Racial and Ethnic Diversity in Studies Funded Under the Best Pharmaceuticals for Children Act

The study population reflected all participants enrolled in completed or ongoing BPCA-funded studies occurring from 2008 through June 2020 whose data are maintained with the BPCA data coordinating center. No study had inclusion and exclusion criteria that explicitly referenced race or ethnicity. The data extracted included age, race, ethnicity, study site, study protocol, and the month and year of enrollment. Studies enrolling mother–infant dyads were retained in the analysis, but those enrolling adults only were excluded. The geographic site of enrollment was identified for each participant along with reference census data for the general population in that city as derived from the US Census Bureau, World population review, Statistics Canada, United Kingdom Office for National Statistics, Israeli local council data, and Department of Statistics Singapore.^18–23  All participating study sites, irrespective of country, were required to record race and ethnicity according to the standards laid out by the Office of Management and Budget as used by the US Census Bureau.²⁴  The observed racial and ethnic enrollment distribution across all studies was compared with the expected racial and ethnic enrollment distribution determined by sampling the general population at the same geographic frequency as in the BPCA-funded studies.

Racial and ethnic enrollment was examined categorically by age, sex, and geographic region, whereas trends in minority enrollment were estimated over the entire evaluation period. Enrollment diversity in each state was explored by calculating the Simpson’s diversity index according to D = 1 − [(Σ n*(n − 1))/(N*(N − 1))], where n represents the total number of children for each racial or ethnic subgroup and N represents the total number of children for all groups.²⁵  The subgroups combine each of 7 racial groups (American Indian or Alaskan Native, Asian American, Black, Hawaiian or Pacific Islander, multiracial, not reported or unknown, white) with two ethnic groups (Hispanic, non-Hispanic) for a total of 14 groupings as recorded in each of the studies. The calculated index “D” ranges from 0 to 1 and represents the probability that two randomly selected individuals, from the same area, belong to a different racial or ethnic group. Converted to percent, 0 represents no diversity and 100 suggests maximal diversity.

The studies from which the participants were drawn were subclassified to investigate comparisons of minority enrollment by study characteristic. Study “type” was categorized as interventional or noninterventional. The perceived burden of participating in the study for any participant (ie, study burden) was classified as none, low, medium, or high: a rating reflecting the composite assignment of 3 pediatric clinician researchers, selected in a block randomized fashion from 6 in total. Raters evaluated each study according to their view on the burden of participation taking into consideration duration of participation, frequency of travel required for study visits, risks associated with study participation (ie, physical, emotional, social), invasive nature of study procedures, and prospect for direct benefit. Discrepancies between raters were adjudicated by a seventh independent clinician researcher. Raters were also asked to classify (yes or no) whether they expected to see racial or ethnic biases in participant diversity for each study (ie, study bias). Classification category was assigned by consensus of all raters on the basis of peer-reviewed literature, publicly available reports, and disease-focused Web sites (eg, Centers for Disease Control and Prevention, March of Dimes, NIH) that address racial and ethnic differences in disease epidemiology, health care use patterns, medication use, access to care, and prescriber treatment bias for the diseases under investigation.

Standard descriptive statistics were used to describe the study population. Comparison of proportions between observed and expected enrollment was accomplished by using a χ² test or z test. Generalized linear models were used to explore differences by sex, study type, and study burden. Linear regression was used to examine trends in enrollment over time. All statistical analyses were performed in SPSS v.23 (IBM, SPSS Statistics, IBM Corporation).

Data from 34 studies were available, representing 11 045 participants. Enrollment occurred at 164 clinical study sites distributed across 46 US states along with sites in Canada, England, Israel, and Singapore. One study in which researchers enrolled pediatric critical care providers (eg, physicians, nurses, and emergency medical technicians; n = 127) was excluded from the analyses, leaving a final data set constituted by 33 studies (Supplemental Table 7) enrolling 10 918 participants (United States: 10 483, international: 435). Studies ranged from retrospective reviews of treatment outcomes to prospective, randomized, placebo-controlled investigations. Thirty studies represented investigations of drugs, whereas the remaining 3 represented medical devices. The overall age of participants in these studies averaged 6.6 ± 8.2 years (range: 0–45 years) and was evenly distributed among male and female individuals (51% male versus 49% female). However, when mothers from studies enrolling mother–baby dyads were removed (n = 439), the participant age lowered to 5.5 ± 6.1 years (0–28 years, Fig 1).

Enrollment across all studies paralleled the expected enrollment based on estimates from the general population (Table 1). Although minimal with respect to percentage points, significantly higher-than-expected enrollment was observed in white, multiracial, American Indian, Hawaiian, and Hispanic children, ranging in absolute increases over expected from +0.2% to +2.6% (P < .05). Asian American children enrolled in lower absolute proportions than expected (−3.7%, P < .001), whereas essentially no difference between observed and expected enrollment was found for Black children (24.6% vs 24.5%, P = .864). There were no significant differences in racial distribution as a function of sex (P = .214), nor were any significant differences observed between the largest of the enrolled populations (infants <3 months, n = 3139) and children >3 months of age (P = .769).

Not surprisingly, enrollment in the United States varied widely by Census region and division (Table 2). A comparison with the expected racial and ethnic distribution by geographic region, when sampled at the same frequency from the same locations as the study participants, reveal a net neutral balance for most population groups. However, absolute percentage deviations of >5% and 10% were observed in 7 of 9 and 4 of 9 census divisions, respectively. Only the West North Central Midwest and the West South Central South saw enrollment parallel the general population, with deviation <5% across all racial and ethnic groups. The 1 exception to otherwise balanced enrollment across the United States was observed for the Asian American population. As illustrated in Table 2, statistically fewer Asian American children were enrolled in all but 1 Census division, and this was most pronounced in Western divisions where a higher overall proportion of the population originates from Asian ancestry.

A more granular look at enrollment diversity by state, as represented by the diversity index, is presented in Fig 2. The shading cutoffs employed in this figure reflect those that have been used, by others, to describe population diversity in counties across the United States.²⁶  In non-US study settings, the diversity was far less varied (Table 3), although wholly aligned with expectations for those regions. Calculated diversity index (converted to %) was highest in Canada (D = 54.0, n = 386), followed by England (D = 32.4, n = 21), with essentially no diversity in Israel (D = 0.0, n = 13) and Singapore (D = 0.0, n = 15).

When examining trends in racial and ethnic enrollment across the 13-year time frame, we observed wide fluctuation from year to year, although the rank order between racial and ethnic groups stayed relatively consistent (Table 4). In 5 racial groups (Asian American, Black, Hawaiian or Pacific Islander, not reported or unknown, white), there was no significant trend in enrollment over time, as evidenced by confidence intervals for the slope that span zero (Table 4). Similarly, there was no trend in enrollment among Hispanic participants. By contrast, enrollment among American Indian and Alaskan Native and multiracial children significantly increased during this time frame at rates of 0.17% and 0.36% per year, respectively (P < .01). Nevertheless, the absolute numbers enrolled for these population subgroups remain modest.

When explored by study type, the only significant racial and ethnic differences observed were lower enrollment for American Indian and Alaskan Natives and higher enrollment for Hispanic individuals in the interventional studies (P < .05, Table 5). More differences in racial and ethnic distribution were observed by study burden; however, these were largely influenced by the higher proportion of participants in the “zero” burden studies in which information on race and ethnicity was incompletely captured (Table 6). When focusing on the studies with some degree of participant burden (ie, those scored in the range of 1–3), white children were represented to a lower extent in moderate burden studies, whereas the inverse was observed for Black children. A higher proportion of multiracial children were enrolled in high burden studies (versus low or moderate burden) and a higher proportion of American Indian children were enrolled in moderate versus low burden studies.

Finally, we explored whether observed enrollment tracked with the over- or underrepresentation bias that we predicted on the basis of reported epidemiological, sociodemographic, and health care use patterns. For studies of conditions in which we expected higher enrollment in the Black population (eg, prematurity, obesity, sickle cell disease), we observed that the proportion of Black children enrolled was nearly 10 percentage points higher than seen in studies in which no bias between Black and non-Black children was expected (33% vs 24.3%, P < .001). In contrast, assumptions for studies in which we expected disproportionate enrollment in white (eg, mental health, infantile hemangioma) or Hispanic (eg, status epilepticus, renal transplant, obesity) children proved largely incorrect. For example, in studies in which a higher proportion of white children were expected, enrollment percentages were nearly 6 points lower than in studies in which no bias in white representation was expected (58.2 vs 63.9%, P < .001). Similarly, studies in which we predicted lower Hispanic enrollment demonstrated a near doubling in the opposite direction as compared with studies in which bias was not expected (29.8% vs 15.3%, P < .001).

This study reveals that the pediatric investigators responsible for executing federally funded drug and device studies under BPCA have ensured ethnic and racial representation among study participants. The minority enrollment observed in these studies was comparable to, or greater than, what would be expected on the basis of estimates from the general population. Thus, although the legislation never required prespecified racial and ethnic representation in pediatric studies, legislators may be reassured that racial and ethnic minorities are appropriately represented in trials supported by the off-patent component of BPCA, which is required by law to be reviewed and renewed by the US Congress every 5 years.

It is possible that the equity observed in this study arose because federally funded investigators are attuned to NIH initiatives surrounding race and ethnicity. However, the wide range of minority representation reflected in the most recent triennial inclusion reports for the various NIH institutes argues that this is unlikely the sole reason for our findings.²⁷  It is also possible that pediatric-trained clinician-investigators feel a greater sense of responsibility to ensure that broad enrollment is sought from all eligible participants and families for whom they care. However, this remains speculative because, to our knowledge, this has not been documented quantitatively for other large NIH-sponsored pediatric clinical trial programs. Another possibility is that multidisciplinary, pediatric, clinician-investigator teams that are experienced in the conduct of drug and device trials approach trial design with an eye toward simplifying inclusion and exclusion criteria, eliminating undue emotional distress, and minimizing participation burden wherever possible. However, we would be remiss not to acknowledge that the findings might simply reflect the balance of children presenting to the participant institutions. If this latter explanation were proven to be the case, it would underscore the need to purposefully select study sites that serve a broad cross-section of the population.

The one exception in these studies was children of Asian ancestry, who were represented at significantly lower proportions than the census would have suggested. Reasons for this finding remain unclear. With no significant disparity in overall rates of medically uninsured among Asian Americans, economic factors may not serve as the primary explanation.²⁸  It is well described that cultural norms influence health care seeking behaviors in the Asian American community, resulting in lower health care use,^29–33  and this appears to extend to Asian American children as well.^34,35  However, without the demographic distribution of patients receiving care at each of the 164 participating study sites, this theory cannot be confirmed. It is also possible that limited English proficiency, coupled with a lower likelihood of permission or assent and consent forms being translated into Asian languages versus more prevalent languages (eg, Spanish), could have played a role. Unfortunately, we did not have access to the permission or assent and consent forms created for these studies to explore this theory.

We observed no discernible trends in participation by study burden across the entire cohort, although unique differences were noted within populations. However, the conclusions that can be drawn from these findings are limited by the fact that our views of research burden (ie, the views of clinical investigators with extensive clinical trials experience) may be different from those of participants. Where medical and economic risks to clinical study participation are objectively easy to identify, psychological burdens are perhaps less so.³⁶  In the context of pediatric research, it is unclear whether concerns of the parent or guardian are projected onto the child. On a separate note, the overrepresentation of unknown or unreported participants in the zero burden studies, which largely reflects retrospective chart reviews, speaks to the inherent limitations of research conducted by using Electronic Health Record systems in which race and ethnicity are incompletely captured.

Interestingly, our predictions of enrollment bias, as informed by the medical literature, were largely incorrect. Although we did see higher Black participant enrollment in studies in which bias in favor of Black participation was expected, this was not reproduced for the white and Hispanic populations. A posteriori examination of our miscalculations for expected bias in white participants appeared to be influenced by the inclusion of studies dealing with pregnancy and lactation in which our perceptions were driven by reported racial differences related to adequate prenatal care. However, we were unable to account for the composition of the population served by the enrolling centers. When these studies were removed from the analysis, we did observe bias in the predicted direction (69.1% vs 63.9%, P = .009), albeit moderate. Our findings in the Hispanic population, in which lower estimates of enrollment were expected, may also be driven by patient demographics at the enrolling centers. However, it is more likely that our predictions reflect an incomplete understanding of the factors that influence the rate and extent to which patients present for care.

Despite perpetuated myths about minority participation in clinical research,³⁷  our data do not indicate that underrepresented pediatric populations are less likely to participate in clinical research. However, we cannot discriminate between the rate at which families from racial and ethnic minorities were approached for participation in these studies and the proportion that elected to participate. Nevertheless, these data reassure us that minority children have not been overlooked in the pediatric studies conducted with funding from BPCA.

On the basis of the nature of this investigation, we were unable to identify whether the individual studies were powered to explore the impact of race and ethnicity on study outcomes and whether such analyses were undertaken. However, ensuring balance in the acquisition of knowledge and the generation of evidence will be an important initiative as we look to refine our understanding of drug and device response as it shapes the risk/benefit balance for these interventions in children of different racial and ethnic backgrounds. Failure to do so will only serve to exacerbate the health disparities that exist among racial and ethnic minorities.³⁸ 

Historically, the biomedical research community has done an inadequate job of ensuring the broad and balanced enrollment of underrepresented minorities in drug and device trials.^39–44  This study revealed no evidence of racial or ethnic bias in pediatric clinical trial enrollment for investigator-initiated studies conducted with funding from the BPCA: a reassuring finding in light of the racial and ethnic disparities that persist in health care. However, these findings are not reflective of all contexts wherein pediatric clinical research takes place. In clinical research settings where similar findings have not been observed, or in settings where clinical trial planning is ongoing and adequate racial representation is paramount (eg, coronavirus disease therapeutic trials), collaborative discussions between stakeholders may offer solutions to resolve these disparities.

We acknowledge Gina Simone and Dr Ravinder Anand at the BPCA data coordinating center and Dr Perdita Taylor-Zapata at NICHD for their assistance and input on the execution of this study.

BPCA

Best Pharmaceuticals for Children Act

FDA

US Food and Drug Administration

NICHD

Eunice Kennedy Shriver National Institute for Child Health and Human Development

NIH

National Institutes of Health

PTN

Pediatric Trials Network