Recruitment and retention of women in a large randomized control trial to reduce repeat preterm births: the Philadelphia Collaborative Preterm Prevention Project

Randomized control trials (RCTs) are widely regarded as the most robust research designs for assessing the efficacy of medical and behavioral interventions [1, 2]. However, in some cases, barriers pertaining to the recruitment and retention of patients for RCT studies are formidable, and may result in compromised execution of studies and prohibitive costs [3]. Perhaps most importantly, unanticipated recruitment or retention problems pose threats to study integrity, either in terms of generalizeability of the findings, or bias associated with selective attrition of intervention or control group subjects [3, 4].

Successful recruitment and retention of sufficient numbers of subjects drawn from a diverse urban population may be especially challenging [5]. A lack of understanding of the purpose of medical research and research designs, general distrust of doctors and hospitals, and structural barriers stemming from lack of transportation, child care obligations, inflexible work schedules, and the hassles of everyday life have all been noted by some observers as potential problems that may undermine the recruitment and retention of subjects in RCT studies [3, 4, 6, 7]. These problems are widely thought to be more prevalent in minority, low income populations, or those living in distressed neighborhoods, where services may be inferior, housing conditions tend to be poor, rates of crime are high, and mobility may be seriously limited [3, 4, 6, 8]. Hence the potential for under-representation as a result of low recruitment or high attrition of disadvantaged, vulnerable populations warrants serious attention. Not surprisingly, a large body of research has emerged in an attempt to identify strategies to address this issue, including locating services and intervention sites where participants can be easily reached, providing appropriate and adequate incentives for enrollment and continued participation, and increasing the sensitivity and ability of staff to deal with cultural differences which may contribute to suspicion and reluctance of patients to consent or adhere to the study protocols [5, 9, 10].

While the conventional wisdom is that problems associated with the recruitment and retention of subjects for RCTs tend to disproportionately 'screen-out' minority, low income, or other residents facing difficult life circumstances, data which directly document the issue are surprisingly scant [6]. One systematic study of recruitment data pertaining to 70 health intervention studies in the U.S. led the researchers to conclude that, contrary to conventional wisdom, participation rates were not consistently lower for minority as opposed to non-minority persons [7]. As some observers have argued, factors such as the real and perceived value of what the study offers to the community and potential subjects themselves could outweigh or at least counterbalance those associated with inconvenience, accessibility, residential instability, fear, or suspicion [5, 6].

Reported here are the recruitment and retention data pertaining to the Philadelphia Collaborative Preterm Prevention Project (PCPPP), a large randomized control trial involving Philadelphia resident women who previously delivered premature (< 35 weeks gestation) infants. Described in detail below, the study was designed to address an array of risk factors for future preterm delivery and thus reduce the rate of repeat preterm births in the intervention group. The protocols for recruitment were without regard to race/ethnicity or other socioeconomic factors, but a concerted effort was made to avoid as well as document any bias in enrollment or selective attrition, especially that associated with membership in minority or low socioeconomic status groups, the group most at-risk for preterm and repeat preterm births. The PCPPP study protocol called for continuous and regular risk assessments, including living conditions, and physical and mental health status of both intervention and usual care participants, for up to a two-year period from baseline enrollment. An advanced database system was developed to track the follow-up status and assessment completion rates, and it proved to be a critical tool for the purposes of cohort maintenance. The data captured by this system were particularly useful for documenting baseline recruitment and retention rates for an urban study population with a substantial proportion of low income and minority residents, and for exploring the extent to which these rates varied by socio-demographic and other relevant factors.

Preterm birth (PTB) is an adverse pregnancy outcome which in the U.S. accounts for 65-75% of infant mortality and nearly 50% of long-term handicapping conditions in children [11, 12]. Despite reductions in infant mortality over the past 40 years, the PTB rate has remained stable, and has been a persistent source of significant race/ethnic and socioeconomic disparities. Although the causes of PTB are not well-understood, a substantial number of cases occur in association with an inflammatory process marked by increased membrane production and elevated amniotic fluid levels of various proinflammatory cytokines, prostaglandins, and metalloproteases [13–16]. In many cases of PTB, a uterine infection is present, but in others, the origin of the inflammatory process is found elsewhere in the mother's body, or not at all. Infections associated with PTB include periodontal disease, urinary tract infections, bacterial vaginosis, pneumonia and influenza [17–20]. In addition to infection, many of the other risk factors for PTB are proinflammatory (e.g., smoking, obesity, depression and psychosocial stress), and are also associated with increased levels of systemic inflammatory markers (SIM) such as, IL-6, C-reactive protein and serum amyloid A [16, 21]. These behavioral risk factors may act through a common inflammatory pathway leading to enhanced risk of PTB.

The hypothesis that seemingly disparate yet well-known risk conditions lead to PTB through a common physiological pathway may explain the persistence of population-based race/ethnic disparities, as well as the individual propensity for repetitive PTB. First, at the population level, these risk conditions are highly prevalent and tend to co-occur in socioeconomically disadvantaged populations. The co-occurrence of multiple risk factors associated with poverty and minority status may put entire groups of women at heightened risk for generalized systemic inflammation (SI) [21–24]. Secondly, at the individual level, the best predictor of PTB is a previous PTB, such that women with a previous PTB are at 2 to 6-fold greater risk for a second PTB, compared to women with a previous full term birth [25]. This suggests that the sources of individual physiological risk persist over time. Furthermore, the earlier the gestational age of any PTB, the greater the risk for a repeat PTB. In one completed progesterone trial, for example, more than 50% of women in the usual care group with a prior history of PTB at approximately 32 weeks gestational age (GA) had a subsequent PTB [26].

Virtually all interventions aimed at reducing PTB, including those that target the inflammatory pathway, have been conducted during pregnancy [12]. These include interventions to improve access to high quality prenatal care, bed rest, nutrition counseling, caloric or vitamin/mineral supplementation, smoking, drug or alcohol cessation programs, and treatment for pelvic infections. The delay of interventions until pregnancy is achieved may explain why most of the interventions have failed to reduce PTB, since once the inflammatory cascades are initiated, it may be too late to interrupt the processes. Furthermore, it may take weeks or months after the behavioral risk factor is modified to register consequent reductions in SI.

For these reasons, we proposed a multifaceted behavioral risk reduction approach during the interconceptional period in a population of women who have already experienced a previous PTB at < 35 weeks gestation. The conceptual model for the intervention is depicted in Figure 1.

To demonstrate the efficacy of the intervention, we carried out a randomized trial, recruiting Philadelphia resident women from the twelve hospitals providing 70% of the obstetric services in the region. Eligibility for enrollment was defined as follows: 1) delivery of live born singleton infant at <35 weeks of gestation; 2) English or Spanish speaking; 3) Philadelphia residency; and 4) not receiving operative sterilization before discharge from the hospital. During the postpartum hospital stay, project staff explained the study and attempted to obtain written informed consent. For those women who refused to participate, we obtained consent to collect pertinent medical records for the index pregnancy, the outcome of the next pregnancy via vital records linkage, and to conduct a much abbreviated survey to obtain data on important demographic, psychosocial and behavioral risk factors. Prior to the beginning of the study, based on individual birth record data for a historical six-year birth cohort, we calculated that approximately 1,670 <35 week singleton PTB's would occur during the 18-month enrollment period, and that 85% of the women would be eligible and agree to participate (n = 1,420). We also determined that 43% of the women would experience a subsequent live birth within the following 2-year period. In addition, of the women with a subsequent live birth, 50% of the second pregnancies were estimated to be <35 weeks gestation.

Ongoing data collection and risk assessment were an integral part of the study design with the objective of continuously identifying, referring and providing treatment for all women in the intervention group who had any of the risk factors associated with SI noted above. As reflected in Table 1, risk assessment/data collection visits for the Usual Care group closely mirrored that for the Intervention group, both in terms of the timing, follow-up period and scope of information collected. If a woman in either group became pregnant following the index birth, the protocol called for the last formal assessment/data collection time point to occur at 20 weeks gestation of the subsequent pregnancy, and for all relevant data pertaining to the period since childbirth to be abstracted from the medical chart. Regardless of when women were enrolled during the 18-month enrollment window, which began 11/01/2004, all risk assessments/data collection visits were scheduled to end on 9/1/2008 under the study protocol. Thus, women who enrolled after 9/1/2006 were not eligible for the full two-year follow-up visits and interventions.

The recruitment rate for the PCPPP study indicated that about 78% of all women who met the study's eligibility criteria consented to participate and were successfully randomized into an intervention or usual care group. A very high percentage of minority women agreed to participate (85%) and, contrary to findings published in other reports [29, 30], at considerably higher rates when compared to non-minority women (58%). In addition, independent of race/ethnicity, women residing in relatively low income households, women who reported having ever been homeless, and unmarried as opposed to married women were more likely to consent to participate.

These women fit the classic profile of those typically assumed to be less willing, or likely to volunteer for, or enroll in clinical trials. Thus, the results may be regarded as 'unexpected'. Our success is likely to be related to our recruitment protocols which were carefully designed to ameliorate and compensate for factors which might have otherwise led to disproportionate under-recruitment of minority, low SES, or other women facing especially difficult or challenging life circumstances.

Yet, it is important to note that our findings are not without precedent. For a randomized control trial designed to reduce behavioral risk factors associated with poor pregnancy outcomes conducted in Washington, D.C., El-Khorazaty et al. reported that 90% of all minority women asked to participate agreed to do so [6]. The study was limited to minority women, and so recruitment and retention rates for minority women could not be compared to those of non-minority women. However, in a systematic review of over 70 studies reporting recruitment data by race/ethnicity, Wendler et al. reported no consistent pattern of lower enrollment rates by minority status [7]. Moreover, for a subset of the 28 studies reviewed by Wendler and colleagues, involving seven surgical trials, recruitment rates were actually significantly higher for minority (African American and Hispanic) than for non minority (non-Hispanic White) women. Taken together with the findings presented here, this evidence suggests that minorities are at least as willing, and in some cases may even be more willing, to participate in health research.

Assumptions about the willingness of minority, as well as low income and other vulnerable populations, to participate in health studies are important to verify empirically [31–33]. Evidence of elevated risks pertaining to a wide range of chronic and acute illnesses in such populations is overwhelming and the need to reduce racial/ethnic and SES health disparities is widely recognized as a major public health priority [33–36]. Unwarranted or avoidable exclusion from randomized control trials and other legitimate health related research, of race/ethnic minorities or low income individuals, may inhibit discovery and thus the availability of effective treatments for chronic and other illnesses underpinning the persistent and pervasive SES and racial health disparities [32, 33].

We found no differences in follow-up rates according to race/ethnicity, SES or other factors which might reasonably be used to define 'hard-to-reach' or 'difficult to find' populations -- again in contrast to the findings of several prior studies reporting difficulties retaining minority, low income, residentially unstable, or otherwise distressed individuals. As was the case with recruitment, every reasonable effort was made in this study to maintain the cohort, so as to overcome distrust, inconvenience, lack of resources or the presence of other factors which otherwise might have led to relatively poor follow up rates for women facing numerous barriers to participation.

It may be that, when such factors are adequately addressed, neither recruitment rates nor retention rates will necessarily be biased -- that is, compromised by under-representation of minority, low income, and other vulnerable populations. As other observers have suggested [7, 8, 37], it may be that the prospects for successful, reasonably proportionate recruitment and retention of 'hard to reach' or 'difficult to find' groups depends on both appropriately addressing the potential barriers and the nature of the study itself. Specifically, it may be that the perceived 'costs' in the form of inconvenience and/or reluctance in the form of suspicion of motives, are often weighed against the perceived possible 'benefits' of participation. In that light it is important to note that the PCPPP study offered a fairly broad array of what were arguably familiar and highly valued services and medical treatments, over as many as two years. The overall purpose of the intervention itself may also have been perceived to be of extremely high value by the women eligible for enrollment, especially by 'at-risk' women who may have otherwise been reluctant to agree to participate and adhere to the study protocols for ongoing assessments. At stake was the reduction of the risk of another preterm delivery, and all that implies in terms of increasing the general well being and indeed the chances of survival of a subsequent offspring.

As noted earlier, the intervention components of the study described in Table 1 were chosen based on empirical evidence of the relationship between systemic infection and the known risk factors for PTB. The implementation of the interventions were specifically designed to augment existing medical or behavioral services in the city, remove existing barriers to those services, or provide such services when generally unavailable. In all cases, all costs associated with receiving the interventions were paid for from project funds, removing any burden of cost from all study participants. Many of the services that were provided as intervention components were either not available at all in the city or not routinely available, at least at the level or depth that they were offered to the intervention group. For example, screening for BV for postpartum women is not routine practice by providers; the housing services provided - including rental assistance, and tenant/landlord mediation are well known to be difficult to find in the city; and routine screening for depression and careful case-management, counseling and referral services are also well-beyond what is available to most women, whether insured or not. It is also well known that, because the costs are so prohibitive, low income women -- whether Medicaid insured or uninsured -- typically have very limited access to dental services, especially as they pertain to the diagnosis and treatment of periodontal disease. In summary, the services associated with the interventions were clearly distinguishable from what is generally considered to be "usual care", especially where factors related to cost, inconvenience, or lack of access to care is concerned. Given that virtually all cost factors were effectively removed, that physical and other barriers pertaining the accessibility of health care services in poor neighborhoods were addressed, and that the otherwise limited scope of services provided under Medicaid were expanded, it seems plausible to suggest that low income and minority women, in particular, would indeed perceive the interventions as having considerable benefit.

The overall PCPPP follow-up rates for the first (one month) postpartum assessment of 80% are similar to those published for other studies. Retention rates for the Washington D.C. study mentioned above, for example, indicated that 79% of the enrolled women were successfully followed-up using a telephone survey at 8-10 weeks postpartum [6]. Wall et al. reported that 69% of the women randomized into a study designed to reduce postpartum smoking relapse were successfully followed-up by a telephone survey at 12 months postpartum [38]. Katz and colleagues reported a successful postpartum follow-up rate using telephone surveys at 12 months for their study, involving a multi-site parenting intervention, of 59% [5]. Postpartum follow up rates for similar studies conducted in other countries with a universal system of care, however, appear to be somewhat better. In postpartum smoking relapse prevention RCT in Canada, for example, follow-up rates were reported to be 95% at 12 months [39].

In the ongoing, large scale Fragile Families observational study, which recruited subjects from 20 major U.S. cities, and involved a national sample of more than 4,700 index births, 89% and 86% of families were successfully followed-up by telephone survey at one and three year time points, respectfully [40]. However, follow-up rates for the in-home assessment portion of the study were considerably lower. According to the published data, only 54% of the intended in-home assessments at three years (2,581 of the 4,789 baseline sample) were successfully completed. This would appear to be a somewhat higher rate of successful follow-up than reported here (47.3% of the expected final postpartum PCPPP assessments were completed). The lower rate for the PCPPP assessments presented here may be attributable to the assessments for the intervention and usual care group requiring on-site clinical visits, rather than simple follow-up via a telephone survey or home visit.

Although the financial incentives for ongoing participation for both groups was equal, and both groups were afforded the same accommodations and reimbursements associated with all on-site clinic assessments, assessments pertaining to all time periods were completed at higher rates by the intervention compared to the usual care group. Because, by definition, the groups were randomized, these differences are unlikely to be caused by factors other than the nature of the intervention itself. Consequently, the results are consistent with the notion that the extent and nature of the medical and behavior services constituting the PCPPP intervention arm were highly valued and, in and of themselves, served as an incentive for the intervention group to remain in the study.

With regard to the comparison of women who consented vs. those who did not, it is important to note that, of all those who did not consent, only 36% (n = 118) agreed to complete the short form survey. Hence, the 'true' characteristics of all the women who did not consent were estimated based on this subsample. As with any subsample which is not randomly selected it was subject to selection bias. In theory, the bias could have resulted in a disproportionately greater number of higher income, more educated, non-minority women having completed the short-form survey. In that case what appeared to be evidence of greater willingness on the part of minority and low SES women to participate in the larger study could instead be a reflection of selection bias with respect to the characteristics of the subsample of women who agreed to complete the short-form survey. To address this concern we conducted a supplementary analysis of birth certificate data for all women who delivered preterm infants at the hospitals where women were recruited, during the same time period active recruitment occurred. The distributions for age, marital status, education, and race/ethnicity calculated using these data were virtually identical to those we calculated for the combined data gathered from all women approached at delivery - that is, for both those who completed the short-form survey and those who consented to the larger study. If there were any substantial selection bias related to sociodemographic factors we would expect these distributions to be discernibly different, with the data from birth records reflecting the 'unusual' group who declined the short-form survey. Since the distributions were not different, we concluded that the data from the short-form surveys were representative of all non-consenting women, and thus that differences observed between those who consented and those who did not were not the result of any selection bias.

In summary the findings from the analysis of recruitment, randomization and retention data presented here adds to a small but growing body of literature which increasingly challenges a widely held belief that low income and minority women are necessarily averse to enrolling in clinical trials or community studies. This study demonstrates that if the barriers to participation of disadvantaged populations are appropriately addressed, high participation rates can be achieved. Much of the concern about the under-participation in randomized control trials and community studies by racial/ethnic minorities, in particular, has focused on the past abuses or ethical lapses pertaining to inadequately informing (or in some cases deliberately misleading) potential subjects about potential risks [8, 37, 41]. In no way would we wish to suggest that such concerns are unwarranted, or that they need not be carefully and fully addressed. Rather, the important point is that these concerns should be weighed in the context of the real and perceived benefits of the study interventions and findings to both the individuals involved and to the communities where they live.