Race and Ethnicity Reporting and Representation in Pediatric Clinical Trials

Records of clinical trials submitted to ClinicalTrials.gov through March 9, 2020, were extracted with the Aggregate Analysis of Clinical Trials.gov tool.^4,11  Inclusion criteria included US-based trials, with a maximum enrollment age of <18 years, a “complete” study status, and available study results. Only trials registered on or after October 1, 2007, were included to account for the introduction of the results database and increased accessibility of ClinicalTrials.gov after passage of the Food and Drug (FDA) Administration Amendments Act in September 2007.^12–14  Trials performed outside of the United States were excluded. This report applies similar methods to and extends our previous analysis on pediatric clinic trial discontinuation, results reporting, and publication.¹⁵  An exemption from institutional review board review was granted given that all data are publicly available.

The 2019 US Census was referenced for <18-year-old national population statistics.¹⁶  We determined disease prevalence for select childhood conditions using the 2019 National Survey of Children’s Health (NSCH). Hospitalization data were based on the 2016 Kids’ Inpatient Database (KID), a nationally representative database of pediatric hospitalizations.^17,18  NSCH survey responses and International Classification of Diseases, 10th Revision, codes corresponding to the primary diagnosis in the KID database were used to calculate estimates of prevalence and hospitalization, respectively, by each race and ethnicity category.

We identified 5 racial and ethnic categories that most closely align with the predominant racial and ethnic composition of the United States: American Indian (including Alaskan Native), Asian American (including Pacific Islander and Native Hawaiian), Black, Hispanic, and white (Supplemental Table 2). We manually reviewed entries for race and ethnicity data and matched specified race and ethnicity groups to 1 of the aforementioned categories.

Of note, only a subset of registered trials are required to report baseline sociodemographic features to ClinicalTrials.gov. Per Section IV.C, the Final Rule (effective January 2017) mandates submission of participant race and ethnicity information for applicable clinical trials, which include: phase 2 to 4 interventional studies; controlled studies of FDA-regulated drugs, biologics, or devices; and studies conducted under an investigational new drug application or investigational drug exemption.¹⁹ 

We selected the following trial characteristics from Clinical Trials.gov as exposure variables: funding source (industry, government, academia, other), enrollment size, blinding, randomization, phase, intervention type, and number of participating sites (Supplemental Information).^20–23 

Pediatric diseases included in the evaluation of clinical trial enrollment relative to disease burden were determined a priori (Supplemental Table 3). For prevalence, we selected among the most common chronic conditions captured by the NSCH: asthma, diabetes, autism, attention-deficit disorder and attention-deficit/hyperactivity disorder, and obesity. Using the KID, leading causes of hospitalization were identified: bronchiolitis, sickle cell disease, asthma, major depressive disorder, and diabetes.

To identify studies associated with diagnoses of interest, the study team manually reviewed all eligible trials and assigned 1 or multiple therapeutic foci on the basis of the study title, abstract, and detailed trial description and in accordance with previous analyses and Medical Subject Headings.

Our 2 outcomes were (1) race and ethnicity reporting, and (2) trial diversity. Race and ethnicity reporting was defined as the proportion of all pediatric trials on ClinicalTrials.gov that reported any race and ethnicity characteristics. We only examined trials that reported results to ClinicalTrials.gov. To quantify trial diversity, we determined the median proportional enrollment of each racial and ethnic category among trials reporting race and ethnicity data. Each racial and ethnic group was analyzed independently. If a racial or ethnic group was not reported, it was assumed that there were no participants of that group. We performed a sensitivity analysis with 4 different models for estimating trial diversity to account for inconsistencies in trial reporting. Given the limited enrollment and lack of prevalence and hospitalization data, the analysis of American Indian representation was restricted to descriptive statistics.

We summarized descriptive analyses with 2-sided Pearson’s Χ² tests. We assessed temporal trends using compound annual growth rates (CAGR) with Mann-Kendall significance tests for both race and ethnicity reporting and representation. Only complete years and those with >10 completed trials were included in the assessment of reporting rates over time.

We performed multivariable logistic regression analysis to determine the association between clinical trial characteristics and reporting of race and ethnicity, controlling for all other trial features. Missing data were recorded in 0.5% (n = 60) of included trials and was handled with multiple imputation with chained equations. We generated and pooled 20 imputed data sets in accordance with Rubin’s Rules.²⁴ 

Secondary analyses were conducted to approximate the extent of disparities in enrollment compared with the US population and estimates of disease burden. We defined the enrollment-Census difference (ECD) as the absolute difference between the proportion of trial participants and the proportion of the US population for each race and ethnicity. The enrollment-prevalence difference and enrollment-hospitalization difference (EHD) compares the distribution of race and ethnicity among trials studying select diagnoses with the proportion of patients of a given race and ethnicity accounting for the overall prevalence (from the NSCH) or hospitalization rate (from the KID database) for these conditions, respectively. As a relative metric of disparities in trial representation, we defined the enrollment-Census ratio (ECR), enrollment-prevalence ratio, and enrollment-hospitalization ratio as the proportion of trial participants from each race and ethnicity divided by the aforementioned reference points. A 1-sample Wilcoxon signed-rank test was performed against a value of 0 for the median difference measures and against a value of 1 for the median ratio measures. Similar approaches for measuring trial diversity have been previously described.^2,25 

Statistical analyses were 2-sided with a threshold of statistical significance of α = 0.05. We performed all analyses in R version 3.6.2.

We identified 1183 US-based pediatric clinical trials encompassing 405 376 participants registered on ClinicalTrials.gov between 2007 and 2020 with available study results (Fig 1). Nearly half (49.6%) of included trials were funded by academic institutions and most were randomized (70.8%), single site (65.2%), and with an enrollment of <100 participants (Table 1).

A total of 43.3% (n = 512) of pediatric trials, amounting to 201 825 total participants, reported any race and ethnicity data. In trials reporting any race and ethnicity, white, Black, Asian American, Hispanic, and American Indian race and ethnicity were reported in 93.1%, 90.8%, 80.1%, 68.8%, and 73.8% of trials, respectively. A total of 49% of trials reporting any race and ethnicity data included all 5 racial and ethnic groups. Over the study period, the proportion of trials reporting any race or ethnicity significantly increased from 27.5% in 2008 to 86.7% in 2018 at a compound annual growth rate of 12.1% (Fig 2). Similar growth was observed for reporting of each race and ethnicity (Supplemental Table 4).

On multivariable regression analysis, trials funded by industry (adjusted odds ratio, 0.36; 95% confidence interval, 0.24–0.53) and academic institutions (adjusted odds ratio, 0.38; 95% confidence interval, 0.27–0.54) were significantly less likely to report race and ethnicity than government-funded trials (Supplemental Table 5).

Among trials reporting race and ethnicity data, white children constituted most trial participants (median, 66.4%; interquartile range [IQR], 41.5%–81.6%). This was followed by Black children, 12% (IQR, 2.9%–28.4%); Hispanic, 7.1% (IQR, 0%–18.6%); Asian American, 0.6% (IQR, 0%–3.7%); and American Indian, 0% (IQR, 0%–0%) (Fig 3A). The median enrollment of Hispanic participants significantly increased from 2008 to 2018 at a CAGR of 7.8%, while remaining relatively stable for other race and ethnicity groups (Supplemental Table 6) (Supplemental Fig 5). Our sensitivity analysis of alternative approaches to race and ethnicity classification demonstrated similar estimates of trial diversity (Supplemental Table 7).

Race and ethnicity enrollment varied with trial characteristics (Supplemental Table 8). There was a significantly greater median proportion of Black (15.5% [IQR, 6.7%–29.4%] vs 10% [0%–25.9%]), and Hispanic (10.3% [0%–21.7%] vs 4.5% [0%–16.7%]) participants in multicenter trials compared with single-center trials. Increased Black, Hispanic, and Asian American representation was also observed with larger trial enrollment (>100 participants).

White participants were significantly overrepresented (ECD, 12.1; IQR, −12.7 to 27.3; P < .001; ECR, 1.2; IQR, 0.7–1.5; P < .001) compared with US population estimates (Fig 3). Conversely, Black (ECD, −5.9; IQR, −15.1 to 10.4; P = .012; ECR, 0.7; IQR, 0.2–1.5; P = .012), Hispanic (ECD, −15.9; IQR, −23 to −4.3; P < .001; ECR, 0.3; IQR, 0–0.8; P < .001), and Asian American (ECD, −6.7; IQR, −7.3 to −3.6; P < .001; ECR, 0.1; IQR, 0–0.5; P < .001) participants were significantly underrepresented.

We identified racial and ethnic disparities between trial enrollment and measures of disease burden for select childhood conditions (Fig 4) (Supplemental Fig 6). White participants were consistently either overrepresented or appropriately matched to expected proportions on the basis of disease prevalence and hospitalizations. The largest disparities were observed in Asian American and Hispanic participants, who were underrepresented across most conditions. Relative to disease prevalence, Black participants were underrepresented in trials of autism and diabetes, and overrepresented in asthma trials. The enrollment-hospitalization disparity and enrollment-hospitalization ratio were unfavorable for Black participants in diabetes trials.

In this cross-sectional analysis of ClinicalTrials.gov, we identified ongoing challenges in the documentation of race and ethnicity information and representation of diverse populations in US-based pediatric clinical trials. Although less than half of trials reported race and ethnicity overall, considerable progress has been made, with rates more than tripling over the study period. Among trials reporting race and ethnicity, we found significant disparities in the enrollment of Black, Hispanic, and Asian American participants relative to population statistics and measures of disease burden. Our work provides a comprehensive picture of racial and ethnic characteristics of the pediatric clinical trial portfolio, evaluation across trial features and diagnoses, and temporal trends.

Studies of journal publications have generally found higher rates of race and ethnicity reporting than those observed in our analysis of ClinicalTrials.gov.^26–28  Journals have enforced comparatively stricter disclosure requirements for sociodemographic characteristics.²⁹  The significant improvement in ClinicalTrials.gov reporting over time, now commensurate with publication estimates, may be in part related to expanded regulatory guidelines. Most recently, the Final Rule strengthened the requirements for race and ethnicity reporting to ClinicalTrials.gov established by the FDA Administration Amendments Act.¹⁹  Reporting among applicable trials has been shown to be significantly greater since its passage.²⁸  In this analysis, government-funded trials were most likely to report race and ethnicity, indicating that policy impacts have extended to the pediatric population.

Transparency of sociodemographic data aligns with modern clinical trial practices and are needed to evaluate the external validity of study findings. By extension, these data permit the surveillance of disparities in trial enrollment at a systemic level. Investigators, databases, and publications should be held to high standards of precision and consistency when handling race and ethnicity information. Our analysis focused only on documentation of study populations by race and ethnicity; however, this should be considered the bare minimum. Ongoing efforts to improve reporting conventions call for more nuanced explanations concerning methods for categorization and reasons for invoking race and ethnicity as a covariate.³⁰ 

We found that minoritized children were underrepresented in pediatric trials relative to the US population. Enrollment patterns remained largely unchanged over the study period, despite these racial and ethnic groups emerging as the fastest growing segments in the United States.³¹  The magnitude of disparities may be even greater than estimated, given Black and Hispanic populations have been historically undercounted in the US Census.^32,33  Importantly, race and ethnicity representation varied across trial characteristics, including proportionally greater enrollment of minoritized children in larger (>100 participants), advanced phase, and multicenter trials. Greater diversity may thus be achieved in the studies with the most significant public health and clinical implications.

Racial and ethnic minoritized groups shoulder a disproportionate burden of disease morbidity and mortality, and should therefore constitute a larger share of those eligible for clinical trial participation. Yet, this is largely not reflected in our exploratory analyses of specific conditions. We found that Hispanic and Asian American participants, and, to a lesser degree, Black participants, were underrepresented relative to the prevalence and hospitalization rates of both medical and psychiatric diagnoses. One notable exception was the enrollment of Black children in asthma trials, which was either greater than or appropriately matched to the expected proportions on the basis of disease prevalence and hospitalizations, respectively. Asthma has become a central case study in scientific, policy, and environmental dimensions of pediatric and adult health disparities.^34–37  It is plausible that these efforts have spurred a heightened sense of accountability to act on the structural factors responsible for racial disparities in asthma.

There is variability in the directionality and degree to which minoritized children are enrolled across other reviews of pediatric trial diversity.^{1,9,10,38–41}  In a recent study of clinical trials published in pediatric and general medicine journals, for example, Black and Hispanic participants were proportionally well represented relative to US population statistics, whereas white participants were underrepresented.¹⁰  We have previously shown that clinical trials that disclose their results on ClinicalTrials.gov do not consistently publish in a peer-reviewed journal, and vice versa.¹⁵  That the disparities observed in ClinicalTrials.gov were not necessarily recapitulated in Rees et al might suggest a publication bias toward studies with greater diversity, inclusive study designs, and larger cohorts. Altogether, this discrepancy highlights the importance of evaluating across data sources to establish a more complete profile of the pediatric clinical trials ecosystem.

Differences in enrollment benchmarks can also lead to differences in the interpretation of trial diversity. Abdel-Rahman et al evaluated trials funded by the Best Pharmaceuticals for Children Act and concluded that all race and ethnicity groups were equitably represented.⁸  The authors derived expected proportions from Census data from the same geographic region as the study sites. Catchment areas, however, do not necessarily reflect the US population demographics nor disease epidemiology. The unequal distribution of trial sites may disproportionately impact access for racial minoritized, low income, and rural populations.^42–44  Although setting a universal target for evaluating representation in pediatric research is challenging and location-specific characteristics can be meaningful, a comparative strength of our study is the use of multiple reference points and national-level data sets.

Overcoming the structural inequities to diversity in pediatric trials demands broad stakeholder engagement. Policies, journals, and institutional review boards can establish rigorous standards for representation as a condition for approval or publication.⁴⁵  In turn, study sponsors must establish operational measures that address barriers to participation. These include financial reimbursement, expanded eligibility criteria, multisite enrollment, and multilingual protocols.^46–48  Additionally, investigators must prioritize partnerships with historically marginalized communities to not only align with local priorities and values, but also to redress longstanding mistrust in medical and research institutions.^49,50  Recruitment strategies should be contextualized within the unique legal and ethical considerations of pediatrics. Racial and ethnic minoritized parents have been found to be more likely than white parents to refuse consent for their child to participate in research with limited efficacy of material incentives.^51–53  An even greater emphasis should thus be placed on understanding caregiver concerns and empowering informed decision-making.

This study has several limitations. Firstly, inconsistencies exist with respect to the methods used by trial investigators to encode and describe race and ethnicity. It was difficult to ascertain the source of classifications and there was considerable heterogeneity in how race and ethnicity was categorized. More generally, the narrow definitions of race and ethnicity are difficult to reconcile with complex notions of self-identification.⁵⁴  The lack of standardization affects both our aggregate estimates of diversity and ability to integrate across data sources, as in our exploratory analysis of disease burden. The omission of American Indian children because of data gaps exemplifies this issue. We attempted to harmonize definitions of race and ethnicity and address incomplete reporting through extensive manual review and sensitivity analyses.

Secondly, we have previously shown that results from pediatric trials registered on ClinicalTrials.gov are often not submitted in a timely fashion.¹⁵  The inability to analyze these studies, as well as those that disclose their findings without race and ethnicity characteristics, introduces an potential selection bias in our cohort. Concerns have also been raised about the reliance on self-reported data entry and changes to the ClinicalTrials.gov database structure over time.^13,55,56 

Lastly, only trials performed exclusively in the United States were included and our findings are therefore not generalizable to other countries. Nonetheless, as a social construct, the concept of race and ethnicity must be considered within specific cultural and historical contexts.

Although race and ethnicity reporting has considerably improved, shortcomings in representative enrollment among pediatric clinical trials persist. Addressing the scientific and social justice imperatives of clinical trial diversity require complex structural-, community-, and individual-level solutions. The complete documentation of racial and ethnic characteristics represents a crucial first step. As structural racism becomes increasingly recognized as a public health crisis, the time could not be more opportune to hold pediatric clinical trials to a standard that advances the promises of biomedical progress for all children.

CAGR

compound annual growth rates

ECD

enrollment-Census difference

ECR

enrollment-Census ratio

FDA

Food and Drug Administration

IQR

interquartile range

KID

Kids’ Inpatient Database

NSCH

National Survey of Children’s Health