Distributed Counterfactual Modeling Approach for Investigating Hospital-Associated Racial Disparities in COVID-19 Mortality

  1. Mackenzie Edmondson
  2. Chongliang Luo
  3. Nazmul Islam
  4. David Asch
  5. Jiang Bian
  6. Yong Chen

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Several studies have found that black patients are more likely than white patients to test positive for or be hospitalized with COVID-19, but many of these same studies have found no difference in in-hospital mortality. These studies may have underestimated racial differences due to reliance on data from a single hospital system, as adequate control of patient-level characteristics requires aggregation of highly granular data from several institutions. Further, one factor thought to contribute to disparities in health outcomes by race is site of care. Several differences between black and white patient populations, such as access to care and referral patterns among clinicians, can lead to patients of different races largely attending different hospitals. We sought to develop a method that could study the potential association between attending hospital and racial disparity in mortality for COVID-19 patients without requiring patient-level data sharing among collaborating institutions. We propose a novel application of a distributed algorithm for generalized linear mixed modeling (GLMM) to perform counterfactual modeling and investigate the role of hospital in differences in COVID-19 mortality by race. Our counterfactual modeling approach uses simulation to randomly assign black patients to hospitals in the same distribution as those attended by white patients, quantifying the difference between observed mortality rates and simulated mortality risk following random hospital assignment. To illustrate our method, we perform a proof-of-concept analysis using data from four hospitals within the OneFlorida Clinical Research Consortium. Our approach can be used by investigators from several institutions to study the impact of admitting hospital on COVID-19 mortality, a critical step in addressing systemic racism in modern healthcare.

Article activity feed

  1. ScreenIT

    SciScore for 10.1101/2021.09.14.21263596: (What is this?)

    Please note, not all rigor criteria are appropriate for all manuscripts.

    Table 1: Rigor

    NIH rigor criteria are not applicable to paper type.

    Table 2: Resources

    Experimental Models: Organisms/Strains
    SentencesResources
    Our primary interest for this analysis is to investigate the association between attending hospital and in-hospital mortality differences for Non-Hispanic Black (NHB) and Non-Hispanic White (NHW) patients.
    Non-Hispanic White
    suggested: None
    Software and Algorithms
    SentencesResources
    Illustrative Example Using Multi-Hospital Real-World Data: We illustrate our application of the dPQL algorithm to study hospital-associated racial disparities in COVID-19 mortality using data from the OneFlorida Clinical Research Consortium, a centralized data repository comprising data for over 74% of Floridians.
    OneFlorida Clinical Research Consortium
    suggested: None

    Results from OddPub: We did not detect open data. We also did not detect open code. Researchers are encouraged to share open data when possible (see Nature blog).

    Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:
    Due to the limitations of our applied analysis, results presented should not be interpreted clinically. Rather, they are meant to illustrate the type of analysis that can be performed using our counterfactual modeling approach with multi-site data. In future work, it would be worthwhile to compare different approaches for distributed generalized linear mixed modeling to counterfactually model in-hospital mortality. The dPQL method is one approach, but more are likely to become available in the near future. It could also be beneficial to continue to investigate whether the association between attending hospital and racial disparities in mortality has changed over time. Our limited analysis did not suggest a trend in either direction, but analysis of a more complete, heterogeneous collection of patient data would provide more convincing conclusions in this regard. Despite the limitations of our own real-world analysis, we believe this novel application of the dPQL algorithm can be used by researchers as a tool for identifying hospital-level inequities in patient outcomes associated with race. In the event that inequities are present and thought to be related to quality of care, hospital-level interventions may be needed to help close gaps in performance between predominately white and predominately black hospitals. While this would likely take considerable time to accomplish, we hope our method can help to highlight underlying disparities and aid in the process of addressing sy...

    Results from TrialIdentifier: No clinical trial numbers were referenced.

    Results from Barzooka: We did not find any issues relating to the usage of bar graphs.

    Results from JetFighter: We did not find any issues relating to colormaps.

    Results from rtransparent:
    • Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
    • Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
    • No protocol registration statement was detected.

    Results from scite Reference Check: We found no unreliable references.

    About SciScore

    SciScore is an automated tool that is designed to assist expert reviewers by finding and presenting formulaic information scattered throughout a paper in a standard, easy to digest format. SciScore checks for the presence and correctness of RRIDs (research resource identifiers), and for rigor criteria such as sex and investigator blinding. For details on the theoretical underpinning of rigor criteria and the tools shown here, including references cited, please follow this link.

  2. Version published to 10.1101/2021.09.14.21263596v1 on medRxiv