Our retrospective investigation of 123 men enrolled in a prostate cancer imaging study found that repeat non-adherence contributed to participant withdrawal more than SES. This finding is significant because it relates to the contemporary discussion regarding increasing diversity in clinical research.

Literature has consistently shown that clinical trials lack the inclusion of marginalized groups [4, 14, 15]. While the reasons for this are multifaceted and may be attributable to a lack of access, financial and logistical barriers, and general distrust, a significant reason may be due to provider perception. Fisher et al. suggest that “physician bias, false perceptions, and prejudices” have a significant impact on who they believe will comply with the often-demanding regimens involved with clinical research [16]. Our findings show that low SES, and/or minority patients, were not at higher risk for withdrawal and should not be excluded by physicians for this reason.

In our study, we used DCI as a measure for socio-economic evaluation. Various proposed measures for SES include insurance status, race, household income or employment status, or combination of these, however, the number of data points’ chosen to assess patients’ SES limit the ability to study the true impact of SES across various studies. A comprehensive evaluation of social, economic, and financial status is necessary to estimate SES and study its impact on health disparities and clinical trials. The Economic Innovation Group, a “bipartisan public policy organization”, introduced DCI, to understand the spatial distribution of US economic well-being. DCI combines seven distinct and complementary socio-economic indicators into a single score and creates a composite ranking by zip code. This score has been demonstrated to be an accurate measure of a community’s socio-economic distress and has been found to improve risk-adjusted outcomes after surgery [17].

One interesting but the anticipated finding is that participants with more than 1 non-adherent event were at a higher risk for eventually withdrawing from the research study. The majority of participants with two non-adherences withdrew from the study, nearly all participants with three non-adherences withdrew from the study, and all participants with four non-adherences withdrew from the study. It is important to note that not all non-adherences or withdrawals were directly attributable to the participant. Coordinator phlebotomy skill, inadequate screening, and clinic scheduling processes attributed to several non-adherences or withdrawals (Supplementary Table 2). Another important aspect regarding repetitive non-adherence is the investigators’ and participants’ perceptions regarding these events. As the number of non-adherence increases, while the participant might opt out of the study fearing their futile role in the study, the investigator might also withdraw these participants from the study, owing to undue burden on the research staff and introduction of bias in the study [18]. Non-adherence is known to increase variance, lower study power, and reduce the magnitude of treatment effects in a study and hence, International Society for CNS Clinical Trial Methodology (ISCTM) Working Group on Nonadherence in Clinical Trials proposed several recommendations to identify and mitigate its negative effects [18]. These measures predominantly include statistical analyses of nonadherence data and modification in study designs so as to address non-adherence at each step of the study.

We anticipate that future applications of our important finding would encourage real-time monitoring of non-adherent instances within clinical research studies in order to intervene early and prevent withdrawal from the study. The concept of real-time monitoring stems from successful remote monitoring to prevent and document adverse events, especially in phase 1, 2, and 3 clinical trials [19,20,21,22]. We specifically identified that the “adherence audit” should occur after the first non-adherence to identify the barriers of the participant or the coordinating team to implement immediate solutions. At that point, the team could assess study-team processes, environment, resources, or patient factors leading to the non-adherence and apply adaptive countermeasures to prevent further issues [12]. While our results on SES are favorable to expand enrollment, the finding by no means assumes there are no barriers caused by SES in clinical trials. We agree with acknowledging and addressing the barriers up front, to prevent non-adherence in a patient-centric manner [23]. Dockendorf et al advocated the use of digital health technologies comprising of smart dosing, outpatient sampling, and digital monitoring to increase participant retention with reduced burden on patients [23]. However, further research will be required to identify common barriers and solutions encountered in real-time during clinical trial activity.

A major limitation of this study is that our analysis derives from an observational study and not a clinical trial where intervention and randomization might further affect subject recruitment and retention. Furthermore, while we do have diversity of race and ethnicity, we lack age and gender diversity. Additionally, non-adherence data was gathered retrospectively resulting in unknown reasons for incomplete study procedures, a lack of details regarding the reason and time-point of withdrawal and therefore the decision to count any incomplete study procedure as a non-adherence regardless of retention status, and missing information regarding the barriers faced by subjects and coordinators. As the participants were required to travel to the study site, their travel distance could also be a potential factor for non-adherence to the study, however, this could not be studied. Also, small sample size and low number of events (withdrawals) need larger studies for definitive inference. Future studies are needed to examine various populations and different types of clinical research studies, specifically clinical trials that include real-time non-adherence data collection.