Do hospitals that participate in COVID-19 research differ from non-trial hospitals? A cross-sectional study of US hospitals

The burden of critical illness from coronovirus disease 2019 (COVID-19) took a tremendous adverse toll on hospitals and health systems across the world as most struggled to care for infected patients who were disproportionately from marginalized racial and ethnic minority groups, while at the same time many patients with non-COVID-19 conditions suffered from delayed care. To address these negative outcomes, COVID-19 has heralded a massive effort to study novel pharmaceuticals and technological interventions to reduce mortality and to alter disease course among hospitalized patients [1]. Many of these studies are randomized controlled trials (RCTs), the gold standard for evaluating safety and efficacy [2]. However, these trials have limitations, including the diversity of included patients [3]. Patient diversity in clinical trials has important implications for COVID-19 due to the well-documented excess burden of morbidity and mortality experienced by minority groups [4]. Indeed, recent articles have highlighted geographic disparities in access to COVID-19 clinical trials [5, 6].

Less is known about hospital-level disparities in access to clinical trials in pandemics. Thus, in this study, we describe the hospitals participating in COVID-19 inpatient clinical trials and assess whether there are differences between hospitals that did and did not participate in these trials.

Between April 1 and August 24, 2020, there were 541 hospitals participating in 298 randomized, clinical trials evaluating treatment options for patients hospitalized for COVID-19. After excluding hospitals not available in the RAND Hospital Dataset, there were 488 trial hospitals and 4232 non-trial hospitals. We believe these sites were missing because a small number of hospitals in the USA are not registered with the CMS and several medical centers listed in the clinicial trials are eligible to perform trials but are actually large clinics and not hospitals.

We found that 79.5% of trial hospitals were teaching hospitals, compared to 20.1% of non-trial hospitals. Trial hospitals were also larger (427 beds vs 110.8 beds, p < 0.001). The average case-mix index and disproportionate share percentage for trial hospitals were also higher compared to non-trial hospitals (Table 1). There were also demographic differences: trial hospitals were in counties that had higher annual median incomes ($3.19 vs $2.73), had a lower percentage of white residents, and had a higher percentage of Latinx residents.

We found that, after controlling for other demographic and hospital characteristics, teaching status (OR: 2.11, 95% CI: 1.52–2.95), having a greater number of hospital beds (OR 1.003, 95% CI: 1.003–1.004), and higher Medicare case-mix index (OR 7.48, 95% CI: 4.39–13.1) were independently associated with higher odds of trial participation at the hospital level. Similarly, being in counties with higher annual median incomes (OR 1.32, 95% CI: 1.16–1.51) and lower percentage of White residents (OR 0.98, 95% CI 0.98–0.99) as well as being in the Northeast census region (OR 1.83, 95% CI 1.18–2.85) were each independently associated with statistically significant higher odds of participating in a COVID-19 trial.

We found that early in the COVID-19 pandemic, there were multiple differences between trial and non-trial hospitals. We found differences in geographic access to trials, with hospitals in the Northeast having the highest odds of trial participation [5, 6]. Additionally, we found that larger size, teaching status, and locations in wealthier counties with fewer White residents were each independently associated with higher odds of trial participation during the early stage of the COVID-19 pandemic. We also found that higher patient acuity was also associated with increased odds of trial participation, which we believe is important for inpatient clinical trials focusing on severe disease.

Although we controlled for teaching status, we did not have a marker for major tertiary and quaternary academic centers. As such, some of the differences may be driven by access to academic medical centers. It is possible that academic medical centers may have been the best equipped to mobilize the resources necessary for trial participation [6]. This has important implications for the generalizability of trial results; efficacy and safety outcomes generated at these highly specialized centers may not necessarily be replicated in community centers with fewer resources. Other factors, such as the number of ICU beds and whether or not the hospital has an accident and emergency department, may also affect the presence of clinical trials, which we did not control for.

Our study has limitations. First, we used county-level demographic characteristics as a proxy for the demographics for each hospital’s patient population. While hospitals’ patient populations may not always reflect the surrounding county demographics, we felt that this was a suitable proxy in the setting of the COVID-19 pandemic, where patients may not be traveling as frequently to receive care from further hospitals. However, this may affect the validity of our analysis. Additionally, we limited our study sample to inpatient trials; our results may not be generalizable to trials evaluating outpatient therapeutics. Finally, COVID-19 policies ranging from non-pharmaceutical interventions to hospital policies varied greatly from region to region. For example, restaurant closures and mandatory masking were imposed at different times across regions. Our analysis does not take into account the spatial and temporal differences in these policies.

Our analysis did not find differences in racial compositions, which we believe is due to limitations of our data. Hospitals that participated in clinical trials are clustered in urban regions. These regions are more racially diverse, but hospitals do not necessarily serve the entire population in the geographic regions we considered. Furthermore, income inequality has risen substantially in these more densely populated urban regions such as San Francisco, Los Angeles, and New York City. While the disproportionate share was higher among hospitals that participated in trials, this does not necessarily mean that clinical trials included more individuals from lower socioeconomic backgrounds. In one of the earliest trials, ACTT-1, which studied the benefit of remdesivir among hospitalized patients, 43% of participants idenfied as Black or Hispanic, but data on income or insurance status were not reported [12]. We encourage clinical trials to release demographic information in standardized formats to allow further analysis to be done.

In conclusion, we found that not only are there significant geographic differences in access to trials, there are also significant hospital-level differences, including teaching status, size, and location within wealthier counties. Trials remain heavily concentrated in urban areas, widening the rural–urban gap in access to care. These differences have important implications for the generalizability of trial data. Further efforts should promote the expansion of clinical trial sites to a more diverse set of hospitals, including those in low-income, rural communities such as those in the Southeast and Midwest, to increase generalizability of trial results.