Explaining COVID-19 related mortality disparities in American Indians and Alaska Natives

American Indian and Alaska Native (AI/AN) individuals are more likely to die with COVID-19 than other groups, but there is limited empirical evidence to explain the cause of this inequity. The objective of this study was to determine whether medical comorbidities, area socioeconomic deprivation, or access to treatment can explain the greater COVID-19 related mortality among AI/AN individuals. The design was a retrospective cohort study of harmonized electronic health record data of all inpatients with COVID-19 from 21 United States health systems from February 2020 through January 2022. The mortality of AI/AN inpatients was compared to all Non-Hispanic White (NHW) inpatients and to a matched subsample of NHW inpatients. AI/AN inpatients were more likely to die during their hospitalization (13.2% versus 7.1%; odds ratio [OR] = 1.98, 95% confidence interval [CI] = 1.48, 2.65) than their matched NHW counterparts. After adjusting for comorbidities, area social deprivation, and access to treatment, the association between ethnicity and mortality was substantially reduced (OR 1.59, 95% CI 1.15, 2.22). The significant residual relation between AI/AN versus NHW status and mortality indicate that there are other important unmeasured factors that contribute to this inequity. This will be an important direction for future research.


Study design
The COVID EHR Cohort at the University of Wisconsin (CEC-UW; 30 ) is a retrospective cohort study supported by the National Cancer Institute (ClinicalTrials.govNCT04506528) that included 21 health systems from across the US (see Figure S1 in Supplemental Materials).Data extractions were performed using customized extraction code altered to accommodate unique health system specific EHR features.Each data extraction captured data on new patients meeting inclusion criteria and follow-up data on existing cohort members.Participating health systems provided selected data elements from the EHR of all COVID-19 patients encountered during the study period (February 1, 2020 to January 31, 2022).Data were transferred to the CEC-UW Coordinating Center in Madison, Wisconsin, where they were harmonized and merged.Harmonization, merging, and data analysis occurred September 30, 2021 through July 3, 2023.This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines 31 .

Ethics statement
The CEC-UW study was initially approved in May 2020 by the University of Wisconsin-Madison Health Sciences Minimal Risk Institutional Review Board (MR-IRB) with approval for the collection of de-identified EHR data from the 21 health systems.The MR-IRB also determined that the study met criteria for a human subjects research exemption and qualified for a waiver of informed consent under the Federal Common Rule.All participating health systems provided written notice of either their own institution's IRB approval or determination of exemption status before sharing EHR data.In February 2021, the MR-IRB approved a change of protocol for a Limited Data Set, allowing the collection of additional information (e.g., death dates, five-digit zip codes) but excluding direct patient identifiers.Each patient in the data set from each health system was assigned an enduring cryptographically processed Patient ID based on the SHA256 algorithm, which yielded a 64-character unique and private hash-based message authentication code (HMAC).

Analysis samples
The full CEC-UW inpatient cohort included 145,944 adult patients hospitalized with COVID-19 who had prior contact with the health system and who completed their hospitalization from February 1, 2020 through January 31, 2022 at a participating health system 30 .
This study focused on three analysis samples.One sample comprised all CEC-UW inpatients who identified as AI/AN (N = 546) [78% were non-Hispanic]) or NHW (N = 78,128).The other two were matched samples

Primary outcome
The primary outcome was all-cause in-hospital mortality during the index COVID-19 hospitalization.Because cause of death was not extracted from the EHRs of the participating health systems, we could not definitively attribute it to COVID-19 as some patients could have died during their hospitalization from other causes.Also reported were rates of three severity indicators (admission to the intensive care unit, intubation for ventilator use, and days in the hospital [among patients who did not die prior to their discharge]).The mean age at death was reported among those who died.

Patient characteristics
The following patient characteristics were extracted from the electronic heath records: sex, age, cigarette smoking status, co-occurring medical conditions (obesity, Type 2 diabetes, kidney, liver, and heart disease, cancer, alcohol use disorder, and drug use disorder [see Supplemental Text S1 for the ICD-10 codes corresponding to the medical conditions]), insurance type, receipt of antiviral medication during the hospitalization (see Supplemental Text S2 for a list of the antiviral medications prescribed), and COVID-19 vaccination status (see Supplemental Text S3).The date of the index hospitalization was extracted to incorporate a variable indicating the year in which the hospitalization occurred.(Note that although it is included as a race category in the EHR, we do not refer to AI/AN as a race based on the recommendation of our Indigenous consultant.The rationale is that AI/ANs are represented by 574 federally recognized tribes that are sovereign political governments and are not a single unified group.) Additional variables were linked to the EHR data based on the patient's home ZIP code: region (Northeast, Midwest, South, and West), urbanicity 33 , area social deprivation 34 , and distance to hospital.Urbanicity was based on Rural-Urban Commuting Area (RUCA) codes 33 ; codes 1-6 were classified as "urban", and codes 7-10 were classified as "rural."Area social deprivation was based on seven census-based indicators of deprivation (see Table S1 in the Supplemental Materials) that were combined via factor analysis 34,35 (see Supplemental Table S2).The area social deprivation index was calculated at the ZIP code tabulation area level; results are presented for each quintile based on the full sample.The ZIP code of the facility in which treatment was received was used to compute the distance to treatment using a SAS function that calculated the geodetic distance in miles between the centroids of two ZIP code locations, in this case the home and treatment facility ZIP codes.Because distance to treatment was very skewed and kurtotic, it was dichotomized at 60 miles, a distance that represents a significant barrier to getting timely treatment (that is, about 60 min travel time) while still including adequate numbers in each group for analyses.
These variables were included as covariates in adjusted multivariable models (covariate categorizations are given in Table 2).

Statistical analysis
Matching was conducted to create samples that were aligned on key variables while allowing for variability in important predictors.Two different matched samples were created.One set of matched samples (Match 1) was matched on age, region, month/year, and sex.The other set of matched samples (Match 2) was matched on region, month/year, and sex (not age).Because age was confounded with AI/AN versus NHW status in the full sample (see Table 2), NHW patients were matched to the AI/AN patients based on age (± 5 years) in Match 1. AI/AN and NHW patients were not matched on age when the outcome was age at death because age at admission and age at death for those who died during their hospitalization were nearly perfectly correlated (Match 2).Matching on region of residence and time of hospitalization minimized potential differences in availability of SARS CoV-2 vaccinations or exposure to different COVID-19 strains for AI/AN and NHW (Match 1 and 2).Both samples were matched based on sex (Match 1 and 2) to maintain the nearly equal proportions of men and women.
Analyses were conducted in the matched samples to identify factors that may contribute to disparities in mortality.First, we examined whether there were differences between AI/AN and NHW inpatients for each of the patient characteristics to identify potential explanations for the inequities in mortality.Second, we examined the extent to which each of these patient characteristics predicted mortality individually and after controlling for all the other characteristics in adjusted models.Of particular interest was the extent to which the association between AI/AN versus NHW status and mortality were reduced after patient characteristics, such as smoking and medical comorbidities, access to treatment (as indicated by distance to treatment, insurance type, residence in a rural or urban region, and receipt of antiviral medication while hospitalized), and area social deprivation were considered.The impact of controlling for these characteristics was quantified by examining the percentage attenuation in the effect sizes obtained in unadjusted and adjusted models: 100 × (B unadjusted model − B adjusted model) / (B unadjusted model ) cf. 36,37 .Third, in the fully adjusted models, we examined whether any of the patient characteristics differentially predicted mortality in AI/AN and NHW inpatients.The purpose was to determine whether a potential risk factor exerted a greater or lesser effect on mortality among AI/AN compared to NHW individuals.Analyses were conducted in SPSS 38 .Multilevel generalized linear models with a binomial distribution and a logit link were fit.Multilevel analysis was used to account for the clustering of patients within the 21 health systems.
Bootstrapped confidence intervals around percentages and means and t-tests were estimated in SPSS.The Benjamini-Hochberg procedure 39 was applied to control the false discovery rate in the multivariable analyses and results of such correction are shown in relevant tables.www.nature.com/scientificreports/

Results
Differences in mortality and other outcomes between AI/AN and NHW inpatients AI/AN patients were more likely to die during their hospitalization than were NHW patients (Table 1 and Fig. 1).AI/AN patients died at a significantly younger age than NHW patients in the full and matched samples, which was not surprising given that AI/ANs were admitted at a younger age.AI/ANs were also more likely to require intubation, be admitted to the intensive care unit, and spend more days in the hospital than NHWs; Mann-Whitney tests indicated that AI/ANs and NHWs did not significantly differ in the number of days hospitalized in the full sample but did in the matched sample (Table 1).Mortality   www.nature.com/scientificreports/

Differences in patient characteristics between AI/AN and NHW inpatients
Differences in patient characteristics between AI/ANs and the full and matched NHW samples are presented in Table 2.As mentioned, AI/AN patients were significantly younger than the full NHW sample but, as expected, were not younger than the matched NHW sample.Some of the significant differences observed in the full sample (smoking status, obesity, heart disease, drug use disorder, insurance type, and vaccination status) were no longer significant in the matched samples, and one new difference (chronic renal failure) was revealed by the matching process.
In the matched samples, AI/AN patients who were hospitalized with COVID-19 were significantly more likely than NHW patients to be diagnosed with the comorbid conditions of Type 2 diabetes, chronic renal failure, liver disease, and alcohol use disorder.They also had significantly higher comorbidity burden as indicated by a count of the number of comorbid disorders (median 2.0) than NHW patients (median 1.0); Mann-Whitney test (U = 842,344.00,z = 6.47, p < 0.001).
Although the AI/AN patients were predominantly urban dwelling (86.4%), they were more likely than NHW patients to live in a rural area and to live 60 or more miles from the facility in which they received treatment.AI/AN patients were significantly more likely than NHW patients to live in socioeconomically deprived areas.For example, they were more likely to live in areas with a higher percentage of residents living below the federal poverty level, with less than a high school education, who did not own a car, who lived in a crowded housing unit and in households headed by a single parent (see Supplemental Table S1).In summary, these results suggest that comorbid medical conditions, access to treatment, and area social deprivation are potential contributors to the inequity in mortality between AI/AN and NHW individuals with COVID-19.AI/AN and NHW patients did not significantly differ in smoking status, some medical comorbidities (obesity, heart disease, cancer, and drug use disorder), type of insurance coverage, receipt of antiviral medication, year admitted to the hospital, and vaccination status.These results suggest that these are unlikely to be contributors to the inequity in mortality between AI/AN and NHW individuals with COVID-19.

Associations of AI/AN versus NHW status and patient characteristics with mortality
As presented in Table 3, several comorbidities were associated with in-hospital mortality (after covariate adjustment and control for false discovery rate): chronic renal failure, liver disease, and heart disease.Several indicators of health care access were associated with in-hospital mortality in unadjusted analyses: being a Medicare recipient, receipt of antiviral medication, living a greater distance from the treatment facility, and living in a rural area; being in the top quintile of area social deprivation was also associated with in-hospital mortality, but after covariate adjustment and control for false discovery rate, being a Medicare recipient and receipt of antiviral medication were the only ones that remained significant predictors of mortality.
The unadjusted odds ratio of the association between AI/AN versus NHW status and mortality was 1.98 (95% CI 1.48, 2.65).After adjusting for comorbidities, access to treatment, area social deprivation and vaccination status, the odds ratio was reduced to 1.59 (95% CI 1.15, 2.22), which represents a 32% reduction in the association between AI/AN versus NHW status and mortality.
Analyses were conducted to identify the source of this diminution in the association between AI/AN versus NHW status and mortality.In a model that adjusted for only comorbidities (excluding those that were inversely associated with mortality in the adjusted model), the odds ratio was reduced to 1.82 (95% CI 1.35, 2.46), which represents a 12% reduction in the association between AI/AN versus NHW status and mortality.This effect was further probed by individually examining each of the six comorbid disorders.The single disorder that accounted for most of the mortality disparity was any liver disease (odds ratio reduced to 1.87 [95% CI 1.40, 2.51], representing a 9% reduction; the others ranged from 1.92 to 1.99).When specific liver diseases were included in the adjusted model presented in Table 3, the single best predictor of mortality with COVID-19 was hepatic failure, not elsewhere classified (OR 8.05, 95% CI 4. 22, 15.36).Given that liver disease is often secondary to hepatitis B and C and HIV infections 40 , we conducted post hoc analyses to determine that the relation between liver disease and mortality could not be explained by comorbid hepatitis or HIV (see Supplemental Text S4).
In a model that adjusted for only access to treatment (distance to treatment, insurance type, residence in a rural or urban region, and receipt of antiviral medication while hospitalized), the odds ratio was reduced to 1.82 (95% CI 1.34, 2.47), also representing a 12% reduction.In a model that adjusted for only area social deprivation, the odds ratio was reduced to 1.79 (95% CI 1.32, 2.43), representing a 15% reduction.These results suggest that comorbidities (especially comorbid liver disease), access to treatment and area social deprivation all contribute to the disparity in COVID-19 related mortality between AI/AN and NHW individuals.The significant residual relation between AI/AN versus NHW status and mortality after accounting for these candidate explanatory variables indicate that they do not fully account for this disparity.

Moderation of associations between AI/AN versus NHW status and mortality
We examined whether any of the patient characteristics differentially predicted mortality in AI/AN and NHW inpatients in fully adjusted models by including an interaction term between AIAN versus NHW status and each of the patient characteristics.After control for false discovery rate, there was no evidence that any of the potential risk factors exerted a greater or lesser effect on mortality among AI/AN compared to NHW individuals (see Supplemental Table S3).Before control for false discovery rate, however, there was a significant interaction between AIAN versus NHW status and alcohol use disorder (see Supplemental Text S5).The magnitude of the association between AI/AN versus NHW status and mortality did not change from 2020 to 2022.

Discussion
This study compared the outcomes of 546 AI/AN to 78,128 NHW individuals from 21 US health systems who were hospitalized with COVID-19 from February 1, 2020 to January 31, 2022.To better isolate differences, the AI/AN sample was also compared to a sample of 2645 NHW individuals matched on age, sex, region, and month/year of hospital admission.As expected, based on previous epidemics and recent studies [4][5][6][7] , the odds of in-hospital mortality were doubled in the AI/AN compared to the matched sample of NHW individuals with COVID-19.Novel to this study was the direct examination of the extent to which comorbidities, area social deprivation, and access to treatment could contribute to the AI/AN disparity in mortality.

Comorbidity and mortality differences between AI/AN and NHW
AI/AN inpatients were compared to NHW inpatients on eight diseases known to be more common among AI/ AN 12,25 and that have been linked to severe COVID-19 outcomes 13,14 .In this inpatient sample, AI/AN patients were more likely to have been diagnosed with three of these (diabetes, chronic renal failure, and liver disease); liver disease accounted for the largest portion of the comorbidity-related mortality disparity between AI/AN and NHW patients.

Area social deprivation and mortality differences between AI/AN and NHW
AI/AN inpatients were more than twice as likely to reside in the most socially deprived areas relative to NHW inpatients.Consistent with prior research 19,41 higher area social deprivation increased the risk of COVID-19 inpatients' dying in the hospital.Furthermore, adjusting for area social deprivation attenuated the association between AI/AN versus NHW status and hospital mortality.The present findings extend the limited number of studies that have linked area social deprivation with patient-level COVID-19 clinical outcomes.

Access to treatment and mortality differences between AI/AN and NHW
Higher rates of intubation, ICU admission, and days hospitalized suggest that AI/AN inpatients may have presented to inpatient care later in the COVID-19 disease course compared to NHW inpatients.In this study, 23% of AI/AN inpatients lived 60 or more miles from their treatment facility, compared to only 10% of the matched sample of NHW.It is likely that this greater distance to treatment may account for the greater severity of COVID-19 illness and ultimately higher rates of death in AI/AN than NHW patients.The association between AI/AN versus NHW status and mortality was reduced after adjusting for distance to treatment.More research is needed to identify other obstacles to timely treatment that AI/AN individuals are more likely to encounter than their NHW counterparts 42 .In particular, transportation barriers 43 and lack of internet access 17,18 may also impede timely treatment.

Differences between AI/AN and NHW in age at death
Previous studies have documented younger ages at death for AI/AN than NHW individuals with COVID-19 9,44 .
In the full sample of the present study, AI/AN patients were younger (d = 0.74) and had younger in-hospital ages at death (d = 0.74) than NHW patients.Of those who died during their hospitalization, 60% of the AI/ ANs compared to only 34% of the NHWs were less than 70 years old.The contributors to mortality disparities previously discussed likely contributed to disparities in age at death.AI/ANs were more likely than NHWs to be living in crowded housing, which may have promoted greater viral load.Living further from treatment may have led to greater delays in getting or seeking treatment and more advanced illness at hospitalization for AI/ANs compared to NHWs.
Disparities in age at death might also be explained by the concept of "weathering" that posits that the cumulative impact of repeated experience with social or economic adversity and political marginalization may lead to physiological deterioration 45 .This cumulative wear and tear has been termed "allostatic load" 45 .An empirical demonstration of weathering was conducted in a community-based sample of Black and White individuals showing that the allostatic load of a 40-year-old Black person was equal to that of a 50-year-old White person 46 .To our knowledge, studies of the causes of disparities in age at death have not been conducted among AI/AN individuals but should be a top priority for future research.The profound loss of lifespan among AI/AN has widespread effects given the value that Indigenous communities place on their elders 28,47 .

Other studies quantitatively explaining COVID-19 related mortality disparities
Several studies have systematically documented COVID-19 related mortality disparities among minority populations (e.g., 48,49 ).To our knowledge, only two studies, both conducted in the UK, have attempted to quantitatively explain the potential causes of these mortality disparities 36,50 .In one UK study 36 , comorbidities explained 10% of the association between Black/White status and mortality and 39% of association between South Asian/White status and mortality; social factors (educational attainment, occupational attainment, household size and area deprivation) explained 28% of the association between Black/White status and mortality and 4% of association between South Asian/White status and mortality.In the other UK study 50 , comorbidities and social factors (household size and area deprivation), but not lifestyle factors (smoking, body mass index), explained about 40% of the association between Black/White status and mortality and between South Asian/White status and mortality.(Access to treatment was not included in either of the UK studies.)In the present study, comorbidities explained 12% of the association between AI/AN status and mortality and area social deprivation explained 15% of the association between AI/AN status and mortality.Taken together, results from the UK studies and the present study suggest that comorbidities and area social deprivation account for similarly small fractions of the

Limitations
First, the sample included only hospitalized AI/AN and NHW patients during their first hospitalization for COVID-19, so it does not reflect the course of COVID-19 and mortality differences in the broader AI/AN population.Second, outcomes occurring post-discharge or outcomes occurring at nonparticipating health systems were not captured.Third, results across time could not be linked with type of COVID-19 variant.The analyses were conducted over the first 2 years of the pandemic, suggesting that the data obtained were contemporaneous with high prevalence of alpha, delta, and early omicron variants 51 .Fourth, risk factors for COVID-19 related mortality were considered in isolation when they were more likely to act in concert or be stages in a causal chain.For example, area social deprivation and distance to treatment may be barriers to adequate prevention and intervention for comorbid liver disease 52 .Future research should model the process by which risk factors combine to influence COVID-19 related mortality.
Fifth, the CEC-UW inpatient cohort is not a representative sample.There was selection of the participants based on having a diagnosis of COVID-19 and for being hospitalized.It is well known that studies based on EHR data are plagued by collider bias [53][54][55] .In this case, COVID-19 may be a collider associated with both AI/AN versus NHW status and mortality and could have induced distorted or spurious findings 56 .It is reassuring that the association between AI/AN versus NHW status and mortality observed in this study was similar to results obtained from other sources, such as state-level surveillance systems 5,7,8 .However, the positive association of receipt of antiviral medication with mortality may have been due to collider bias (but also possibly due to sicker patients being more likely to be prescribed medication).Although the association with mortality may have been distorted, it was an important observation that AI/AN were not less likely than NHW to receive pharmacologic treatment for their COVID-19.

Conclusions
Notwithstanding limitations, this study is an important contribution to the literature because it represents the first attempt to explain COVID-19 related mortality disparities among AI/AN.Comorbidities, area social deprivation, and access to treatment were all important contributors to the mortality disparity between AI/AN and NHW inpatients with COVID-19.Nonetheless, the significant residual relation between AI/AN versus NHW status and mortality after accounting for the candidate explanatory variables of comorbidity burden, neighborhood socioeconomic deprivation and reduced access to health care indicate that there are other important unmeasured factors that contribute to this inequity.This likely includes living conditions, such as multigenerational and crowded housing 57 , being a frontline worker 57 , and having inadequate access to transportation 43 and to the internet 17,18 .Accounting for the unexplained causes of disparities among AI/AN will be an important direction for future research.
Health disparities among AI/AN are not a new problem but reflect "legacies of failing to address historical and ongoing inequities" ( 58 , p. 2739).Results of the present study likely extend beyond the current COVID-19 pandemic and may apply to many other past, current, and future health disparities experienced by AI/AN.The availability of quality data on the disparate impacts of health threats such as COVID-19 on AI/ANs is essential in the effort to reduce disparities and enhance health equity.
13:20974 | https://doi.org/10.1038/s41598-023-48260-9 Native and Non-Hispanic White adult inpatients from the CEC-UW COVID-19 cohort on in-hospital mortality and other outcomes.ICU intensive care unit, df degrees of freedom, SD standard deviation, OR odds ratio, mdn median.a Match 1 sample used for matched analyses.b Match 2 sample used for matched analyses.

Figure 1 .
Figure 1.In-hospital mortality, intubation, and ICU admission among matched samples of American Indian/ Alaska Native and Non-Hispanic White inpatients from the CEC-UW COVID-19 cohort.Vertical bars are bootstrapped 95% confidence intervals, analyses are based on the Match 1 sample.

Table 3 .
Associations of comorbid disorders, treatment access, and area deprivation with mortality among matched samples a of American Indian/Alaska Native and Non-Hispanic White adult inpatients from the CEC-UW COVID-19 cohort.OR odds ratio, CI confidence interval, B-H? did the p value survive Benjamini & Hochberg (1995) test for multiple comparisons.Italic rows indicate variables on which groups were matched.a Match 1 sample used for all analyses.b Adjusted analyses estimate the effect of each characteristic after adjusting for the influence of every other characteristic.