The worldwide pandemic of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has presented the largest global public health problem in several generations. A vast amount of rapid research has taken place to find an effective therapeutic agent to manage 2019 novel coronavirus disease (COVID-19), which is caused by SARS-CoV-2. Vitamin D (which will henceforth be referred to as cholecalciferol, i.e., in its “parent”, therapeutic form, vitamin D3
) has been proposed as a potential adjuvant to therapy for COVID-19 in a number of recent studies [1
], as cholecalciferol has previously been suggested to have an antiviral effect [5
]. Serum 25-hydroxyvitamin D (25(OH)D) is the measurable metabolite of cholecalciferol that is used to determine an individual’s vitamin D status. Other reviews have suggested that replete vitamin D status (serum 25(OH)D) >50 nmol/L) may be important in preventing severe manifestations of COVID-19 [6
]. Other sources have proposed the interleukin (IL)-6 inhibitor tocilizumab as a potential treatment for COVID-19, and due to its modulator effect on IL-6, cholecalciferol has again been postulated as a potential therapeutic option [8
A UK-based study by Panagiotou et al. found that low serum 25(OH)D levels in COVID-19 in-patients were associated with a more severe disease course [9
], but this small study of 134 patients only looked at serum levels and not concurrent cholecalciferol therapy. Furthermore, two meta-analyses have identified low serum 25(OH)D levels as a potential predictor of more severe COVID-19 disease outcomes [10
], again without addressing any effect of treatment. To our knowledge, no observational study has addressed the effect of cholecalciferol therapy on outcomes on an individual-level basis, and very few studies exist regarding serum 25(OH)D levels and the risk of COVID-19 mortality following in-patient admission.
Therefore, the primary research question of this study was to determine whether serum 25(OH)D levels and/or deficient vitamin D status affect mortality in COVID-19 infection. Our secondary objective was to determine whether any other patient characteristics were associated with COVID-19 mortality. To reach these objectives, we carried out a retrospective multi-centre cross-sectional observational study.
2. Materials and Methods
For the primary analysis, patients were opportunistically recruited from an acute hospital trust in the UK, namely Tameside and Glossop NHS Foundation Trust (recruitment from Tameside General Hospital). Tameside General Hospital is a district general hospital serving a population of 250,000 people living in both urban and rural areas in Greater Manchester and Derbyshire. For the validation analysis, patients were recruited from two additional acute hospital Trusts, also in the UK: Lancashire Teaching Hospitals NHS Foundation Trust (recruitment from Royal Preston Hospital), and University Hospitals of Leicester (UHL) NHS Trust (comprising of Glenfield Hospital, Leicester General Hospital, and Leicester Royal Infirmary). Patients were recruited from all three sites comprising UHL, and samples were processed at the same laboratory, based at Leicester Royal Infirmary. Royal Preston Hospital serves a population of 370,000 people living in both urban and rural areas in Lancashire and South Cumbria. The UHL NHS Trust serves 1,000,000 patients across Leicestershire; Leicester Royal Infirmary has the county’s only emergency department.
Ethical approval was granted by the Health and Care Research Wales Research Ethics Committee (IRAS number 285337). The study was also registered on Clinicaltrials.gov (reference number NCT04386044), prior to commencement. Although this was not an interventional study, we registered it on an open-access database for transparency. Informed written consent was not required as this was rapid COVID-19 research carried out prior to March 2021, as per Health Research Authority (HRA) guidance in the UK [12
]. The UK Government announced emergency arrangements for the use of confidential patient information without consent for COVID-19 research purposes, and this was included in our ethical application, which was approved following fast tracking.
Patients recruited were admitted between 27th January 2020 and 5th August 2020, and data were collected retrospectively between 26th June 2020 and 7th August 2020. Although the first case of COVID-19 reported in the UK was on approximately 29th January 2020 and the first patient recruited to the study was admitted on 27th January 2020, this patient developed COVID-19 as an in-patient following admission for a different condition. Data collection commenced as soon as ethical approval had been obtained, and stopped once the investigators felt that the first peak of COVID-19 admissions had passed.
In-patients with a clinical diagnosis of COVID-19 identified by clinical coding (emergency use ICD code U07·1, COVID-19 confirmed by laboratory testing, and code U07·2, COVID-19 diagnosis where laboratory confirmation is inconclusive or not available [13
]) were all included in the study. Laboratory testing for COVID-19 was carried out using throat ± nasal swab, and samples were tested for SARS-CoV-2 viral RNA following amplification using real-time PCR. A clinical diagnosis of COVID-19 was made if laboratory testing was negative, but patients had symptoms and signs suggestive of SARS-CoV-2 infection, such as persistent dry cough, low oxygen saturations (SpO2
), fever, dyspnoea, bilateral interstitial infiltrates on a chest radiograph or computed tomography (CT) scan, etc. Patients were excluded if they were younger than 18 years of age or if the final clinical diagnosis was not COVID-19. Demographic and clinical data were obtained from the hospital admission associated with the diagnosis of COVID-19 using a combination of electronic patient records (EPR), hard-copy patient records, and hospital laboratory data.
The primary outcome measure, COVID-19 mortality, included deaths in hospital, and deaths following admission recorded during the data collection period, e.g., following transfer or discharge. Potential predictors of mortality consisted of: baseline serum 25(OH)D levels, deficient vitamin D status (serum 25(OH)D < 25 nmol/L), treatment with cholecalciferol using high-dose booster therapy (approximately ≥ 280,000 IU in a time period of up to 7 weeks [14
]), age, sex, non-Caucasian ethnicity, hospital-acquired COVID-19, clinical parameters on admission (SpO2
, C-reactive protein (CRP), creatinine, random glucose), progression to either continuous positive airways pressure (CPAP) therapy or invasive mechanical ventilation (IMV), length of stay, and common medical comorbidities of interest (including both type 1 and type 2 diabetes), as listed by clinicians in patient medical records. A full list of variables measured and how they were obtained is listed in Table S1
. Patients received cholecalciferol booster therapy if they were recognised as being either vitamin D insufficient (serum 25(OH)D 25–50 nmol/L) or deficient as part of routine clinical care. Hospital-acquired COVID-19 was defined as: (i) if a patient had been admitted with a different acute condition and had gone on to develop COVID-19 whilst an in-patient; or (ii) if a patient had been re-admitted within the 14-day incubation period and the second admission was for COVID-19. All laboratory measurements (including serum 25(OH)D levels) were carried out as part of routine clinical care of patients during their acute in-patient admissions.
2.2. Measurement of Serum 25(OH)D Levels
Serum 25(OH)D was measured using the UniCel Dxl 800 Access Immunoassay System (Beckman Coulter Life Sciences, Indianapolis, IN, USA) at Tameside General Hospital, the cobas e 801 analytical unit (Roche, Basel, Switzerland) at Royal Preston Hospital, and the ADVIA Centaur XPT Immunoassay System (Siemens Healthineers, Erlangen, Germany) at UHL. As serum 25(OH)D measurement was not part of an established care protocol at any participating site, it was at the discretion of physicians caring for patients whether to order this test. Hence, some participants still have missing values, as measurements were carried out as part of routine clinical care, and not specifically for participation in this study. Serum 25(OH)D measurements were ordered if patients were deemed to be at risk of insufficiency or deficiency (e.g., non-Caucasian ethnicity, elderly, lack of exposure to sunlight). In addition, as evidence emerged over the course of the COVID-19 pandemic regarding serum 25(OH)D and disease outcomes, patients with suspected or confirmed COVID-19 were also included on the list of patients at risk of insufficiency or deficiency. Serum 25(OH)D levels up to 12 weeks prior to admission with acute COVID-19 were included if not measured during each participant’s in-patient stay, in order to increase power. This time period was set to mitigate for seasonal variation; we chose not to include older serum 25(OH)D measurements for this reason.
The clinical laboratories at Tameside General Hospital participate in the Randox International Quality Assessment Scheme (RIQAS) [15
] in order to ensure external quality assessment of all assays, including 25(OH)D. The clinical laboratories at Royal Preston Hospital participate in the Vitamin D External Quality Assessment Scheme (DEQAS) [16
] to ensure analytical reliability of its 25(OH)D assays. The clinical laboratories at UHL participate in DEQAS, as well as the UK National External Quality Assurance Scheme (NEQAS) [17
] for external quality assessment of all assays.
2.3. Statistical Methods
The Wilcoxon rank–sum test was used to determine whether 25(OH)D assays were significantly different by centre, due to non-parametric distributions of this variable. In both the primary and validation cohorts, logistic regression was used to analyse predictor variables for potential associations with COVID-19 mortality, with adjustment for the following variables, which are known to be associated with COVID-19 mortality: age, sex, obesity, non-Caucasian ethnicity, and diabetes (types 1 and 2 combined). Median values were used to convert linear variables to binary high/low variables, as none of these variables had a parametric distribution. Binary variables were created separately for both primary and validation cohorts, as cut-off values were slightly different between the two independent populations. Variables with significant associations were placed into multivariate logistic models to adjust for any potential interactions between predictors and potential confounders. In the validation cohort, analysis was additionally adjusted for the centre from which participants were recruited. Missing values were treated as missing data, and values were not imputed, because of the nature of the clinical data collected. All analysis was carried out using Stata (StataCorp LLC, College Station, TX, USA), version 14.0.
To our knowledge, this is the largest observational study of hospital in-patients with COVID-19 to examine any potential associations between the treatment of the acute infection and vitamin D status, and cholecalciferol treatment. Serum 25(OH)D levels were not associated with COVID-19 mortality in both primary and validation cohorts, and deficient vitamin D status was not associated with COVID-19 mortality in the primary cohort. However, treatment with cholecalciferol appeared to be protective against mortality, regardless of baseline serum 25(OH)D levels, and this replicated across both cohorts.
Our findings regarding 25(OH)D levels appear to fit with a study utilising participants from the UK Biobank, which found no association between serum 25(OH)D levels and risk of COVID-19 infection [18
]. The UK Biobank study looked at 348,598 participants, of whom, 449 had a confirmed diagnosis of COVID-19 as defined by a positive laboratory test for SARS-CoV-2 (only 0.13% of study population). However, it is likely that the COVID-19 cases from that study were managed in a mixture of hospitals and the community, and serum 25(OH)D was measured between 2006 and 2010, and not contemporaneously with COVID-19 infection 10–14 years after recruitment to the UK Biobank. Our study adds extra information regarding patients who, by their nature, have more severe disease, as they have been hospitalised. Additionally, our study provides information on 25(OH)D levels as close to acute COVID-19 infection as possible (as opposed to up to 14 years before contracting COVID-19, as in the UK Biobank study), giving a more accurate picture of any interactions; we imposed a limit of 12 weeks on 25(OH)D levels prior to admission to mitigate for seasonal variation, whilst also including as many measured 25(OH)D levels as possible to maximise power.
Rhodes et al. suggest that countries at a latitude above 35 degrees North have experienced increased mortality from COVID-19, suggesting a potential role of cholecalciferol therapy in COVID-19 treatment [19
], but our findings do not implicate 25(OH)D levels in the role of increasing mortality rates in these countries. In this editorial, statistics of COVID-19 mortality by country are compared to latitude, and because countries above 35 degrees North are shown to have increased mortality per million population compared to those south of this latitude, the authors postulate that vitamin D insufficiency may be a contributing factor, due to a lack of sunlight during the winter months. This is not an experimental epidemiological study and does not include data on 25(OH)D levels or vitamin D status, in comparison to our study, which includes both.
Two independent studies from Israel [20
] and the USA [21
] found that deficient vitamin D status was associated with an increased risk of COVID-19. These studies differ from our own because they only studied the risk of SARS-CoV-2 infection, and not of mortality. Our study also differs in that it only includes hospitalised patients, who are already known to have developed COVID-19, so we were unable to assess whether vitamin D status was associated with the risk of SARS-CoV-2 infection, due to the nature of our study population. Our findings appear to differ from some of these other studies, but this could be due to power issues or a differing population. Given the emerging nature of this research, large meta-analyses will be required in the future when more data are available from multiple international sites.
Interestingly, treatment with high-dose cholecalciferol booster therapy was associated with a reduced odds of death, even following adjustment for baseline serum 25(OH)D levels [14
]. This could be due to a number of reasons. Firstly, it may be because it is not clear what an adequate amount of cholecalciferol supplementation is required to maintain immune health. UK guidance is that serum 25(OH)D levels > 25 nmol/L are required to maintain musculoskeletal health [22
], but this is within the range for deficiency [14
] and does not take into account cholecalciferol’s role outside of musculoskeletal health. It is possible that serum 25(OH)D levels might need to be higher than the recommended range in order to provide protection from more severe COVID-19 outcomes. An alternative hypothesis might be that not all patients had 25(OH)D levels measured during admission (755/986 participants), and the patients that had levels measured and acted on may have received overall more intensive treatment, resulting in better outcomes, i.e., a proportion of patients had 25(OH)D levels measured, but may not have been prescribed replacement cholecalciferol treatment, for unknown clinical reasons. From the level of data collected, it is not clear what the mechanisms are behind our findings.
It may seem paradoxical that whilst treatment with high-dose cholecalciferol reduces the risk of COVID-19 mortality, neither baseline serum 25(OH)D levels nor vitamin D deficient status had an effect on mortality risk in the primary cohort. Vitamin D deficient status was associated with increased mortality in the validation cohort following adjustment for potential confounders, so the lack of association in the primary cohort may have been due to superior power to detect the association in the validation cohort.
However, our findings may have been due to the concept of the “personal vitamin D response” put forward by Carlberg et al. [23
]. Two studies in independent populations (one in elderly pre-diabetic patients [24
], and another in young, healthy subjects [25
]) demonstrated changes in mRNA expression following administration of cholecalciferol. A range of clinical and biochemical parameters were also tested, but individual participants appeared to express these parameters differently from one another within the same studies. The authors classified participants as high/moderate/low responders to cholecalciferol based on the number of altered parameters that they exhibited following cholecalciferol administration, and they found that up to 25% of participants were low responders [23
Given that our study was only carried out on hospital in-patients (i.e., those with the most severe clinical manifestations of COVID-19), if Carlberg et al.’s hypothesis [23
] holds true, it is possible that in our study population there could be patients who are low responders to cholecalciferol who have seemingly adequate levels. Conversely, a high responder with seemingly deficient serum 25(OH)D levels could have only mild or no COVID-19 symptoms and not require hospitalisation, as they are able to utilise much lower 25(OH)D levels to better effect. Therefore, by analysing only patients who have been hospitalised, our population may include vitamin D high responders with low serum 25(OH)D levels, as well as vitamin D low responders with sufficient serum 25(OH)D levels. Statistically, these participants could cancel out the effect of the other, explaining why no association is seen between serum 25(OH)D levels and mortality. Once patients are treated with a high enough dose of cholecalciferol, we can then see an effect on reduction in mortality risk, as the dose will be high enough to overcome low vitamin D responder status. However, this is merely supposition, and large-scale clinical studies would need to be carried out in order to validate this theory.
Nonetheless, cholecalciferol as a potential therapeutic option for COVID-19 is an attractive prospect, given its wide availability and low cost, particularly in developing nations, as well as its relatively safe side-effect profile, in conjunction with regular monitoring of serum levels and serum adjusted calcium. Small-scale clinical trials are already beginning to populate the literature. In a pilot study of 76 patients hospitalised with COVID-19 in a single Spanish centre, Entrenas Castillo et al. found that fewer patients who were treated with calcifediol (hydroxylated cholecalciferol, also known as 25-hydroxyvitamin D3
) were admitted to the intensive therapy unit when compared to controls [26
]. Interestingly, 25(OH)D levels were not available on participants in this trial, although it is now being scaled up to include 15 hospitals across Spain. In addition, Rastogi et al. have recently completed a randomised placebo-controlled trial of high-dose cholecalciferol therapy (60,000 IU cholecalciferol for 7 days) in 40 SARS-CoV-2-positive patients who were asymptomatic or only mildly symptomatic, based in a single centre in India [27
]. A greater proportion of patients in the treatment group achieved SARS-CoV-2 negativity at the end of the 14-day study period, and fibrinogen levels (which were used as a biomarker of inflammatory response) were significantly lower in the treatment arm. These studies are making headway in ascertaining the role of cholecalciferol/calcifediol therapy both at the pre-hospital and post-hospitalisation stage, and further work must be done to ascertain optimum dosage regimens and whether prophylactic therapy is of benefit in the setting of COVID-19 infection.
It is unsurprising that predictors such as age > 73 years, IHD and baseline creatinine > 83 μmol/L are associated with increased risk of death from COVID-19 (as seen in the multivariate analysis in the validation cohort), as these are likely to represent patients with a poorer baseline of health with less physiological reserve to adequately cope with acute COVID-19 infection. These patients are also in a group at risk of vitamin D insufficiency or deficiency. The June 2020 report from the Office of National Statistics, ONS (covering England and Wales, where this study population was recruited from), shows an exponential rise in age-specific mortality rates as age increases [28
], and this fits with our data. Furthermore, a large multi-centre study in Italy found that older age, chronic kidney disease and coronary artery disease were more common in patients who died [29
], and our findings agree with this. We found the opposite with the age > 74 years variable in the primary cohort following adjustment for multiple confounders, but this is likely to be a type 1 error, due to the smaller sample size (and hence, reduced power) of the primary cohort. We do not feel that this weakens any associations, however, as the one finding that has replicated across both cohorts is that high-dose cholecalciferol booster therapy is protective of COVID-19 mortality.
This study’s strengths lie in its recruitment of almost 1000 acute COVID-19 hospital in-patients from three separate centres, with a high proportion of patients with available serum 25(OH)D levels. It is also one of the largest observational studies, to date, to assess the benefit of cholecalciferol therapy in reducing the risk of COVID-19 mortality. The fact that our findings replicate across participants recruited from independent study populations strengthens the association between high-dose cholecalciferol booster therapy and a reduced risk of COVID-19 mortality. Our study population is potentially generalisable to the rest of the UK, with similar demographics across centres compared with the ONS figures on COVID-19 admissions that were most up to date at the time that recruitment closed [28
]. Patients were recruited from throughout the pandemic, so our population reflects changing treatment recommendations as the pandemic evolved and evidence increased.
There are potential limitations to our study. For instance, not all patients had serum 25(OH)D levels available, so power may have been improved with more values. In addition, while results are potentially generalisable to the UK, results would need to be replicated in different international populations to assess transferability of findings globally. Another limitation lies in the fact that, although we imposed a 12-week time limit on pre-admission serum 25(OH)D measurements, the half-life of this form of vitamin D is 15–25 days, less than our 84-day limit, which we imposed to mitigate for seasonal variation whilst also aiming to include as many values as possible. However, only 14/227 participants (6.2%) with available serum 25(OH)D levels had measurements over 25 days pre-admission, with those 14 participants’ values having a median of 51 days (IQR 41, 63) prior to admission. Finally, due to the cross-sectional nature of this study, we are unable to ascertain cause and effect between associations, and we do not have a mechanistic understanding of our findings as yet. A longitudinal analysis of outcomes must be carried out in the future to determine any long-term sequelae of deficient vitamin D status during acute COVID-19 infection. There is also the potential for studies in whole blood and/or tissue to understand the mechanisms behind vitamin D status and COVID-19 severity.