Association of Vitamin D Status and COVID-19-Related Hospitalization and Mortality

Background The relationship between vitamin D status and COVID-19-related clinical outcomes is controversial. Prior studies have been conducted in smaller, single-site, or homogeneous populations limiting adjustments for social determinants of health (race/ethnicity and poverty) common to both vitamin D deficiency and COVID-19 outcomes. Objective To evaluate the dose-response relationship between continuous 25(OH)D and risk for COVID-19-related hospitalization and mortality after adjusting for covariates associated with both vitamin D deficiency and COVID-19 outcomes. Design Retrospective cohort study. Patients Veteran patients receiving care in US Department of Veteran Affairs (VA) health care facilities with a positive severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) test and a blood 25(OH)D test between February 20, 2020, and November 8, 2020, followed for up to 60 days. Main Measures Exposure was blood 25(OH)D concentration ascertained closest to and within 15 to 90 days preceding an index positive SARS-CoV-2 test. Co-primary study outcomes were COVID-19-related inpatient hospitalization requiring airborne, droplet, contact, or other isolation and mortality ascertained within 60 days of an index positive SARS-CoV-2 test. Key Results Of 4,599 veterans with a positive SARS-CoV-2 test, vitamin D deficiency (< 20 ng/mL) was identified in 665 (14.5%); 964 (21.0%) were hospitalized; and 340 (7.4%) died. After adjusting for all covariates, including race/ethnicity and poverty, there was a significant independent inverse dose-response relationship between increasing continuous 25(OH)D concentrations (from 15 to 60 ng/mL) and decreasing probability of COVID-19-related hospitalization (from 24.1 to 18.7%, p=0.009) and mortality (from 10.4 to 5.7%, p=0.001). In modeling 25(OH)D as a log-transformed continuous variable, the greatest risk for hospitalization and death was observed at lower 25(OH)D concentrations. Conclusions Continuous blood 25(OH)D concentrations are independently associated with COVID-19-related hospitalization and mortality in an inverse dose-response relationship in this large racially and ethnically diverse cohort of VA patients. Randomized controlled trials are needed to evaluate the impact of vitamin D supplementation on COVID-19-related outcomes. Supplementary Information The online version contains supplementary material available at 10.1007/s11606-021-07170-0.

pandemic has resulted in unprecedented suffering, morbidity, and mortality worldwide. 1,2 The COVID-19 vaccine has promised to decrease COVID- 19 prevalence, yet persistent vaccine hesitancy and barriers to vaccine access in racial and ethnic minority and underserved populations coupled with emerging COVID-19 variants are leading to new surges of infection. [3][4][5][6] Thus, interventions to mitigate COVID-19 disease severity remain highly relevant due to disparities in individuals' ability to prevent or access effective therapies for COVID-19 and because of critical shortages of hospital beds. [7][8][9] Vitamin D deficiency, typically defined as 25(OH)D < 20ng/mL, 10,11 is widespread and considered a global public health problem. 12,13 Vitamin D deficiency or low 25(OH)D is more prevalent in non-White individuals, those > 65 years and/or obese, and those residing in Northern latitudes with less sunlight. [14][15][16][17] Several studies have demonstrated an independent association between vitamin D deficiency and testing positive for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). [18][19][20] Other observational studies have reported poorer clinical outcomes from SARS-CoV-2 infection in patients with vitamin D deficiency, but these studies have been relatively small or single site. [21][22][23] To date, randomized clinical trials of vitamin D supplementation on COVID-19-related outcomes have demonstrated mixed results. [24][25][26] Moreover, vitamin D deficiency has been associated with a variety of chronic health conditions (e.g., diabetes, cardiovascular disease), yet randomized controlled trials have failed to demonstrate that vitamin D supplementation prevents or ameliorates these chronic conditions. [27][28][29] This suggests that vitamin D may instead function as a marker for general health, nutritional status, and outdoor physical activity. 30 A challenge in studying the relationship between vitamin D deficiency and COVID-19 outcomes is that risk factors for vitamin D deficiency are also associated with COVID-19 disease severity (e.g., obesity, medical comorbidities) as well as social determinants of health (e.g., non-Whiterace/ethnicity and poverty). 9,12,15,31 Some prior COVID-19-related observational studies and trials have failed to fully adjust for potential confounding, particularly race/ethnicity, owing to the lack of diversity in the populations studied. 19,24,25 Because of the controversy surrounding vitamin D supplementation, in July 2020, the US National Institutes of Health concluded "there are insufficient data to recommend either for or against the use of vitamin D for the prevention or treatment of COVID-19." 32 Using a Department of Veterans Affairs (VA) clinical database of veteran patients across the USA with positive SARS-CoV-2 tests linked to recent blood 25(OH)D test results, we investigated independent dose-response relationships between blood 25(OH)D concentrations and risk for COVID-19-related hospitalization and mortality. This study adds to the literature in including a large sample of geographically and racially and ethnically diverse patients, allowing adjustment for social determinants of health common to both vitamin D deficiency and COVID-19.

Study Population
This is a retrospective cohort analysis of veterans enrolled in VA health care systems across the USA tested for SARS-CoV-2 from February 20, 2020, to November 8, 2020, with up to 60 days of follow-up after the first or "index" SARS-CoV-2 test until study end (December 8, 2020) ( Figure 1). Of 681,183 patients, 71,175 had positive SARS-CoV-2 tests, and of these, 4,872 had 25(OH)D tests within 15 to 90 days of their SARS-CoV-2 tests. Patients with 25(OH)D tests within 14 days of a positive SARS-CoV-2 test were excluded because vitamin D may act as a negative acute phase reactant, leading to reverse causality. 33 Of these 4,872 patients, an additional 273 were excluded who were missing covariate data or were inpatients admitted more than 3 days prior to the SARS-CoV-2 test. This resulted in a final analytic cohort of 4,599 patients. This study was approved by the Committee on Human Research, University of California, San Francisco, and the San Francisco VA Health Care System Human Research Protection Program.

Data Sources
The primary data source was the VA COVID-19 Shared Data Resource that included SARS-CoV-2 PCR test results, medical comorbidities associated with SARS-CoV-2 infection risk and disease severity (e.g., diabetes, chronic obstructive pulmonary disease (COPD), cardiovascular disease (CVD), smoking), health services utilization (including COVID-19related hospitalization), and mortality data. The VA Corporate Data Warehouse (CDW) was linked to the COVID-19 Shared Data Resource to provide additional sociodemographic data and laboratory data, in this case, blood 25(OH)D and serum albumin test results. Data were prepared using SAS 9.4 (SAS Institute, Cary, NC), 34 36,37 In binary analyses, vitamin D concentrations <20ng/mL represented vitamin D deficiency, consistent with clinical practice guidelines. 38 Based on the distribution of 25(OH)D concentrations within the entire VA COVID database (N=237,934), 25(OH)D concentrations used in this study falling below the 5th percentile (i.e., 13.6ng/mL) and above the 95th percentile (i.e., 60.2ng/mL) were reset to 13.6 and 60.2 ng/mL respectively because small numbers of patients at either extreme could exert undue influence in a logtransformed model (see below). Covariates in multivariable analyses included sociodemographics, such as age at index SARS-CoV-2 test date with relative risk calculated per 5-year increase; race/ethnicity; and poverty with relative risk calculated per 5% increase in the proportion of residents in patients' zip code areas living below the federal poverty line. 39 We included race/ethnicity and a proxy for poverty given these are two social determinants of health known to be associated with both SARS-CoV-2 infection and COVID-19 severity as well as vitamin D deficiency. 12,40 Additional comorbidities associated with both COVID-19 severity and vitamin D deficiency were included, namely obesity (body mass index (BMI) as a continuous variable in kilograms/meters 2 (kg/m 2 )), smoking, COPD, obstructive sleep apnea, CVD including hypertension, cancer, diabetes, chronic kidney disease (CKD), liver disease, human immunodeficiency virus (HIV), drug dependence, and alcohol dependence. 15,31 The International Classification of Diseases, Tenth Edition (ICD-10) codes denoting each condition associated with VA outpatient and inpatient encounters that occurred within 2 years preceding the index SARS-CoV-2 test were included in these analyses. 41

Statistical Analyses
Descriptive univariate analyses were used to calculate the probabilities of individual associations between vitamin D deficiency (25(OH)D concentrations < 20 ng/mL) and sociodemographic factors and medical comorbidities, summarized as unadjusted pairwise relative risks (RR). Relative risks with p-values < 0.05 were considered significant. Both unadjusted and adjusted log-linked generalized linear models with Poisson distributions and cluster-robust error variances were used to estimate outcome probabilities and relative risks to describe the dose-response relationship between continuous 25(OH)D concentrations and COVID-19-related hospitalization and mortality. 42 Since the relationship between 25(OH)D concentrations and the risk of outcomes may be non-linear, multiple functional forms of the model were investigated (y=x, y=x+ ln(x), y=x+x 2 , y=x+x 1/2 , y=x+x 1/3 , and y=x+x 1/4 ). Given the potential for type 1 error inflation, resampling with replacement with 500 resamples was implemented for all functional forms of the model to reduce the impact of outliers and increase reproducibility of the results. The model with the consistently lowest average BIC across resamples was chosen as the final model functional form (y=ln(x)). Interactions between blood 25(OH)D and serum albumin (accounts for a small proportion of 25(OH)D binding) 43 were tested and were not significant in any models. Sensitivity analyses that included serum albumin as a covariate did not significantly alter study findings (Appendix Tables 5 and 6).
Three generalized linear models were fit for each of the coprimary study outcomes (hospitalization and mortality): an unadjusted model with only log 25(OH)D concentration, a model adjusted for age and sex, and a model adjusted for all independent covariates. These continuous models were presented in terms of adjusted probabilities of the outcome-CO-VID-19-related hospitalization or mortality-conditional upon different blood 25(OH)D concentrations of interest. This involved using a post-estimation technique of predictive margins to calculate conditional probabilities and confidence intervals. 44 In addition, relative risks for comparisons between outcome probabilities for any pair of blood 25(OH)D concentrations (e.g., 15 ng/mL vs. 20, 25, 30, 40, 50, or 60 ng/mL) could be made using margin contrasts, calculated as the ratio of the log of the margins followed by exponentiation. 44

RESULTS
Of 4,599 eligible patients with a positive index SARS-CoV-2 test and 25(OH)D measurement within 15 to 90 days, 588 (12.8%) were female, mean age was 62.6 years (SD +/−15.1 years); 1,578 (34.3%) identified as non-White, and 487 (10.6%) as Latinx ( Table 1). The mean follow-up time was 54.9 days, with a range of 31 to 60 days. The median 25(OH)D concentration was 32.6 ng/mL (interquartile range=24.3-42.6 ng/mL) and 665 (14.5%) had 25(OH)D concentrations < 20 ng/mL, indicating vitamin D deficiency. In unadjusted analyses, Black or African American patients were at the very highest risk for low 25 (OH) D concentrations (RR=2.63; 95% confidence interval (CI)=2.28-3.04, p<0.001), as well as patients residing in areas with higher concentrations of poverty and those with alcohol and drug use dependence (Table 1).
The inverse dose-response relationship between continuous increasing 25(OH)D concentrations (from 15 to 60 ng/mL) and corresponding decreasing probability of COVID-19related hospitalization (from 24.1 to 18.7%) in fully adjusted analyses is shown in Figure 2 and Appendix Table 2. Figure 2 is non-linear demonstrating that among patients with lower 25(OH)D concentrations, increases in 25(OH)D are associated with larger reductions in probability for hospitalization than among those with higher 25(OH)D concentrations. Table 3 provides examples of representative concentrations of 25(OH)D compared to 15 ng/mL. For example, after adjusting for all other covariates, patients with a positive SARS-CoV-2 test and a 25(OH)D concentration of 15 ng/mL compared to 40 ng/mL had a 20% greater risk of hospitalization (ARR=1.20, 95% CI=1.05-1.37, p=0.009) ( Table 3).
Of the 4,599 patients with positive SARS-CoV-2 tests, 340 (7.4%) died within 60 days of their index SARS-CoV-2 test. Of those who died, 48 (14.1%) had vitamin D concentrations <20 ng/mL. Increasing 25(OH)D concentrations were independently associated with decreasing risk of COVID-19related mortality after adjusting for all covariates (p=.001) ( Table 4). Other risk factors significantly associated with increased risk for COVID-19-related mortality were age (44% per 5-year increase), diabetes (42%), CVD including hypertension (30%), and chronic kidney disease (64% increase) ( Table 4). The inverse dose-response relationship between continuous increasing 25(OH)D concentrations (from 15 to 60 ng/mL) and decreasing probability of COVID-19-related mortality within 60 days (from 10.4 to 5.7%, p=0.001) is illustrated in Figure 3 and Appendix Table 3. Figure 3 is non-linear demonstrating that among patients with the lower 25(OH)D concentrations, increases in 25(OH)D are associated with larger reductions in mortality than among those with    Table 5).

DISCUSSION
To our knowledge, this is the first observational study to demonstrate a dose-response relationship between the exposurelower concentrations of 25(OH)Dand the outcome increased COVID-19-related hospitalization and mortality in a large geographically and racially and ethnically diverse cohort of VA patients testing positive for SARS-CoV-2. Furthermore, this relationship was non-linear. Patients with the lowest 25(OH)D concentrations had the greatest associated These adjusted relative risks and confidence intervals reflect comparisons between pairs of 25(OH)D values displayed in Figure 2 (fully adjusted model shown only) which were predicted from the models in Table 2  Relative risk reflects 5% increase in proportion of residents in patient ZIP code living below federal poverty line.  Table 4 which was adjusted for age, sex, sociodemographics (race, ethnicity, proportion of residents below federal poverty line), medical comorbidities (obesity, diabetes, cardiovascular disease including hypertension, obstructive sleep apnea, obstructive sleep apnea, chronic obstructive pulmonary disease, cancer, chronic kidney disease, liver disease, human immunodeficiency virus), and health risk behaviors (smoking status, non-alcohol drug dependence, alcohol dependence) reduction in hospitalization and mortality as 25(OH)D concentrations increased. These findings remained significant even after adjusting for known risk factors for both vitamin D deficiency and COVID-19 severity. These findings contribute to a growing evidence-base suggesting that vitamin D deficiency may be associated with more severe outcomes related to COVID-19. 19,21-25 Indeed, pre-clinical studies demonstrate that 25(OH)D stimulates immune and respiratory epithelial cells to secrete cathelicidin, an anti-microbial peptide that clears respiratory pathogens. 25(OH)D also initiates adaptive immunity to dampen down pro-inflammatory cytokines ("cytokine storm") leading to adverse COVID-19 outcomes. 45,46 This study found that Black or African American patients had over twice the risk for low 25(OH)D concentrations consistent with 25(OH)D deficiency. In Black and other non-White individuals, darker skin pigmentation blocks ultraviolet light; thus, more solar radiation may be required to produce similar concentrations of 25(OH)D, although the etiology of low 25(OH)D in non-White individuals remains controversial. 12,15,47,48 In addition, stay-at-home orders or quarantining related to COVID-19 may have resulted in even less sun exposure, putting some individuals at greater risk for vitamin D deficiency. 49 Furthermore, ad hoc analyses of the larger cohort of 71,175 with positive SARS-CoV-2 tests (Figure 1) revealed that patients who were Caucasian, older, female, and had a variety of chronic health conditions including CVD, COPD, and cancer were significantly more likely to have been tested for 25(OH)D in the 15-90 days prior to their index positive SARS-CoV-2 test (Appendix Table 4). This may explain why some patients in this cohort with chronic conditions were less likely to be vitamin D deficientthey had been tested and likely supplemented with vitamin D (a finding confirmed by additional ad hoc analyses). In contrast, Black or African American patients were significantly less likely than Caucasian patients to have been tested for 25(OH)D within the VA in the past year (Appendix Table 4), hence less likely that vitamin D deficiency would have been detected or treated. This is consistent with studies revealing disparities in access to care within VA among ethnic/racial minority groups, which may also explain the lower 25(OH)D concentrations observed in non-White populations. 50,51 Black or African American populations have been disproportionately impacted by the COVID-19 pandemic through higher infection rates and more serious clinical outcomes, thought to be related to increased exposure, poorer access to health care, and to a lesser extent, underlying pre-existing conditions. 9,52 Latinx patients and patients residing in areas with higher levels of poverty were also at significantly greater risk for vitamin D deficiency. In addition, patients with drug and alcohol dependence were more likely to have low 25(OH)D concentrations, and those with alcohol dependence to be hospitalized for COVID-19-related complications. Individuals with substance use disorders, as well as those living in poverty, may have lower intake of vitamin D enriched foods or supplements, coupled with a lack of or increased requirement for sunlight to produce 25(OH)D. Thus, testing for 25(OH)D and supplementing when indicated, especially in non-Whiteracial/ethnic groups, poor individuals, and those with substance use disorders, may represent one way to mitigate social disparities associated with poorer COVID-19-related clinical outcomes. 40,52 Vitamin D deficiency was strongly associated with mortality related to COVID-19 infection. Steadily decreasing concentrations of vitamin D from 60 to 15 ng/mL corresponded to a continuous and significant increase in mortality in patients hospitalized for COVID-19 infection after adjustment for sociodemographics, including race/ethnicity and medical comorbidities. Only a handful of other smaller studies have demonstrated this association between vitamin D deficiency and COVID-19-related deaths; [22][23][24] thus, more research is needed to rule out the possibility that 25(OH)D is instead serving as a marker for other unmeasured predictors of mortality in patients with COVID-19. Nevertheless, the non-linear dose-response relationship between 25(OH)D concentrations and risk of both COVID-19-related hospitalization and mortality provides tentative support for providing vitamin D supplementation.
This analysis has limitations. First, results are based on VA administrative data that incompletely capture all possible risk factors for COVID-19 hospitalization and mortality, in addition to missing, unknown, or misclassified race/ethnicity data and a lack of income and educational data. 53,54 Second, the retrospective cohort of US veterans was not representative of all veterans or the US population; it was also largely male, and thus our results may not completely generalize. Third, results may be biased as to which VA patients are tested for 25(OH)D; as ad hoc analyses revealed, those with 25(OH)D These adjusted relative risks and confidence intervals reflect comparisons between pairs of 25(OH)D values displayed in Figure 3 (fully adjusted model shown only) which were predicted from the models in Table 4 15 to 90 days of the positive test did not act as a negative acute phase reactant, even outside the prodromal or infectious period. 33 Finally, we were unable to reliably investigate the role of vitamin D supplementation since VA pharmacy data only capture vitamin D dispensed through VA pharmacies. Considering these limitations, our findings suggest that lower 25(OH)D concentrations have a significant and independent dose-response relationship with adverse clinical outcomes of COVID-19 infection, namely hospitalization and mortality, as it did in this large, diverse cohort of VA patients tested both for COVID-19 and 25(OH)D. Compared to other COVID-19 therapies, vitamin D supplementation is relatively inexpensive, widely available, and safe for most individuals at therapeutic doses. Therefore, consideration should be given to testing patients for 25(OH)D who are SARS-CoV-2 positive or at high risk for COVID-19-related complications and providing vitamin D supplementation when deficient. Only a large randomized controlled trial can confirm if vitamin D supplementation can prevent hospitalization and mortality in patients with COVID-19. Another consideration brought to light by this study is that testing for and supplementing those with low 25(OH) D concentrations may differentially benefit vulnerable sub-populations of patients disproportionately negatively impacted by COVID-19. Funding This work was financially supported by the US National Institutes of Health (NIH) through cooperative agreement U24AT009769 and cooperative agreement UG3AT009765/ UH3AT009765 from the National Center for Complementary and Integrative Health (NCCIH at the NIH) (Dr. Seal). This work was also supported by an administrative supplement from the Office of Dietary Supplements at the National Institutes of Health (NIH) (Dr. Seal). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Department of Veterans Affairs, the NIH, or any other institution. The funding source played no role in the design, conduct, and analysis of this study or in the decision to submit the manuscript for publication.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons. org/licenses/by/4.0/.