitz KH, Ahmed RL, Hannan PJ, Yee D. Safety and efficacy of weight training in recent breast cancer survivors to alter body composition, insulin, and insulin-like growth factor axis proteins. Cancer Epidemiol Biomarkers Prev
13. Winters-Stone KM, Dobek J, Bennett JA, Nail LM, Leo MC, Schwartz A. The effect of resistance training on muscle strength and physical function in older, postmenopausal breast cancer survivors: a randomized controlled trial. J Cancer Surviv
14. Yee J, Davis GM, Hackett D, et al. Physical activity for symptom management in women with metastatic breast cancer: a randomized feasibility trial on physical activity and breast metastases. J Pain Symptom Manage
15. Freedman RA, Foster JC, Seisler DK, et al. Accrual of older patients with breast cancer to alliance systemic therapy trials over time: protocol A151527. J Clin Oncol
16. Simon MS, Du W, Flaherty L, et al. Factors associated with breast cancer clinical trials participation and enrollment at a large academic medical center. J Clin Oncol
17. Avis NE, Smith KW, Link CL, Hortobagyi GN, Rivera E. Factors associated with participation in breast cancer treatment clinical trials. J Clin Oncol
18. Gollhofer SM, Wiskemann J, Schmidt ME, et al. Factors influencing participation in a randomized controlled resistance exercise intervention study in breast cancer patients during radiotherapy. BMC Cancer
19. Cheville AL, Girardi J, Clark MM, et al. Therapeutic exercise during outpatient radiation therapy for advanced cancer: Feasibility and impact on physical well-being. Am J Phys Med Rehabil
20. Cantrell MA, Conte T, Hudson M, et al. Recruitment and retention of older adolescent and young adult female survivors of childhood cancer in longitudinal research. In: Oncology Nursing Forum
. NIH Public Access; 2012, p. 483.
21. Hoover JC, Alenazi AM, Alothman S, Alshehri MM, Rucker J, Kluding P. Recruitment for exercise or physical activity interventions: a protocol for systematic review. BMJ Open
22. Higgins JP, Altman DG, Gøtzsche PC, et al; Cochrane Bias Methods Group; Cochrane Statistical Methods Group. The Cochrane Collaboration’s tool for assessing risk of bias in randomised trials. BMJ
23. Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. Introduction to Meta-analysis
. John Wiley & Sons; 2011.
24. Barendregt JJ, Doi SA, Lee YY, Norman RE, Vos T. Meta-analysis of prevalence. J Epidemiol Community Health
25. Hedges LV. Statistic Methodology in Meta-Analysis
. Princeton (NJ): ERIC Clearinghouse on Tests, Measurement, and Evaluation; 1982. 79 p.
26. Duval S, Tweedie R. Trim and fill: a simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics
27. Egger M, Smith GD, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ
28. Gal R, Monninkhof EM, van Gils CH, et al. The trials within cohorts design faced methodological advantages and disadvantages in the exercise oncology setting. J Clin Epidemiol
29. Reuters T. EndNote, Version X8
. Thomson Reuters; 2016.
30. IBM Corp. IBM SPSS Statistics for Windows, Version 25.0
. Armonk (NY): IBM Corp.; 2017.
31. R Core Team. R: A Language and Environment for Statistical Computing [Internet]
. Vienna (Austria): R Foundation for Statistical Computing; 2019. Available from: https://www.R-project.org/
32. Wickham H. tidyverse: Easily Install and Load the “Tidyverse” [Internet]. 2017. Available from: https://CRAN.R-project.org/package=tidyverse
33. Stef van Buuren KG-O. mice: multivariate imputation by chained equations in R. J Stat Softw
34. Viechtbauer W. Conducting meta-analyses in R with the metafor package. J Stat Softw
35. Bennett JA, Lyons KS, Winters-Stone K, Nail LM, Scherer J. Motivational interviewing to increase physical activity in long-term cancer survivors: a randomized controlled trial. Nurs Res
36. Fields J, Richardson A, Hopkinson J, Fenlon D. Nordic walking as an exercise intervention to reduce pain in women with aromatase inhibitor–associated arthralgia: a feasibility study. J Pain Symptom Manage
37. Stan DL, Rausch SM, Sundt K, et al. Pilates for breast cancer survivors: impact on physical parameters and quality of life after mastectomy. Clin J Oncol Nurs
38. Kim SH, Shin MS, Lee HS, et al. Randomized pilot test of a simultaneous stage-matched exercise and diet intervention for breast cancer survivors. Oncol Nurs Forum
39. Korde LA, Micheli A, Smith AW, et al. Recruitment to a physical activity intervention study in women at increased risk of breast cancer. BMC Med Res Methodol
40. Bertram LAC, Chung G, Yu H, Salovey P, Irwin M. Feasibility of institutional registry-based recruitment for enrolling newly diagnosed breast cancer patients in an exercise trial. J Phys Act Health
41. Irwin ML, Cadmus L, Alvarez-Reeves M, et al. Recruiting and retaining breast cancer survivors into a randomized controlled exercise trial: the Yale Exercise and Survivorship Study. Cancer
. 2008;112(Suppl 11):2593–606.
42. Eakin EG, Lawler SP, Winkler EA, Hayes SC. A randomized trial of a telephone-delivered exercise intervention for non-urban dwelling women newly diagnosed with breast cancer: exercise for health. Ann Behav Med
43. Hayes S, Rye S, Battistutta D, et al. Design and implementation of the exercise for health trial—a pragmatic exercise intervention for women with breast cancer. Contemp Clin Trials
44. Loh SY, Lee SY, Quek KF, Murray L. Barriers to participation in a randomized controlled trial of qigong exercises amongst cancer survivors: lessons learnt. Asian Pac J Cancer Prev
45. Johnston MF, Hays RD, Subramanian SK, et al. Patient education integrated with acupuncture for relief of cancer-related fatigue randomized controlled feasibility study. BMC Complement Altern Med
46. McTiernan A, Ulrich C, Kumai C, et al. Anthropometric and hormone effects of an eight-week exercise-diet intervention in breast cancer patients: results of a pilot study. Cancer Epidemiol Biomarkers Prev
47. Westphal T, Rinnerthaler G, Gampenrieder SP, et al. Supervised versus autonomous exercise training in breast cancer patients: a multicenter randomized clinical trial. Cancer Med
48. Al-Majid S, Wilson LD, Rakovski C, Coburn JW. Effects of exercise on biobehavioral outcomes of fatigue during cancer treatment: results of a feasibility study. Biol Res Nurs
49. Myers JS, Mitchell M, Krigel S, et al. Qigong intervention for breast cancer survivors with complaints of decreased cognitive function. Support Care Cancer
50. Lynch SM, Stricker CT, Brown JC, et al. Evaluation of a web-based weight loss intervention in overweight cancer survivors aged 50 years and younger. Obes Sci Pract
51. Ammitzbøll G, Lanng C, Kroman N, et al. Progressive strength training to prevent lymphoedema in the first year after breast cancer—the LYCA feasibility study. Acta Oncol
52. Porter LS, Gao X, Lyna P, et al. Pilot randomized trial of a couple-based physical activity videoconference intervention for sedentary cancer survivors. Health Psychol
53. Daley AJ, Crank H, Mutrie N, Saxton JM, Coleman R. Patient recruitment into a randomised controlled trial of supervised exercise therapy in sedentary women treated for breast cancer. Contemp Clin Trials
54. Rogerino A, Grant LL, Wilcox H 3rd, Schmitz KH. Geographic recruitment of breast cancer survivors into community-based exercise interventions. Med Sci Sports Exerc
55. Aycinena AC, Valdovinos C, Crew KD, et al. Barriers to recruitment and adherence in a randomized controlled diet and exercise weight loss intervention among minority breast cancer survivors. J Immigr Minor Health
56. Broderick JM, Guinan E, Kennedy MJ, et al. Feasibility and efficacy of a supervised exercise intervention in de-conditioned cancer survivors during the early survivorship phase: the PEACH trial. J Cancer Surviv
57. Penttinen H, Nikander R, Blomqvist C, Luoto R, Saarto T. Recruitment of breast cancer survivors into a 12-month supervised exercise intervention is feasible. Contemp Clin Trials
58. Zhou ES, Dunsiger SI, Pinto BM. Proactive versus reactive recruitment to a physical activity intervention for breast cancer survivors: does it matter?Clin Trials
59. Ott CD, Twiss JJ, Waltman NL, Gross GJ, Lindsey AM. Challenges of recruitment of breast cancer survivors to a randomized clinical trial for osteoporosis prevention. Cancer Nurs
60. Pinto BM, Trunzo JJ, Rabin C, et al. Recruitment strategies for a home-based physical activity intervention for breast cancer patients. J Clin Psychol Med Settings
61. Hirschey R, Kimmick G, Hockenberry M, et al. A randomized phase II trial of MOVING ON: an intervention to increase exercise outcome expectations among breast cancer survivors. Psychooncology
62. Seiler A, Murdock KW, Garcini LM, et al. Racial/ethnic disparities in breast cancer incidence, risk factors, health care utilization, and outcomes in the USA. Curr Breast Cancer Rep
63. Duma N, Vera Aguilera J, Paludo J, et al. Representation of minorities and women in oncology clinical trials: review of the past 14 years. J Oncol Pract
64. Yalaza M, Inan A, Bozer M. Male breast cancer. J Breast Health
65. Kreps DM. Intrinsic motivation and extrinsic incentives. Am Econ Rev
66. Grunfeld E, Zitzelsberger L, Coristine M, Aspelund F. Barriers and facilitators to enrollment in cancer clinical trials: qualitative study of the perspectives of clinical research associates. Cancer
67. Oldervoll LM, Loge JH, Paltiel H, et al. The effect of a physical exercise program in palliative care: a phase II study. J Pain Symptom Manage
68. Cormie P, Newton RU, Spry N, Joseph D, Taaffe DR, Galvao DA. Safety and efficacy of resistance exercise in prostate cancer patients with bone metastases. Prostate Cancer Prostatic Dis
69. van den Dungen IA, Verhagen CA, van der Graaf WT, van den Berg JP, Vissers KC, Engels Y. Feasibility and impact of a physical exercise program in patients with advanced cancer: a pilot study. J Palliat Med
70. Leavens ELS, Stevens EM, Brett EI, Molina N, Leffingwell TR, Wagener TL. Use of rideshare services to increase participant recruitment and retention in research: participant perspectives. J Med Internet Res
71. Temel JS, Greer JA, Goldberg S, et al. A structured exercise program for patients with advanced non-small cell lung cancer. J Thorac Oncol
72. Young CL, Dombrowski M. Psychosocial influences on research subject recruitment, enrollment and retention. Soc Work Health Care