Where Inequities Emerge: Racial and Ethnic Differences Across the COVID-19 Hospitalization Continuum

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The study contributes valuable data and is generally well organized, with a clear research objective and appropriate use of statistical analyses. The manuscript would benefit from the suggestions below. With these revisions, this work has the potential to make a meaningful contribution to the literature.

1. The statistical approach appears appropriate overall, but some aspects require clarification. Please specify how continuous variables were handled (e.g., normalization, categorization). Clarify whether assumptions for the statistical models were tested. If multiple comparisons were performed, discuss whether any correction was applied or justify why it was not necessary. Consider adding sensitivity or subgroup analyses if feasible, particularly for key findings.

2. The Discussion appropriately summarizes the findings but could be strengthened. Expand comparison with prior studies, highlighting consistency and discrepancies. A comparison with prior analysis on differences in mortality data (PMID: 41034766) will enhance the discussion.

Discuss potential mechanisms underlying the observed associations. Consider elaborating on the public health or policy implications, particularly for environmental or population-level interventions.

3. The limitations section is present, but could be more comprehensive. Please explicitly address residual confounding, selection bias, and generalizability. If the study population is geographically or demographically specific, clarify how this may affect external validity. Discuss how future studies could address these limitations.

Author Response

The study contributes valuable data and is generally well organized, with a clear research objective and appropriate use of statistical analyses. The manuscript would benefit from the suggestions below. With these revisions, this work has the potential to make a meaningful contribution to the literature.

1. The statistical approach appears appropriate overall, but some aspects require clarification. Please specify how continuous variables were handled (e.g., normalization, categorization). Clarify whether assumptions for the statistical models were tested. If multiple comparisons were performed, discuss whether any correction was applied or justify why it was not necessary. Consider adding sensitivity or subgroup analyses if feasible, particularly for key findings.

Thank you for this helpful request for clarification regarding the statistical approach. We have revised the Methods section to more explicitly describe the handling of continuous variables, assessment of model assumptions, and the rationale for not applying multiple-comparison corrections.

Continuous variables were handled using established, pre-specified categorizations where appropriate (e.g., age groups, Charlson Comorbidity Index categories), consistent with prior literature and interpretability considerations. Length of stay was treated as the time scale in survival analyses. As a sensitivity analysis, we also repeated key models using a continuous Charlson Comorbidity Index specification, with consistent findings (Appendix Table A4).

Assumptions for regression models were evaluated using standard diagnostic approaches. For Cox proportional hazards models, proportional hazards assumptions were assessed using log–log survival plots and Schoenfeld residuals; no meaningful violations were detected.

We did not apply formal multiple-comparison corrections because the analyses were hypothesis-driven, focused on a limited number of pre-specified primary contrasts across the hospitalization continuum, and interpreted using effect estimates and confidence intervals rather than exploratory significance testing.

Finally, several sensitivity analyses were already conducted to assess robustness, including length-of-stay restrictions, alternative comorbidity specifications, and alternative propensity score–matching approaches; these are described in the Methods and presented in the Appendix. We have clarified this in the revised manuscript text.

Updates:

“Continuous variables were categorized where appropriate to increase interpretability and consistency with prior health services and COVID-19 outcomes research.” (lines 172-174).

“Proportional hazards assumptions for Cox models were assessed using standard diagnostic methods, including log–log survival plots and Schoenfeld residuals, with no substantive violations observed. This was a hypothesis-driven comparative analysis; therefore, multiple testing corrections were not applied to avoid overly conservative inference.” (line 208-212).

2. The Discussion appropriately summarizes the findings but could be strengthened. Expand comparison with prior studies, highlighting consistency and discrepancies. A comparison with prior analysis on differences in mortality data (PMID: 41034766) will enhance the discussion.

Thank you for this suggestion. We have revised the Discussion to strengthen contextualization of our findings within the existing literature and to more explicitly articulate potential mechanisms and public health implications. Specifically, we clarified how the observed racial and ethnic inequities align with national patterns of COVID-19 hospitalization and outcomes, while emphasizing the added contribution of examining inequities across successive stages of the hospitalization continuum.

We also expanded discussion of plausible upstream mechanisms—including occupational exposures, housing conditions, structural barriers to care, comorbidity burden, and healthcare-seeking patterns—that may contribute to disparities observed at the point of hospital admission. In addition, we strengthened the public health and policy implications by highlighting the importance of population-level and environmental interventions, such as workplace protections, timely outpatient evaluation, vaccination, and early treatment, particularly for Medicaid populations facing structural barriers to care.

These revisions enhance the interpretive clarity and public health relevance of the Discussion without altering the study design, analyses, or conclusions.

Updates (please as a whole):

“For instance, the distribution of COVID-19 hospitalizations by race and ethnicity in this study mirrored patterns observed nationally during the pandemic [6,19]. Greater exposure risk stemming from structural inequities—including frontline and essential occupations, crowded living conditions, and barriers to timely outpatient care—likely contributed to higher infection risk and subsequent hospitalization among non-Hispanic Black and Hispanic communities [20]. National surveillance has repeatedly documented substantially higher COVID-19 hospitalization rates among non-Hispanic Black, American Indian/Alaska Native, and Hispanic populations com-pared with non-Hispanic White populations during the early pandemic period [6,19].

Importantly, within Kentucky’s Medicaid population, these patterns persisted despite uniform insurance coverage, suggesting that coverage alone was insufficient to mitigate disparities in severe COVID-19 illness requiring hospitalization. This underscores the role of upstream population- and environmental-level factors—such as occupational exposures, housing conditions, structural barriers to care, comorbidity burden, and healthcare-seeking patterns—in shaping which patients arrive at the hospital with COVID-19.” (lines: 370-377).

3. The limitations section is present, but could be more comprehensive. Please explicitly address residual confounding, selection bias, and generalizability. If the study population is geographically or demographically specific, clarify how this may affect external validity. Discuss how future studies could address these limitations.

Thank you for this helpful suggestion. We have revised the Limitations section to more explicitly address residual confounding, potential selection bias related to hospitalization-based analyses and differences in severity at admission, and limitations to generalizability given the focus on Medicaid beneficiaries in Kentucky during the early pandemic period. We also added a brief statement outlining how future studies incorporating clinical severity measures, longitudinal patient-level data, and broader geographic coverage could further clarify these findings.

Updates:

“Third, hospitalizations—not individual patients—were the unit of analysis. Repeating admissions by the same individual could contribute multiple observations and may influence estimates despite adjustment. Differences in the severity of illness at the time of hospital admission across racial and ethnic groups may contribute to selection bias, particularly when examining disparities at the point of hospitalization.” (lines: 444-448).

“Finally, because the analysis was limited to Medicaid beneficiaries in Kentucky, findings may not be fully generalizable to other populations or geographic settings. Future studies incorporating patient-level longitudinal data, clinical severity measures, and broader geographic coverage may further clarify the mechanisms underlying observed disparities.” (lines: 457-461).

Reviewer 2 Report

Comments and Suggestions for Authors

Overall, this is a well-written paper with a clear message and a straightforward methodology. It provides a very good explanation of the sample size, a solid review of related literature, and a clear description of the technology used.

I have a few minor suggestions. The first relates to the regression model: it would be interesting if the author could interact the ratio category with the CCI score, particularly for CCI scores greater than three. As I see in Table 2, scores between 1 and 2 are not significant, so these could serve as a reference group. We could also examine scores greater than five for comparison. The idea would be to see whether patients with higher comorbidity scores still do not receive hospitalization or care.

My second comment concerns Figure 2. There appear to be slight outliers in the data; perhaps limiting the hospital stay to 150 or 180 days would improve the shape of the figure.

Finally, I would like to understand the message of Table 4. Although the results are not statistically significant, what do they mean? I suggest that the authors describe or discuss this in the discussion section.

Author Response

Overall, this is a well-written paper with a clear message and a straightforward methodology. It provides a very good explanation of the sample size, a solid review of related literature, and a clear description of the technology used.

I have a few minor suggestions. The first relates to the regression model: it would be interesting if the author could interact the ratio category with the CCI score, particularly for CCI scores greater than three. As I see in Table 2, scores between 1 and 2 are not significant, so these could serve as a reference group. We could also examine scores greater than five for comparison. The idea would be to see whether patients with higher comorbidity scores still do not receive hospitalization or care.

 Thank you for this thoughtful suggestion. While we agree that effect modification by comorbidity burden is an important question, the proposed interaction analyses would extend beyond the scope of the current study’s pre-specified aims and hypothesis-driven analytic framework.

In this analysis, all models were designed to estimate population-average associations across stages of the hospitalization continuum among already-hospitalized patients. The study design does not allow assessment of whether individuals “received hospitalization or care,” as non-hospitalized individuals were not observed.

To preserve analytic consistency and avoid post hoc model expansion, we did not introduce additional interaction terms. We note this as a valuable direction for future research using data sources that capture pre-hospitalization care pathways.

My second comment concerns Figure 2. There appear to be slight outliers in the data; perhaps limiting the hospital stay to 150 or 180 days would improve the shape of the figure.

 Thank you for this observation. Figure 2 presents unadjusted Kaplan–Meier survival curves and includes the full observed range of hospital length of stay, with discharge alive treated as a censoring event.

Importantly, the primary separation in survival curves occurs early during hospitalization, and findings were robust in sensitivity analyses that excluded very short (<1 day) and long (>60 days) stays. Because the figure is descriptive and intended to reflect the observed data, we retained the full range while addressing potential outliers through sensitivity analyses rather than truncation of the display.

Finally, I would like to understand the message of Table 4. Although the results are not statistically significant, what do they mean? I suggest that the authors describe or discuss this in the discussion section.

Thank you for this comment. We have clarified the interpretation of Table 4 in the Discussion. Although estimates were not statistically significant, these findings indicate that, after adjustment and confirmatory propensity score matching, racial differences in in-hospital mortality were not observed among patients hospitalized with COVID-19.

This pattern supports the interpretation that inequities in this Medicaid population were more pronounced at the point of hospitalization and in the distribution of COVID-19 admissions, rather than during COVID-19–specific inpatient mortality. The Discussion now explicitly addresses this interpretation and places the null findings in the context of prior literature.