Association of pre-pandemic high-density lipoprotein cholesterol with risk of COVID-19 hospitalisation and death: The UK Biobank cohort study

There is growing evidence of, and biological plausibility for, elevated levels of high-density lipoprotein cholesterol (HDL-C) being related to lower rates of respiratory disease. We tested whether pre-pandemic HDL-C within the normal range is associated with subsequent COVID-19 hospitalisations and death. We analysed data on participants from UK Biobank, a prospective cohort study, baseline data for which were collected between 2006 and 2010. Follow-up for COVID-19 was via hospitalisation records (1845 events in 317,306 individuals) and a national mortality registry (458 deaths in 317,833 individuals). After controlling for a series of confounding factors which included health behaviours, inflammatory markers, and socio-economic status, higher levels of HDL-C were related to a lower risk of later hospitalisation. The effect was linear (p-value for trend 0.001), whereby a 0.2 mmol/L increase in HDL-C was associated with a 7% lower risk (odds ratio; 95% confidence interval: 0.93; 0.90, 0.96). Corresponding relationships for mortality were markedly weaker, such that statistical significance at conventional levels were not apparent for both the linear trend (p-value 0.25) and the odds ratio per 0.2 mmol/L increase (0.98; 0.91, 1.05). While our finding for HDL-C and hospitalisations for COVID-19 raise the possibility that favourable modification of this cholesterol fraction via lifestyle changes or drug intervention may impact upon the risk of the disease, it warrants testing in other studies.


Introduction
High-density lipoprotein cholesterol (HDL-C) has traditionally been linked with coronary heart disease and stroke. 1 While conventional epidemiological studies consistently demonstrate that elevated levels of this cholesterol fraction confer protection against vascular events, 2 support for such a gradient has been lacking in people genetically predisposed to low concentrations of HDL-C, 3 Mendelian randomisation studies, 4 and randomized clinical trials utilising HDL-C-elevating medication. 5 More recently, HDL-C has been implicated in the pathogenesis of other health endpoints, including infectious disease. While most investigators testing the link between HDL-C and infection have done so in prognostic studies of patients groups, 8 in one of the few cohort analyses of apparently healthy individuals, people with lower levels of baseline HDL-C experienced a greater subsequent risk of hospitalisation for gastroenteritis, urinary tract infection, and bacterial pneumonia. 9 Plausible mechanisms include the HDL-C-mediated sequestration of pathogen-associated lipids, and regulation of immune cells proliferation, maturation, and function which results in neutralization or clearance of pro-inflammatory endotoxins. 7 This evidence base raises the possibility of a link between HDL-C and COVID-19, the disease caused by severe acute respiratory syndrome coronavirus 2. Using UK Biobank, a prospective cohort study, we have recently shown that an unfavourable pre-pandemic vascular risk factor profile -low HDL-C included -is associated with a higher risk of hospitalization for COVID-19. 10 Whether HDL-C across the normal range offers predictive capacity for COVID-19 hospitalisations is, however, untested, and this is the purpose of the present study. Further, as the present pandemic has unfolded, this cohort has . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint accumulated sufficient deaths from this disease to facilitate analyses with the aim of corroborating any associations with hospitalisations.

Study population
We used data from the UK Biobank, a prospective cohort study, 11

Baseline data collection
At baseline, non-fasting venous blood samples were drawn and assayed for total cholesterol, HDL-C, and triglycerides using a Beckman Coulter AU5800 analytical platform. Low density lipoprotein (LDL)-cholesterol values were calculated using the Friedewald equation. 12 Total blood count (leukocyte, platelet, haemoglobin) as markers of immune function were analysed using an automated Coulter LH 750. Physiciandiagnosed cardiovascular disease (heart attack, angina, stroke), diabetes, cholesterollowering drugs use, cigarette smoking, alcohol intake, highest educational attainment, ethnicity, 13 number of people living in the household, and physical activity in the prior month were self-reported using standard enquiries. 14 Body mass index was computed using direct measurements of height and weight using the usual formulae. 15 Hypertension was defined as elevated measured blood pressure (≥140/90 mmHg) and/or use of antihypertensive medication. Townsend index of neighbourhood deprivation was based on postcode linkage. 14 Provided by Public Health England, data on COVID-19 status in . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Statistical analyses
We used logistic regression analyses to compute odds ratios with accompanying 95% confidence intervals to summarise the relation between HDL-C and later COVID-19 hospitalisation or death. In a first analytical approach, we created an HDL-C variable with eight categories which was designed to examine the shape of the HDL-C-COVID-19  were collapsed owing to a lower number of events. With preliminary analyses suggesting a linear gradient, we were then able to summarise the relationship for a unit change in HDL-C (0.2 mmol/L increase). In taking both approaches, we first adjusted for age and sex (comparator model) and then, in the multivariable model, a series of covariates which included inflammatory markers, lifestyle factors, and socioeconomic circumstances.

Results
In 317,306 (171,466 women) participants with complete data on baseline covariates, there were 869 hospitalisations for COVID-19 (427 in women) during the surveillance period. As illustrated in figure 1, in age-and sex-adjusted analyses, relative to the group with the lowest concentration of HDL-C, those in the highest experienced around one third of the . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint risk of hospitalization for COVID-19 (odds ratio; 95% confidence interval: 0.31; 0.19, 0.52). There was also evidence of a stepwise relationship (p-value linear trend <0.001) such that lower disease risk was apparent in people with higher level of this cholesterol fraction. Summarising this trend, a 0.2 mmol/L increase in HDL-C was associated with a . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Discussion
To the best of our knowledge, this is the first study to examine the shape of the relationship between pre-pandemic HDL-C levels and risk of later COVID-19 events. Our salient finding was that, net of an array of confounding factors, higher concentrations of HDL-C were associated with protection against hospitalisation for, and death from, the disease. That COVID-19 events ascertained using different approaches revealed very similar effects increasing confidence in our novel results. In other analyses from the present dataset we have reported what are now regarded as established associations of classic vascular risk factors such as raised levels of glycosylated haemoglobin, 16 body weight, 15 and blood pressure 10 with COVID-19, as apparent in studies based in the US, 17 Italy, 18 China, 19,20 and Brazil. 21 We have also shown higher rates of hospitalisation for the disease in people of increased age, male sex, socioeconomic disadvantage, 14 and ethnic minority groups. 13 Comparison with existing studies Levels of HDL-C appear to change in the presence of coronavirus disease 2019 (COVID- 19), such that, relative to healthy controls, patients with COVID-19 have lower levels of HDL-C concentrations in conjunction with acute elevation in systemic inflammatory markers. 22,23 While a lowering of HDL-C levels appears to be a consequence of COVID-19, the reverse, whereby pre-pandemic HDL-C levels may offer some predictive capacity for this disease, has been little examined. Taking a Mendelian Randomisation approach, investigators using UK Biobank data have shown an inverse association between baseline HDL-C and hospitalisations for any infectious disease up to 10 years later. 24 In contrast, using a conventional epidemiological study design, analyses based on two Danish . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint community-based, prospective cohort studies with up to 20 years of follow-up revealed a 'U'-shaped relationship whereby the greatest risk of hospitalisation for any infection was apparent at opposing ends of the HDL-C continuum. 9 There was no clear evidence of such a quadratic effect in the present study. Studies linking with respiratory disease and earlier measurement of HDL-C are scarce and we are unaware of any other analyses with COVID-19 as the endpoint.

Study strengths and weaknesses
The strengths of this study include the measurement of biomarkers that preceded the onset of COVID-19, so ruling out reverse causality. While large, the study sample is also well-characterised. That UK Biobank participants represent only 6% of the target population, however, means that the present data cannot be used to estimate prevalence or incidence in the general population, although established risk factor associations appear generalisable. 27 HDL-C levels were measured up to 14 years before COVID-19 case assessment and this raises concerns about their utility for current values, however, in a reassessment a mean of 4.4 years after baseline examination they showed high test-. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint 9 retest stability (correlation coefficient 0.85, p<0.001) in a subsample (N=13,430). While the HDL-C-COVID-19 gradient was robust to the adjustment of various covariates, it is plausible that unmeasured confounding factors might explain the association. To this extent, because the data are observational, we cannot be dogmatic about causality.
Further scrutiny of our results, including the application of the Mendelian Randomisation approach where a genetic proxy for HDL-C is used as the exposure of interest, is required.
In conclusion, the novel association between higher levels of HDL-C and lower risk of hospitalisation and death due to COVID-19 warrant testing in using other study designs.
. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted February 9, 2021.  Multiply adjusted odds ratios are adjusted for age, sex, ethnicity, education, number in household, area deprivation, body mass index, leisure time physical activity, alcohol intake, smoking habit, diagnosed diabetes, cardiovascular disease, or hypertension, cholesterol-lowering medication, LDLcholesterol, triglycerides, haemoglobin, white blood cell, and platelet count.
. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint Multiply adjusted odds ratios are adjusted for age, sex, ethnicity, education number in household, area deprivation, body mass index, leisure time physical activity, alcohol intake, smoking habit, diagnosed diabetes, cardiovascular disease, or hypertension, cholesterol-lowering medication, LDLcholesterol, triglycerides, haemoglobin, white blood cell, and platelet count.
. CC-BY-NC 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 9, 2021. ; https://doi.org/10.1101/2021.01.20.21250152 doi: medRxiv preprint