Multimorbidity, polypharmacy, and COVID-19 infection within the UK Biobank cohort

Background It is now well recognised that the risk of severe COVID-19 increases with some long-term conditions (LTCs). However, prior research primarily focuses on individual LTCs and there is a lack of data on the influence of multimorbidity (≥2 LTCs) on the risk of COVID-19. Given the high prevalence of multimorbidity, more detailed understanding of the associations with multimorbidity and COVID-19 would improve risk stratification and help protect those most vulnerable to severe COVID-19. Here we examine the relationships between multimorbidity, polypharmacy (a proxy of multimorbidity), and COVID-19; and how these differ by sociodemographic, lifestyle, and physiological prognostic factors. Methods and findings We studied data from UK Biobank (428,199 participants; aged 37–73; recruited 2006–2010) on self-reported LTCs, medications, sociodemographic, lifestyle, and physiological measures which were linked to COVID-19 test data. Poisson regression models examined risk of COVID-19 by multimorbidity/polypharmacy and effect modification by COVID-19 prognostic factors (age/sex/ethnicity/socioeconomic status/smoking/physical activity/BMI/systolic blood pressure/renal function). 4,498 (1.05%) participants were tested; 1,324 (0.31%) tested positive for COVID-19. Compared with no LTCs, relative risk (RR) of COVID-19 in those with 1 LTC was no higher (RR 1.12 (CI 0.96–1.30)), whereas those with ≥2 LTCs had 48% higher risk; RR 1.48 (1.28–1.71). Compared with no cardiometabolic LTCs, having 1 and ≥2 cardiometabolic LTCs had a higher risk of COVID-19; RR 1.28 (1.12–1.46) and 1.77 (1.46–2.15), respectively. Polypharmacy was associated with a dose response higher risk of COVID-19. All prognostic factors were associated with a higher risk of COVID-19 infection in multimorbidity; being non-white, most socioeconomically deprived, BMI ≥40 kg/m2, and reduced renal function were associated with the highest risk of COVID-19 infection: RR 2.81 (2.09–3.78); 2.79 (2.00–3.90); 2.66 (1.88–3.76); 2.13 (1.46–3.12), respectively. No multiplicative interaction between multimorbidity and prognostic factors was identified. Important limitations include the low proportion of UK Biobank participants with COVID-19 test data (1.05%) and UK Biobank participants being more affluent, healthier and less ethnically diverse than the general population. Conclusions Increasing multimorbidity, especially cardiometabolic multimorbidity, and polypharmacy are associated with a higher risk of developing COVID-19. Those with multimorbidity and additional factors, such as non-white ethnicity, are at heightened risk of COVID-19.

For example, in both the UK and the USA, age (�70 in UK and �65 in USA), severe obesity (BMI �40 kg/m 2 ), being immunocompromised, and certain LTCs (e.g. cardiac/respiratory/ renal disease) are used to identify those at higher risk of severe disease [14,15]. Individuals who meet any of these criteria have been asked to be more cautious and adhere to public health guidance (e.g. social distancing) more strictly than the general population. In the UK, this higher-risk group is distinct from those with specific LTCs (e.g. leukaemia) who are considered 'extremely high risk' and, as a result, are asked to 'shield' and not leave their homes at all [16].
To date, LTC prognostic factors for severe COVID-19 primarily involve single conditions and there is a lack of data on the influence of multimorbidity on the risk of COVID-19 [5,13,17]. Multimorbidity prevalence is increasing worldwide and is associated with higher mortality, with different disease clusters associated with even higher mortality [18][19][20][21]. Polypharmacy, closely linked to multimorbidity [22], is also associated with adverse health outcomes [23]. Therefore, it is plausible that the number of LTCs, type of LTCs, and polypharmacy are associated with a higher risk of developing severe COVID-19. Furthermore, it is plausible that the adverse consequences of multimorbidity on COVID-19 risk may be greater for population subgroups (e.g. people aged �65 years, or those with a BMI �40kg/m 2 ). Such knowledge would benefit clinicians, as this is routinely collected information in many clinical settings. Using a large UK population cohort, UK Biobank, we aimed to investigate: 1. the association between multimorbidity (by number and type of LTC, and by level of polypharmacy) and COVID-19.
2. the potential effect modification of known COVID-19 sociodemographic and physiological risk factors on the association between multimorbidity and COVID-19.

Study design and data collection
Data came from UK Biobank, a longitudinal population-based cohort of 502,503 participants aged between 37-73 years old at baseline from England, Wales and Scotland [24]. Baseline data were collected across 22 assessment centres between 2006-2010. UK Biobank contains detailed biological measurements and self-reported demographic, lifestyle, and health information elicited by touch-screen questionnaire and nurse-led interview. COVID-19 test samples were collected and processed between 16 March 2020 and 18 th May 2020. COVID-19 test results were provided by Public Health England (PHE) [25]. Data presented here are from participants recruited from 16 assessment centres located in England only (5 participants with COVID-19 test data were excluded as they attended baseline assessment centres in Scotland or Wales). Participants who had died prior to the last available mortality register extraction (14 February 2018) were also excluded. This resulted in a final eligible study population of 428,199 participants. This study was conducted as part of UK Biobank project number 14151 and is covered by the generic ethics approval for UK Biobank studies from the NHS National Research Ethics Service (16/NW/0274).

Outcomes
Our outcome of interest was confirmed COVID-19 infection (defined as at least one positive test result). Whether or not participants received a COVID-19 test was used as an outcome for a secondary analysis, presented in Supplementary Material (S1-S3 Tables).

Measurement of LTCs
Physician-diagnosed LTCs were self-reported and confirmed at nurse-led interview at baseline. Number of LTCs, based on 43 commonly occurring LTCs from previous UK Biobank work [21], were categorised into 0, 1, and �2. When used as a predictor variable for COVID-19, LTCs were modelled using three measures: total number of LTCs; number of cardiometabolic LTCs (diabetes, coronary heart disease, atrial fibrillation, chronic heart failure, chronic kidney disease, hypertension, stroke/transient ischaemic attack, or peripheral vascular disease); and number of respiratory LTCs (asthma, chronic obstructive pulmonary disease, chronic bronchitis, emphysema, or bronchiectasis).

Exposures
As for LTC and polypharmacy measures, all exposures were based on assessment at the time of recruitment. Sex was collected as a binary variable (male/female). Age at time of COVID-19 test was calculated using age at baseline, and dates of baseline assessment and COVID-19 test extract, and categorised as 48-59, 60-69, and 70-86 years. Ethnicity was self-reported and categorised as Asian/Asian British, black/black British, Chinese, mixed, white and other. Socioeconomic deprivation was measured using the Townsend score of participants' postcode of residence derived from Census data on car ownership, household occupancy, unemployment, and occupation and categorised into quintiles [26]. Smoking status was dichotomised as never and current/former. Frequency of alcohol intake was categorised into: "Never or special occasions only", "1-3 times a month", "1-4 times a week", or "Daily or almost daily". Level of physical activity was defined as "none", "low", "medium", or "high" using Metabolic Equivalent Task (MET) scores based on the International Physical Activity Questionnaire (IPAQ) scoring protocol [27]. Assessment centre location, a categorical variable, describes attendance at one of sixteen assessment centres included in this study. BMI was derived from weight and height and categorised as <18. 5, 18.5-25, 25-30, 30-35, or >35 kg/m 2 . The UK Biobank study protocol is described in more detailed online [28].

Statistical analyses
We compared participants who tested positive for COVID-19 with those who tested negative or were not tested, based on sociodemographics (age, sex, Townsend score and ethnicity), lifestyle (smoking status, frequency of alcohol intake, physical activity), BMI, LTCs, and medication counts using χ 2 tests. Poisson regression models were then used to test for an association between outcome measure (confirmed COVID-19 infection vs. no COVID-19 infection, including those with a negative result and those not tested) and number of LTCs (all, cardiometabolic and respiratory LTCs)/level of polypharmacy. We chose Poisson regression models rather than logistic regression to provide more interpretable relative risks as opposed to odds ratios [32]. Participants with no LTCs, no cardiometabolic LTCs, no respiratory LTCs, or not taking any medication formed the respective reference groups. Two models were conducted, adjusting for 1) sociodemographic variables (as above plus assessment centre location), and 2) as for model 1 adjustments plus lifestyle variables and BMI. For both χ 2 tests and Poisson regression models, p<0.01 was considered statistically significant.
Next, we assessed whether there was evidence of effect modification on an additive scale by examining how the association between multimorbidity and COVID-19 differed across strata of known COVID-19 risk factors: sex, age, ethnicity, Townsend score, smoking, physical activity, BMI, systolic blood pressure, and eGFR. For each risk factor, the reference category was participants in the lowest risk category for that risk factor who also had no LTCs. We used Poisson regression models to examine the association between LTC and COVID-19 across the other combinations of LTC and risk factor categories. We tested formally for interactions by performing ANOVA tests between two models for each risk factor: one model containing an interaction term between the risk factor and number of LTCs, and one without the interaction term. Interactions were considered significant if p<0.01 for each ANOVA. As a secondary analysis we re-ran the analyses with the outcome as tested vs. not tested for COVID-19. The number of participants in each model varied depending on the proportion missing data for any included variable, however, the maximum proportion of missing data for any variable of interest was 2.9% (N = 12,350) for systolic blood pressure. All analyses were conducted using R studio v.1.2.1335 operating R v.3.6.1.

Demographic and lifestyle factors
Of 428,199 eligible participants, 1,324 (0.31%) tested positive for COVID-19. Participants who tested positive for COVID-19 were older and more likely to be male, non-white, in the most deprived quintile, current/former smokers, drink alcohol rarely, to have a BMI of �40 kg/m 2 , do no physical activity, to have more LTCs (including cardiometabolic and respiratory LTCs), and to be taking more medications, compared to those who did not have a positive COVID-19 test (Table 1).

Multimorbidity and COVID-19
In the fully adjusted model (

Polypharmacy and COVID-19
Compared to those taking no medications, in the fully adjusted model (Model 2) there was a clear dose relationship whereby the risk of a COVID-19 positive test rose steadily with polypharmacy level (

Interactions between COVID-19 prognostic factors and multimorbidity
For all risk factors examined, there was a general trend of higher risk of a positive COVID-19 test with increasing LTCs and at-risk subgroups (Fig 1; Table 3). Further, for all prognostic factors, in participants with �2 LTCs there was a higher risk of a positive COVID-19 test for each risk factor subgroup when examining ethnicity, physical activity, BMI, systolic blood pressure, and eGFR. For some factors, having �2 LTCs was associated with an increased risk of COVID-19 infection only in the at-risk sub-group: males when examining sex; >65 years when examining age; and current/previous smoker when examining smoking status. When observing the effect of socioeconomic status, significantly higher risk of COVID-19 infection was apparent in the more deprived quintiles. There was no evidence of a multiplicative interaction between all risk factors modelled and number of LTCs. However, there appeared to be an additive effect whereby the combination of multimorbidity and each prognostic factor was associated with greater risk of COVID-19 infection. Of those with no LTCs, being �65 years old was associated with 34% lower risk of COVID-19 compared with those <65 years old. Similar results overall were observed in the secondary analysis, where the above analyses were repeated in those who were tested for COVID-19 versus those who were not (S1-S3 Tables).

Summary of key findings
This study showed that participants with multimorbidity (�2 LTCs) had a 48% higher risk of a positive COVID-19 test, those with cardiometabolic multimorbidity had a 77% higher risk, and those with respiratory multimorbidity a 78% higher risk (albeit not statistically significant, p = 0.02), compared to those without that type of multimorbidity. Importantly, those from non-white ethnicities with multimorbidity had nearly three times the risk of having COVID-19 infection compared to those of white ethnicity, suggesting that those from minority ethnic groups with multimorbidity are at particular risk. COVID-19 prognostic factors appeared to have an additive effect by further increasing the risk of a positive COVID-19 test in those with multimorbidity.

Comparison with previous literature
Previous literature has suggested that the presence of single LTCs such as hypertension, diabetes, or COPD increase the risk of COVID-19 [33]. In an English primary care research cohort of 3,802 patients, confirmed COVID-19 infections were associated with male sex, black ethnicity, deprivation, chronic kidney disease, and urban setting [30]. A preprint report of linked primary care data on 17,425,445 adults in England, showed that older age, male sex, deprivation, and black and Asian ethnicity were associated with higher risk of COVID-19 related in-hospital deaths [11]. However, our study is the first to show a higher risk of a positive COVID-19 test in those with �2 LTCs and particularly in those with cardiometabolic multimorbidity. It has also been suggested that the presence of multimorbidity could further increase the risk of adverse outcomes for people with COVID-19 [33]. Guan et al, in a cohort of 1,590 hospital patients with confirmed COVID-19, found a higher risk of a composite COVID-19 outcome (intensive care admission, invasive ventilation, or death) in those with 1 (HR 1.79 (95%CI 1.16-2.77)) and 2 (HR 2.59 (95%CI 1.61-4.17) comorbidities [9]. However, while they provided details of specific LTCs most likely to be present, especially in severe cases of infection (hypertension, cardiovascular/cerebrovascular diseases, diabetes, hepatitis B infections, COPD, chronic kidney diseases and malignancy), they did not describe which patterns of multimorbidity were associated with the greatest risk. A report on 3,200 patients with COVID-19 from Italy showed that, of the 481 patients who died, 48.6% had 3 or more comorbidities. However, again, while it listed the specific LTCs associated with increased risk, such as ischaemic heart disease and diabetes, it did not describe which patterns of multimorbidity were associated with the highest risk [13].

Strengths and limitations
As a large prospective cohort with rich demographic, lifestyle, health, and anthropometric data linked to COVID-19 test results, UK Biobank provides a valuable opportunity to examine the predictors for COVID-19 [34,35]. In particular, the rich data allowed examination of the risk of a positive COVID-19 test in those with multimorbidity and the influence of a range of known sociodemographic, lifestyle and physiological risk factors for COVID-19. However, our study has a number of limitations. Firstly, the proportion of UK Biobank participants with COVID-19 test data is currently low (1.05%) which resulted in wide confidence intervals for groups with few participants. Secondly, the denominator for the test group included all those who had a negative test result as well as those who were not tested at all. At the time for which COVID-19 test data are available, the strategy in the UK had been to only test those in hospital (emergency department and inpatient) settings. This means the positive COVID-19 participants are likely to have had sufficiently severe clinical signs and symptoms to justify hospital assessment and those with COVID-19 but with mild symptoms are less likely to have been tested. Our results are therefore likely to reflect the associations with more severe COVID-19 disease. It is not known if the associations identified in this study would be similar for those with milder COVID-19 disease. Thirdly, exposures and moderators examined here were assessed at baseline only and may have changed during follow up. Depending on the direction and level of change since recruitment, more up-to-date data on exposures (e.g. smoking, alcohol intake, physical activity, and post code of residence) could have provided different results. However, LTC and medication count are likely to have remained the same or increased with time and therefore our effect size estimates may be conservative. Fourthly, UK Biobank participants are not representative of the general population and are acknowledged to be mostly white British, more affluent, and healthier than the general population [36]. Consequently, absolute values may not be generalisable, however, effect size estimates will be and strongly agree with more representative cohorts [37]. Finally, participants in this study were aged between 48-86 years old and associations between multimorbidity and COVID-19 may be different for younger age groups. This may be particularly important for participants from more deprived backgrounds who are more likely to have multimorbidity at a younger age [38,39].

Implications
Our work has clinical and practical implications as more countries navigate lifting COVID-19 restrictions. It demonstrates that multimorbidity, particularly cardiometabolic multimorbidity and polypharmacy, are strongly associated with COVID-19 infection and suggests that those who have multimorbidity coupled with additional risk factors, such as non-white ethnicity, are at increased risk. Such individuals should be particularly stringent in adhering to preventive measures, such as physical distancing and hand hygiene. Our findings also have implications for clinicians, occupational health and employers when considering work-place environments, appropriate advice for patients, and adaptations that might be required to protect such staff.
Future research is needed to corroborate these findings in other countries and in people from different ethnic backgrounds. We know that patterns of multimorbidity differ across ethnic groups and have different associations with mortality, so this merits further investigation [40]. We also need to explore the implications of different patterns of multimorbidity on COVID-19 related health care outcomes in the short and long term.

Conclusion
This study suggests that multimorbidity, cardiometabolic disease, and polypharmacy are associated with COVID-19. Those with multimorbidity who were also of non-white ethnicity, from the most socioeconomically deprived backgrounds, those who were severely obese, or who had reduced renal function, had more than twice the risk of COVID-19 infection.
More work is required to develop risk stratification for COVID-19 in people with different patterns of multimorbidity in order to better define those individuals who would benefit from enhanced preventive measures in public, work, and residential spaces.
Supporting information S1