Cross-Sectional Analysis of Clinical Trial Availability and North Carolina Neighborhood Social Vulnerability

PURPOSE:

Residents of communities facing social vulnerability (eg, poverty) have limited access to clinical trials, leaving them susceptible to experiencing poor health outcomes. We examined the association between North Carolina county-level social vulnerability and available multiple myeloma (MM) trials.

METHODS:

Using a novel data linkage between ClinicalTrials.gov, the 2019 American Community Survey, and the Centers for Disease Control and Prevention's Social Vulnerability Index, we investigated at the county level (1) availability of MM trial sites and (2) the relationship between Social Vulnerability Index and MM trial site availability using logistic regression.

RESULTS:

Between 2002 and 2021, 229 trials were registered across 462 nonunique trial sites in 34 counties. Nearly 50% of trial sites were in academic medical centers, 80% (n = 372) of all trials were industry-sponsored, 60% (n = 274) were early-phase, and 50% (n = 232) were for patients with relapsed or refractory MM. Counties with low as opposed to high poverty rates had six times greater odds of having ≥ 1 MM trial sites (odds ratio [OR], 5.60; 95% CI, 1.85 to 19.64; P = .004). Counties with the lowest percentage of Black Indigenous Persons of Color and non-native English speakers had 77% lower odds (OR, 0.23; 95% CI, 0.07 to 0.69; P = .011) of having ≥ 1 trial sites. The effect remained significant after accounting for the presence of five academic medical centers (n = 95; OR, 0.18; 95% CI, 0.05 to 0.6; P = .008) and adjustment for metropolitan, suburban, or rural status (OR, 0.25; 95% CI, 0.07 to 0.81; P = .025).

CONCLUSION:

Counties with the lowest poverty rates had more MM trial sites, whereas those with the lowest percentage of Black Indigenous Persons of Color populations had fewer MM trial sites. Multilevel efforts are needed to improve the availability and access to trials for socially vulnerable populations.

INTRODUCTION

Multiple myeloma (MM) is a disease of aging associated with one of the greatest Black-White disparities in incidence and mortality among all cancer types affecting the US population.^1,2 Adults age ≥ 65 years account for 64% of all new MM cases and 80% of all MM-related deaths.^1,2 Cancer trials, including those focused on MM, set the treatment standards for cancer care and quality. However, restrictive eligibility criteria often result in the exclusion of older adults (age ≥ 65 years).^3-5

Some populations (eg, Black Indigenous Persons of Color [BIPOC] and those facing poverty) are prone to social vulnerability (defined as the ability of specific populations or communities to survive and thrive when faced with external stresses on human health),⁶ which influences their ability to access trials.^7-11 Furthermore, vulnerable populations are also likely to receive care within hospital systems without the infrastructure to support trials.¹² Consequently, trials lack representation by diverse participants reflective of the real-world population. This lack of diversity creates differential treatment patterns that further widen age-related¹³ and race-related disparities in cancer-related health outcomes.¹⁴

Community-level (eg, physical environment,¹⁵ education,¹⁶ and rural locations¹⁷) and individual-level factors (income^10,18 and insurance status¹⁹) drive disparate clinical trial access. However, composite measures such as the Centers for Disease Control Social Vulnerability Index (CDC SVI) and the Area Deprivation Index have been developed to account for the co-occurrence of these factors.^6,20,21 Such aggregate measures are used increasingly to evaluate health outcomes and measure the resiliency of communities when exposed to external stressors.^22-27 The CDC SVI uses American Community Survey data to evaluate community-level (county and census tract) socioeconomic status, household composition/disability, minority status/language, and housing and transportation (Appendix 1, online only).⁶

The existing SVI literature has primarily focused on postoperative surgical outcomes for cancer^25,28 and noncancer-related conditions²² such as diseases involving the cardiovascular²⁹ and renal systems³⁰ and COVID-19 infection.^23,31,32 To our knowledge, no previous studies have examined the availability of cancer-related trials while accounting for community-level SVI. We focused on MM-related trials given the disproportionate burden of MM on older adults and Black persons, with the latter being twice as likely to get and die from MM compared with White persons.¹ Accordingly, in this study, we created an extensive county-level data linkage of all North Carolina (NC) counties and used it to investigate the relationship between the SVI and the number of registered MM clinical trial sites from 2002 to 2021.

METHODS

Data Sources and Linkage

We used Medical Subject Headings terms myeloma and multiple myeloma to identify phase I-IV interventional trials from the ClinicalTrials.gov database, registered in NC between October 30, 2002, and March 1, 2021 (Fig 1).

FIG 1.

Study flow diagram showing the selection of trials from ClinicalTrials.gov. AL, amyloid light-chain.

We aggregated MM trials by county using the ZIP codes of each trial site. We then linked the county location of each trial site to county-level information derived from two external sources (Appendix 1). We derived county-level socioeconomic and demographic data from the 2019 American Community Survey estimates³³ and data for the overall SVI and its four themes from the CDC SVI data set.⁶

To address the questions around the availability of MM trials, we focused on NC because of the state's socioeconomic diversity, with 53% (n = 5,343,700) considered BIPOC.⁶ In addition, on the basis of 2013 (Rural-Urban Continuum Codes [RUCC; metropolitan: 01-03, suburban: 04-06, and rural: 07-09]), 21% of NC counties are considered rural, with 20% of all counties facing the highest overall social vulnerability (CDC SVI score 0.8-1.0).⁶

Social Vulnerability Index

The CDC SVI is a validated measure of 15 community-level social factors derived from the American Community Survey and used to characterize the social vulnerability of US counties or census tracts.⁶ Current SVI estimates are based on 2014-2018 American Community Survey data.⁶ Social vulnerability is characterized by four themes: (1) socioeconomic status, (2) household composition/disability, (3) minority status/language, and (4) housing and transportation (Table 1). These indicators are weighted, yielding a composite score (range, 0-1) for each census tract and county.⁶ We used county-level SVI as the smallest unit for this study because of missing geographic data about MM trial sites in the ClinicalTrials.gov database. County-level SVI scores were categorized into terciles (low, moderate, and high) using cut points: (0, 0.333], (0.333, 0.666], (0.666, 1]. Further information on the publicly available CDC's SVI data sets can be found in the Agency for Toxic Substances and Disease Registry.⁶

TABLE 1.

Social Vulnerability Index Components⁶

Outcome

Outcomes were at the county level and included (1) the availability of MM trial sites within the county and (2) the relationship between the SVI and the availability of MM trial sites registered from 2002 to 2021. We defined trial availability as a binary indicator of whether the county had ≥ 1 MM trial sites.

We used a similar matching algorithm described by Wang et al³⁴ by first matching trials to their respective counties using ZIP codes. We identified counties using Federal Information Processing Standards codes. We linked the ZIP codes of the trial sites to county Federal Information Processing Standards codes using the US Department of Housing and Urban Development US Postal Service crosswalk files (Appendix Fig A1, online only).³⁵ We encountered invalid ZIP codes in the ClinicalTrials.gov database but manually fixed them by looking up the trial site. If the ZIP code was unavailable, we developed a matching algorithm to assign a county to each site using its name, city, and state. Finally, we created a city-county-state list and assigned counties to those sites that did not have a valid ZIP code or site name and had information only on city and state. We encountered no sites with missing data on city and state. Therefore, we treated our data set as complete at the time of data cutoff on March 1, 2021.

Descriptive Methods

We described characteristics of trials for all counties and across SVI terciles; setting (academic medical centers [university- or nonuniversity-affiliated] and nonacademic medical centers); sponsor type (National Institute of Health/Cooperative Group and industry-supported); total study enrollment (1-49, 50-100, 101-200, 201-300, 301-400, 401-500, and > 500); study phase, early phase (phase I/II and phase II) and late phase (phase II/III, phase III, and phase IV); disease phase (newly diagnosed MM, relapsed/refractory, transplant-related, maintenance, smoldering myeloma, and others); and study status (active [not yet recruiting, recruiting, active, and not recruiting], inactive [terminated, suspended, withdrawn, and unknown status], and completed). We also described the characteristics of NC counties according to metropolitan, suburban, or rural classification; other socioeconomic indicators; household composition; race/ethnicity/language; and housing/transportation (Appendix Table A1, online only).

The study followed the guidelines for Strengthening the Reporting of Observational Studies in Epidemiology.³⁶ This study used publicly available data and did not constitute human subjects research; therefore, we did not require Institutional Review Board approval or informed consent.

Statistical Methods

Trial characteristics, census, and American Community Survey data were summarized as frequencies for the entire sample and compared across the 5 and 95 counties with and without academic medical centers, respectively. Dot densities were used to show the spatial distribution of MM trials across NC counties. A priori, we considered a two-sided α level of .05 as statistically significant.

We used quasi-Poisson models to examine the number of MM trials per county.³⁷ To investigate the association of SVI with the availability of MM trials, we used a logistic regression model. We defined the outcome as the county-level availability of ≥ 1 MM trial sites and the predictor of interest as the county's overall SVI and each of the four themes (socioeconomic status, household composition, minority/language status, and housing/transportation). We compared SVI groups using high social vulnerability as the reference group. We adjusted for rural-urban status on the basis of 2013 RUCC estimates without adjusting for the following SVI components: county population age ≥ 65 years versus < 65 years and race/ethnicity.

To account for the likelihood that trial sites are more likely to be in an academic medical center, we conducted a sensitivity analysis excluding five counties with major academic medical centers (Durham, Mecklenburg, Forsyth, Orange, and Pitt counties). We also evaluated the association between SVI and industry-sponsored trials on the basis of a priori knowledge of older adults being less likely than their younger counterparts to enroll in industry-sponsored trials.³⁸ All geographic analyses (geocoding, geographic linkage, and figure creation) were performed using ArcGIS 10.3 (Esri, Redlands, CA), data cleaning and management were performed using SAS (version 9.5, Cary, NC), and statistical analyses were performed using R version 4.0.2 (R Core Team, 2020, Vienna, Austria).³⁹

RESULTS

County and Clinical Trial Characteristics

Between 2002 and 2021, we initially identified 2,842 trials and excluded 2,613 (Fig 1). Table 2 summarizes trial site characteristics for 229 MM trials opened across 462 nonunique trial sites according to county-level social vulnerability (low, moderate, and high). Thirty-four percent of counties had registered MM trial sites (Fig 2). We considered the data set comprehensive for our analyses, treating the remaining 66 counties with no registered trials with a zero trial count. Counties with high overall social vulnerability had the lowest number of trial sites compared with those with moderate and low social vulnerability, and the former also had the lowest number (11) of academic medical centers (Fig 2 and Table 2). Nearly 50% of trial sites were in academic medical centers. Eighty-two percent (n = 379) of registered trial sites offered were industry-sponsored, and 52% (n = 197) of these were in an academic medical center. Across all registered trial sites, 60% (n = 274) offered early-phase trials and 50% (n = 232) had registered trials for persons with relapsed or refractory MM. Among the trial sites with early-phase studies, 88% were registered in counties with either moderate or low social vulnerability (Table 2).

TABLE 2.

Trial and Disease Characteristics by SVI

FIG 2.

(A) Dot density map showing myeloma trial sites across North Carolina counties between 2002 and 2021, overall social vulnerability. (B) Availability map showing myeloma trial sites across North Carolina counties 2002-2021, minority status and language theme. (C) Availability map showing myeloma trial sites across North Carolina counties between 2002 and 2021, overall social vulnerability. (D) Context map showing the major cities and urban areas in North Carolina. MM, multiple myeloma; SVI, Social Vulnerability Index.

Counties (5) with academic medical centers had higher population densities than those without (n = 95), but the latter included larger numbers of low-income, higher-poverty, and less-educated individuals. However, counties with academic medical centers had more BIPOC individuals, multiunit structures, and group quarters than those without academic medical centers (Appendix Table A1).

Association Between SVI and MM Trial Availability

There were, on average, five MM trial sites per county during our study period. However, five counties (Durham, Forsyth, Mecklenburg, Orange, and Pitt) with academic medical centers had a combined average of 61 MM trial sites per county (Appendix Table A1).

In unadjusted analyses that included all 100 NC counties, we found a statistically significant association in the odds of having ≥ 1 MM trial sites for the overall SVI and the socioeconomic status and minority/language themes. Specifically, for the overall SVI, the odds were three times greater (odds ratio [OR], 3.10; 95% CI, 1.08 to 9.57; P = .04) if the county had moderate overall social vulnerability as opposed to high overall social vulnerability, but six times greater (OR, 5.60; 95% CI, 1.85 to 19.64; P = .004) if the county had low as opposed to high poverty rates (Table 3). The odds were three times greater for counties with moderate poverty rates and socioeconomic challenges (OR, 3.20; 95% CI, 1.02 to 11.37; P = .055) that at least one site offered a MM trial. Counties with the lowest percentage of BIPOC persons and non-native English speakers had 77% lower odds (OR, 0.23; 95% CI, 0.07 to 0.7; P = .011) of having ≥ 1 MM trial sites than those with the highest percentage of these persons (Table 3). After adjusting for metropolitan, suburban, or rural status, the odds of having ≥ 1 MM trial site remained statistically significant for counties with the lowest percentage of BIPOC persons and non-native English speakers (OR, 0.25; 95% CI, 0.07 to 0.81; P = .025) compared with those with the greatest percentage of these persons (Table 3).

TABLE 3.

Unadjusted and Adjusted Association Between County-Level Availability of ≥ 1 MM Trial Sites and SVI Among All Counties (n = 100)

We observed similar results in sensitivity analyses (unadjusted and adjusted) conducted to account for the expected clustering of trial sites in the five counties with academic medical centers (Appendix Table A2, online only). One exception was the socioeconomic status theme; in unadjusted analyses, the statistically significant association for counties with moderate vulnerability was no longer present (OR, 2.57; 95% CI, 0.79 to 9.34; P = .128; Appendix Table A2). Results of unadjusted and adjusted analyses also remained similar when accounting for the presence of MM trial sites offering industry-sponsored trials (Appendix Table A3, online only). To address the high variability in MM trial sites, we conducted a quasi-Poisson regression using the number of registered MM trial sites as the outcome (Appendix Table A4, online only). Results revealed that counties with the lowest percentage of BIPOC persons and non-native English speakers had, on average, 92% fewer trial sites (incident rate ratio, 0.08; 95% CI, 0.01 to 0.31; P = .003) than those with the highest percentage of these individuals. Furthermore, counties with the fewest multiunit housing structures, mobile homes, group quarters, and homes that were less crowded and had the highest access to a vehicle had 76% fewer trial sites (incident rate ratio, 0.24; 95% CI, 0.04 to 0.96; P = .067; CIs are from profile likelihood and P value from Wald statistics) as compared with counties where there were more multiunit housing structures, mobile homes, group quarters and overcrowded homes, and less access to a vehicle.

Geographic Distributions

County-level dot maps show high-density clustering of trials around five counties with academic medical centers (Durham [SVI: 0.62], Forsyth [0.68], Mecklenburg [0.44], Orange [0.30], and Pitt [0.82]), with no clear trend in clustering according to overall social vulnerability scores (Fig 2). We used a permutation test for Moran's I to test spatial autocorrelation in the Pearson residuals of a logistic model of study availability predicted by overall social vulnerability and metro area status and found no evidence of spatial autocorrelation (Moran's I statistic = 0.420, P = .340).

DISCUSSION

We investigated the distribution of MM-related interventional trials in NC and whether the availability of a MM trial site was associated with the county's overall SVI and each of the four SVI themes, given existing disparities in the enrollment of socially vulnerable populations in trials. We created a novel, extensive data linkage between ClinicalTrials.gov, the CDC's SVI data set, and the 2019 American Community Survey to study the availability of MM interventional trials across NC counties. After adjusting for metropolitan, suburban, or rural status, we found that counties with the lowest population of BIPOC and non-native English speakers had the lowest odds of having ≥ 1 MM trial sites. In addition, counties with the lowest poverty rates and the fewest socioeconomic challenges had the greatest odds of having ≥ 1 MM trial sites.

To our knowledge, the association between area-level social vulnerability and clinical trial availability has not been previously reported. However, previously published literature has demonstrated the association between the CDC's SVI and health-related outcomes and persistent health and health care disparities in the United States.^{25,28,29,32,40,41} The existing literature has also identified potential barriers to trial enrollment of BIPOC and other socially vulnerable groups (eg, older adults) at the individual and interpersonal levels.^42,43 For example, individuals who are uninsured/underinsured, living in rural locations, and unemployed have a lower level of education and are less likely to enroll in trials.^{10,16,18,27,44,45} In addition, BIPOC and older adults face additional barriers to clinical trial enrollment because of a lack of awareness of trial enrollment opportunities, a legacy of medical mistrust, fear of the health care system, and strict age and performance status–based eligibility criteria.^42,46

Consistent with these reports, our study adds to the current knowledge of factors that may contribute to the under-representation of BIPOC and older and other socially vulnerable adults in MM trials by showing that the availability and location of trials (eg, in an academic medical center) are critical determinants. In addition, the observation that 34 of 100 counties had registered trials suggests additional geographic barriers to MM trial participation. Despite our inability to account for travel across county lines to access a MM trial site, we observed a paucity of trial sites in the eastern counties of NC (Fig 1). This finding has important implications as most counties in Eastern NC have either moderate or high social vulnerability (all themes), suggesting more socially vulnerable populations. These findings are critical for informing decision making about clinical trial sites and ensuring that trials are made available at locations accessible by vulnerable groups. It also emphasizes the need for tailored clinical trial outreach efforts like those described by Woodcock et al and Unger et al to enhance the enrollment of these vulnerable populations.^47,48

Our study highlights the role of community-level social indicators in trial availability. It builds on the recent findings from the study by Unger et al examining survival and community-level area deprivation on the basis of the area deprivation index,²¹ a similar area-level metric as the CDC SVI. This study showed that among 41,109 patients enrolled in 55 cancer trials, residing in areas of highest social vulnerability (area deprivation index, 80%-100%) was associated with worse overall survival, irrespective of disease-related prognostic risk, insurance status, or whether they resided in a rural or urban community.²⁷

In our secondary analyses, we included metropolitan-urban-rural status (2013 RUCC data) in the models and found that the effect of socioeconomic status was no longer present across 100 counties. After excluding counties with academic medical centers, the effect of socioeconomic status was no longer significant. However, we believe that these results may be misleading for the following reasons: (1) rural counties are more likely to have a smaller population density, (2) they are less likely to have an academic medical center, and (3) rural residents are more likely to be unemployed and have less postsecondary education and lower median household incomes, and are either underinsured or uninsured than urban residents.⁴⁹ We also hypothesize that some of these factors contributed to findings that counties with the lowest percentage of BIPOC and non-native English speakers had 77% lower odds (OR, 0.23; 95% CI, 0.07 to 0.7; P = .011) compared with those with the highest percentage of these individuals. These findings highlight the difficulty of disentangling the effects of metropolitan-urban-rural status and socioeconomic status and the role of additional individual-level, interpersonal-level, organizational-level, and community-level factors that influence whether BIPOC persons can access available trials.⁵⁰

Our study has some limitations. First, our study used publicly available data from ClinicalTrials.gov, which relies on sponsors and principal investigators of studies accurately reporting complete data related to their trials. In addition, descriptive information about individual participants enrolled in each trial either was unavailable because of incomplete records or was not routinely captured by the data set in ClinicalTrials.gov. Such variables include patient-level socioeconomic factors, demographics, and clinical and disease-related factors of those enrolled in the study. We were, therefore, unable to adjust for age, sex, race, ethnicity, education, socioeconomic status, or insurance status at the clinical trial site level. As a result, adjusting for the truly eligible clinical trial population was impossible. Second, the cross-sectional study design does not allow for causal, individual-level inferences to be drawn about county-level social vulnerability and the number of available MM trials. Third, our study included studies registered between 2002 and 2021; however, the CDC SVI⁶ relies on the 2014-2018 American Community Survey data. Therefore, our study design cannot account for changes in population demographics, population density, or socioeconomic factors. Finally, we limited our analysis to NC and MM trials, enrolling those age ≥ 65 years. These decisions were based on the diversity of the NC population and a priori knowledge about MM-related disparities. Therefore, our study's findings may not be generalizable to other clinical trial sites enrolling participants with other cancer types and age groups.

In conclusion, this study adds to a growing body of literature highlighting the barriers to accessing trials for socially vulnerable groups. Using a novel data linkage and a composite score of a community's social vulnerability, we demonstrate that counties with the highest socioeconomic status had the greatest odds of having a MM trial site. After accounting for metropolitan-urban-rural status, counties with the lowest percentages of BIPOC and non–English-speaking residents had the lowest odds of having ≥ 1 MM trial sites. The design of future MM-related clinical trials should account for these barriers by developing tailored strategies to ensure the availability of trial sites in communities with large percentages of socially vulnerable populations. This study also highlights the need for multilevel approaches to increase participant diversity in MM trials, such as educational activities to increase knowledge and awareness about trial opportunities in these vulnerable communities. Furthermore, since cancer trials generally lack adequate representation from socially vulnerable groups, future studies should examine the association between community social vulnerability and the availability of clinical trials for other cancer types. Such research is critical to ensuring broad clinical trial representation by those who are most likely to use these novel therapies in the real world.

APPENDIX 1. Results

Analyses of the association between Social Vulnerability Index and industry-sponsored trials (Appendix Table A3) revealed results similar to the overall cohort. As an example, adjusted analyses showed that counties with the lowest percentage of Black Indigenous Persons of Color individuals and non-native English speakers had a 75% lower odds (odds ratio, 0.25; 95% CI, 0.07 to 0.81; P = .025) of having at least one site offering a multiple myeloma (MM) trial as compared with those with the highest percentage of these individuals. To address the high variability in MM trial sites, we conducted a quasi-Poisson regression using the number of registered MM trial sites as the outcome (Appendix Table A4). Results revealed that counties with the lowest percentage of Black Indigenous Persons of Color individuals and non-native English speakers had, on average, 92% fewer trial sites (incident rate ratio, 0.08; 95% CI, 0.01 to 0.31; P = .003) than those with the highest percentage of these individuals. Furthermore, counties with the fewest multiunit housing structures, mobile homes, group quarters, and homes that were less crowded and had the highest access to a vehicle had 76% fewer trial sites (incident rate ratio, 0.24; 95% CI, 0.04 to 0.96; P = .067; CIs are from profile likelihood and P value from Wald statistics) as compared with counties where there were more multiunit housing structures, mobile homes, group quarters and overcrowded homes, and less access to a vehicle.

FIG A1.

Flowchart of data linkage. We linked registered clinical trial sites from the ClincalTrials.gov database to county-level characteristics from the 2019 American Community Survey and the CDC ASTDR 2018 SVI data set. ASTDR, Agency for Toxic Substances and Disease Registry; CDC, Centers for Disease Control; HUD, US Department of Housing and Urban Development; SVI, Social Vulnerability Index.

TABLE A1.

Characteristics of NC Counties, All Counties, and 95 Counties Without and Five Counties With Academic Medical Centers

TABLE A2.

Unadjusted and Adjusted Association Between County-Level Availability of ≥ 1 MM Trial Sites and SVI Among 95 Counties

TABLE A3.

Unadjusted and Adjusted Association Between County-Level Availability of ≥ 1 Industry-Sponsored MM Trial Sites and SVI Among All Counties (n = 100)

TABLE A4.

Unadjusted and Adjusted Association Between the County-Level Number of MM Trial Sites and SVI Among All Counties (n = 100)

DATA SHARING STATEMENT

The data underlying this article are available in publicly available sources, including the ClinicalTrials.gov database at https://clinicaltrials.gov/, 2018 CDC/ATSDR Social Vulnerability Index at https://www.atsdr.cdc.gov/placeandhealth/svi/index.html, and 2019 American Community Survey 2019 at https://www.census.gov/programs-surveys/acs.

AUTHOR CONTRIBUTIONS

Conception and design: Shakira J. Grant, Tzy-Mey Kuo, Tanya M. Wildes, Eben I. Lichtman, Marjory Charlot

Collection and assembly of data: Shakira J. Grant

Data analysis and interpretation: Shakira J. Grant, Matthew Jansen, Tzy-Mey Kuo, Samuel M. Rubinstein, Sascha A. Tuchman, Hyman B. Muss, Marjory Charlot

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Cross-Sectional Analysis of Clinical Trial Availability and North Carolina Neighborhood Social Vulnerability

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/op/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).