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

Daniel Kang, Cher X. Huang, Alexander Yuen, Keith C. Norris, Tara VijayanTrials 2023 · DOI

TL;DR

This cross-sectional study reveals that US hospitals participating in COVID-19 clinical trials are systemically different from those that do not. Trial-participating hospitals tend to be larger, teaching-oriented institutions located in wealthier, more urban, and more diverse counties, raising significant questions about the "real-world" generalizability of trial results to smaller community or rural hospitals.

What Problem Does This Paper Address?

When a new drug is tested in a clinical trial, the goal is to determine if it works and is safe. However, the "gold standard" randomized controlled trial (RCT) often takes place in a controlled environment. If the hospitals conducting these trials are fundamentally different from the hospitals where most people actually receive care, the results might not translate perfectly to the general population—a concept known as external validity or generalizability.

While previous research has highlighted disparities in which patients get enrolled in trials, this paper shifts the lens to the institutions themselves. The authors sought to determine if there was a structural bias in which hospitals were able to mobilize and host COVID-19 therapeutic trials during the first wave of the pandemic (April–August 2020).

Methods

The researchers conducted a cross-sectional study comparing two groups: hospitals that participated in at least one interventional COVID-19 inpatient trial and those that did not.

To identify these differences, they employed multivariable logistic regression modeling. This statistical approach allowed them to isolate the impact of specific hospital traits (like size or teaching status) while controlling for other variables. They also used the Bonferroni correction to account for multiple hypothesis testing, ensuring that their findings weren't just the result of statistical "noise."

Datasets

The study integrated three primary data sources to create a comprehensive picture of the US hospital landscape:

  • ClinicalTrials.gov: Used to identify 298 interventional trials and the 488 hospitals participating in them.
  • 2019 RAND Hospital Dataset: Provided institutional metrics, including the number of beds, teaching status, and the Case-Mix Index (CMI)—a proxy for how "sick" the average patient is at that hospital.
  • 2019 American Community Survey (ACS): Provided county-level demographic data (median income, racial/ethnic makeup, and urban/rural designations) linked via Federal Information Processing System (FIPS) codes.

Key Insights and Ideas

  • The "Acuity" Driver: The strongest predictor of trial participation was the Case-Mix Index. Hospitals treating the most severely ill patients were over seven times more likely to host a trial. While logical for a respiratory pandemic, it suggests trial data is heavily skewed toward the most extreme clinical cases.
  • The Academic Advantage: Teaching hospitals were significantly more likely to participate. These institutions typically have the administrative infrastructure (IRBs, research coordinators) to pivot quickly during a crisis, whereas community hospitals may lack the "red tape" bandwidth to join trials.
  • The Wealth and Diversity Paradox: Trial hospitals were more likely to be in counties with higher median incomes but lower percentages of White residents. This suggests that while trials are happening in diverse urban centers, they are still anchored in high-resource geographic "hubs."
  • Geographic Concentration: The Northeast US had a disproportionately high rate of trial participation, likely reflecting both the early intensity of the pandemic in that region and its high density of academic medical centers.

Key Results

The study compared 488 trial hospitals against 4,232 non-trial hospitals. Key findings from the adjusted odds ratios (OR) include:

  • Teaching Status: Teaching hospitals had 2.11 times the odds of participating (95% CI 1.52–2.95).
  • Patient Acuity: High Medicare case-mix index was associated with 7.48 times higher odds of participation (95% CI 4.39–13.1).
  • Hospital Size: Larger hospitals (more beds) were slightly more likely to participate (OR 1.003 per bed).
  • County Wealth: For every $10,000 increase in median annual county income, the odds of trial participation increased by 32% (OR 1.32).
  • Race: Hospitals in counties with more White residents had slightly lower odds of participation (OR 0.98).

Limitations and Future Work

The authors acknowledge several limitations:

  • Proxy Demographics: The study used county-level data as a proxy for hospital patient populations. A hospital in a diverse county might still serve a non-diverse subset of that population.
  • Scope: The analysis was limited to inpatient trials; results might differ for outpatient or vaccine trials.
  • Unmeasured Variables: Factors like the specific number of ICU beds or the presence of specialized emergency departments were not controlled for.

Future Work: The authors argue for a deliberate expansion of clinical trial sites into low-income, rural, and non-teaching community hospitals. Without this expansion, the medical community risks developing "ivory tower" treatments that may not perform as expected in the resource-constrained environments where the majority of patients are treated.

Summary updated

Abstract

Hospitals participating and not participating in COVID-19 inpatient treatment clinical trials differed in many ways, resulting in important implications for the generalizability of trial data.
BibTeX
@inproceedings{kang2023do,
  author = {Daniel Kang and Cher X. Huang and Alexander Yuen and Keith C. Norris and Tara Vijayan},
  title = {{Do hospitals that participate in COVID-19 research differ from non-trial hospitals? A cross-sectional study of US hospitals}},
  booktitle = {Trials},
  year = {2023},
  doi = {10.1186/s13063-023-07450-6},
}
Figure