Association Between Accelerometer-Assessed Physical Activity and Severity of COVID-19 in UK Biobank

Objective To quantify the association between accelerometer-assessed physical activity and coronavirus disease 2019 (COVID-19) outcomes. Methods Data from 82,253 UK Biobank participants with accelerometer data (measured 2013-2015), complete covariate data, and linked COVID-19 data from March 16, 2020, to March 16, 2021, were included. Two outcomes were investigated: severe COVID-19 (positive test result from in-hospital setting or COVID-19 as primary cause of death) and nonsevere COVID-19 (positive test result from community setting). Logistic regressions were used to assess associations with moderate to vigorous physical activity (MVPA), total activity, and intensity gradient. A higher intensity gradient indicates a higher proportion of vigorous activity. Results Average MVPA was 48.1 (32.7) min/d. Physical activity was associated with lower odds of severe COVID-19 (adjusted odds ratio per standard deviation increase: MVPA, 0.75 [95% CI, 0.67 to 0.85]; total, 0.83 [0.74 to 0.92]; intensity, 0.77 [0.70 to 0.86]), with stronger associations in women (MVPA, 0.63 [0.52 to 0.77]; total, 0.76 [0.64 to 0.90]; intensity, 0.63 [0.53 to 0.74]) than in men (MVPA, 0.84 [0.73 to 0.97]; total, 0.88 [0.77 to 1.01]; intensity, 0.88 [0.77 to 1.00]). In contrast, when mutually adjusted, total activity was associated with higher odds of a nonsevere infection (1.10 [1.04 to 1.16]), whereas the intensity gradient was associated with lower odds (0.91 [0.86 to 0.97]). Conclusion Odds of severe COVID-19 were approximately 25% lower per standard deviation (∼30 min/d) MVPA. A greater proportion of vigorous activity was associated with lower odds of severe and nonsevere infections. The association between total activity and higher odds of a nonsevere infection may be through greater community engagement and thus more exposure to the virus. Results support calls for public health messaging highlighting the potential of MVPA for reducing the odds of severe COVID-19.

 I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above.  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above. X A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above. Confirm that all information provided is accurate by checking the box below and then enter your complete name as it appears at the top of this form and date of confirmation in the fields below.
X I certify that all information I have provided is accurate. Each author should meet all criteria below (A, B, C, and D) and should indicate general and specific contributions by reading criteria A, B, C, and D and checking the appropriate boxes.
x A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above.
x B. I have given final approval of the submitted manuscript. Confirm that all information provided is accurate by checking the box below and then enter your complete name as it appears at the top of this form and date of confirmation in the fields below.
x I certify that all information I have provided is accurate. √ A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above.  Confirm that all information provided is accurate by checking the box below and then enter your complete name as it appears at the top of this form and date of confirmation in the fields below.
√ I certify that all information I have provided is accurate.  X A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above. X A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above. X A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above. ☑ A. I certify that  The manuscript represents original and valid work and that neither this manuscript nor one with substantially similar content under my authorship has been published or is being considered for publication elsewhere, except as described in the cover letter submitted with the manuscript, and copies of closely related manuscripts have been provided; and  I agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity or any part of the work are appropriately investigated and resolved; and  If requested, I will provide the data or will cooperate fully in obtaining and providing the data on which the manuscript is based for examination by the editor or their assignees; and  For papers with more than 1 author, I agree to allow the corresponding author to serve as the primary correspondent with the editorial office, to review the edited manuscript and proof, and to make decisions regarding release of information in the manuscript to the media, federal agencies, or both, or, if I am the only author, I will be the corresponding author and agree to serve in the roles described above.  Confirm that all information provided is accurate by checking the box below and then enter your complete name as it appears at the top of this form and date of confirmation in the fields below.
☑ I certify that all information I have provided is accurate. J o u r n a l P r e -p r o o f

Introduction
Poor outcomes from COVID-19 are more likely in people who are older [1] , more deprived [2] , have comorbidities [3] , and/or are from ethnic minority populations [4] . As with chronic disease, research suggests that risk factors related to health behaviours, such as obesity [5] and slow walking pace [6] , also negatively impact on COVID-19 outcomes.
Physical activity is a modifiable health behaviour that may mitigate the risks of COVID-19 [7] . This could occur though reductions in chronic inflammation [8,9] and/or cardiometabolic risk factors [10] , which are associated with an increased risk of COVID-19 [11] , and/or through enhanced immunity [7] . In the early months of the pandemic (up to July 2020), we reported initial observations from Biobank data [12] , which was suggestive evidence for lower odds (up to 20%) of severe COVID-19 per 30 minutes daily moderate-to-vigorous physical activity (MVPA, P = 0.06). Consistent with this finding, Sallis et al. [13] reported that being physically inactive is the strongest modifiable risk factor for severe COVID-19 and stressed the importance of this message for public health. They showed that the risk of hospitalisation or death with COVID-19 up to October 2020 in a sample of US health plan members was over twice as high in people who self-reported consistently being inactive in the two years prior to the pandemic, compared to those who self-reported consistently meeting the guidelines of 150 minutes MVPA per week [13] .
While most of the evidence on associations between physical activity and a wide range of health outcomes has similarly been gleaned from self-report methods [14] , self-report has well-documented limitations; not least the reliance on recall and, consequently, a focus on purposeful activity [14] .
Accelerometers directly measurement movement, reducing measurement error and facilitating a more nuanced consideration of physical activity, e.g., the relative importance of the total amount and/or intensity of physical activity [15,16] . UK Biobank is the largest dataset with accelerometerassessed physical activity, having assessed ~100,000 participants [17] .

J o u r n a l P r e -p r o o f
In this study, we use UK Biobank data from the first and second waves of the COVID-19 pandemic to determine the association between accelerometer-assessed physical activity (measured between 2013 and 2015) and severe and non-severe COVID-19 outcomes. We hypothesised that physical activity would be associated with reduced odds of severe COVID-19. However, for non-severe COVID-19, which reflects community transmission, we hypothesised associations would be attenuated as higher physical activity levels may also reflect greater exposure to the virus.

Study Design
This is a retrospective observational study. Physical activity was assessed between June 2013 and December 2015, 4-7 years preceding the COVID-19 pandemic.

Setting and cohort
This study uses data from UK Biobank (application 36371), a prospective cohort of >500,000 adults aged 40-69 years with baseline assessments conducted between March 2006 and July 2010 [18] .  Figure S1 in Supplementary material.

COVID-19 outcomes
J o u r n a l P r e -p r o o f Two outcomes were reported: 1) severe infection with SARS-CoV-2; 2) non-severe infection for SARS-CoV-2. A positive test for SARS-CoV-2 with hospitalisation or death related to the disease (i.e., any death with an ICD-10 code of U07.1 or U07.2 as the primary cause of death on the death certificate) was considered as evidence of a severe infection [19] . A positive test for SARS-CoV-2 from a community setting without a hospital diagnosis or death was considered evidence of probable mild disease or a non-severe infection. Classifying a positive test for SARS-CoV-2 in those admitted to hospital as a marker of disease severity within UK Biobank is in line with guidance for this dataset [19] .
However, actual disease severity cannot be confirmed from the linkage data available. Therefore, our reference to the composite of a test result for SARS-CoV-2 in those admitted to hospital or death from COVID-19 as indicating "severe" disease is for descriptive purposes only.

Physical activity
A sub-sample of ~100,000 adults were asked to wear the Axivity AX3 wrist-worn accelerometer (Axivity, Newcastle, UK) 24 hours a day for seven days between June 2013 and December 2015 [17] .
For each participant, we extracted the accelerometer data (5-second epoch time series) from UK Biobank [17] and converted it to R-format for processing and analysis with GGIR (version 1.11-0, http://cran.r-project.org) [20] . Participants were excluded if they failed calibration, had fewer than three days of valid wear (defined as >16 h per day), or wear data were not present for each 15 min period of the 24 h cycle [12,15,21] . Accelerometer outcomes, selected to describe total physical activity and its intensity, were:  average acceleration over the 24 h day (proxy for total physical activity, mg)  intensity gradient over 24 h (intensity distribution of activity over the day; higher values indicate a greater proportion of total activity is spent at high intensity [21]  time spent in 1-min bouts of MVPA (acceleration cut-point 100 mg [22] )

J o u r n a l P r e -p r o o f
Logistic regression was used to analyse associations of physical activity with the COVID-19 outcomes:  [5] ).
Three physical activity exposures were considered: total physical activity, the intensity gradient, and MVPA. A mutually adjusted model was also run for total physical activity and the intensity gradient to test whether associations were independent of the alternate activity metric consistent with previous research assessing the relative contributions of total activity and intensity of activity for health [15,23] . The variables were standardized before entry into the models and the odds ratios (ORs) per cohort standard deviation (SD) reported for ease of comparison across exposures [12] . Covariates Analyses were reported for the full population and stratified by sex. Effect modification by sex was tested using an interaction term (sex*physical activity) in the model.
Statistical significance was set at P<0.05; results are reported with a 95% confidence interval (CI).
Interactions were considered significant at P<0.1. All analyses were performed in Stata version 16.1.

Sensitivity analyses
1. All models were further adjusted for co-variates potentially on the causal pathway from physical activity to COVID-19 risk.
2. As testing in the UK has not been universal, particularly in the first wave of the pandemic, there is a risk of selection bias in models 1 and 3, where those with COVID-19 were compared to those with a negative test or no test. Participants with COVID-19 may not have been tested and thus wrongly allocated to the comparator group. To address this, we carried out sensitivity analyses for models 1 and 3, restricting the comparator group to those who tested negative for COVID-19.  (Figure 1a).

Data
Results for a severe infection relative to those with a non-severe infection (Model 2) were consistent with those for Model 1 (Figure 1b).

Non-severe COVID-19 relative to no COVID-19 (Model 3)
There was an association between total physical activity and higher odds of a non-severe infection

Main findings
Higher physical activity was associated with reduced odds of severe COVID-19; intensity of physical activity was the driving factor with 20-25% lower odds per 30 minutes daily MVPA, e.g., walking.
Associations were stronger in women, with 32%-37% lower odds per 30 minutes daily MVPA, relative to 10-16% lower odds in men. Total physical activity appeared to increase the odds of nonsevere COVID-19. As the incidence of non-severe infections reflects community transmission, this finding likely reflects greater exposure to the virus. In contrast, when adjusted for total activity, a greater proportion of high intensity activity was associated with 7-10% lower odds of infection.
Increased odds of severe COVID-19 with lower total activity and MVPA is consistent with the recent findings from self-reported physical activity in the U.S. [13] , and from UK Biobank early in the pandemic [12] . It is not clear why associations with severe COVID-19 tended to be weaker in men than women for metrics reflecting the intensity of physical activity. Men are known to be at higher risk of severe COVID-19 than women and, while most studies include sex in their analyses as a potential confounder, relatively few studies have reported whether associations with risk factors differ by sex (i.e., whether sex is an effect modifier). However, recently Gao et al. [24] reported no difference in associations between BMI and COVID-19 severity by sex. Conversely, higher odds of severe COVID-19 have been reported for women who work shifts outside of healthcare (2.77 [2.14, 3.59]) than for men who work shifts outside of healthcare (1.59 [1.23, 2.05]) [25] .

Total amount and the intensity of activity
The availability of accelerometer-assessed physical activity enabled us to explore whether the total amount of physical activity and the intensity of that activity were associated with COVID-19 outcomes independent of each other. Independent associations for the intensity gradient for the whole cohort and women for severe COVID-19, alongside reduced odds for MVPA, were observed.
This suggests that the proportion of activity taken at a moderate-to-vigorous intensity is key, e.g., walking and brisk walking, consistent with self-report of meeting physical activity guidelines [13] or having a brisk walking pace [6] . As time spent in MVPA is associated with cardiorespiratory fitness, this is consistent with evidence that cardiorespiratory fitness [26] , is associated with reduced risk of hospitalization due to COVID-19 [27,28] .
Low levels of physical activity contribute to chronic disease [10] and chronic inflammation [9] , which could be a factor in the observed association with severe COVID-19 [11] . Given that COVID-19 is an acute inflammatory disease, inactivity may also exacerbate existing chronic inflammation and, alongside other risk factors (e.g., genetic predisposition, psychological factors), be associated with a 'cytokine storm' contributing to this increased risk of severe COVID-19 [9] .
For non-severe infections, when mutually adjusted, the intensity gradient was again independently associated with reduced odds of infection, but total physical activity was associated with elevated odds. MVPA, which combines moderate and vigorous intensity activity, was not associated with reduced odds. This suggests that intensities greater than moderate, e.g., vigorous activity such as brisk walking and running, may be optimal to reduce the odds of a non-severe infection. It is possible that the observed association between total physical activity and increased odds of a non-severe infection are due to a bias in the people who were more likely to be tested for COVID-19 [29] .
However, results were consistent when restricting the comparator group to people who tested negative for COVID-19. Accumulating evidence indicates that occupation type is associated with COVID-19 outcomes, with elevated risk in essential workers [30] . Notably, approximately three-J o u r n a l P r e -p r o o f quarters of the people with non-severe COVID-19 were employed, relative to approximately half of the cohort. It is possible that high levels of total physical activity accumulated at relatively low intensities reflect working in a job that requires high levels of routine movement/walking and more community engagement, thus increased exposure to the virus. While we controlled for employment status and deprivation it was not possible to determine the level of exposure to infection.
Conversely, vigorous intensity activity may reflect exercise or training type activity that strengthens immunity [7] .

Strengths and limitations
The main strength of this study is the availability of accelerometer-assessed physical activity, facilitating precise characterisation of the physical activity profile of participants in a large population with linked COVID-19 data. In addition, UK Biobank is an extensively phenotyped population, differentiating it from many other datasets currently being analysed to better understand COVID-19. However, there are also some limitations. First and foremost, the characteristics of participants were measured, and accelerometer data collected, some years prior to the pandemic. However, data from a longitudinal study on older adults in England [31] shows steady physical activity levels over time, albeit with a slight decline in vigorous activity. Secondly, the definition of severe COVID-19 was a positive test from a hospital inpatient. While this is consistent with the definition proposed by the researchers that developed the linkage method [19] , and with previous research using the UK Biobank dataset to explore risk factors for COVID-19 [2,5,6,12,30] , actual disease severity cannot be confirmed from the linkage data available. It should also be noted that testing in the UK has not been universal, particularly in the first wave of the pandemic making the analyses vulnerable to bias [29] . Further this is an observational study, thus we cannot exclude the risk of residual confounders due to unmeasured confounders or measurement error. Finally, UK Biobank participants are not representative of the wider population; however, participants may not need to be representative when estimating relative risk factor associations [32] . Taking these limitations into J o u r n a l P r e -p r o o f consideration, our results point to the potential importance of physical activity as predictive of later risk of severe and non-severe COVID-19 infection. Notably, however, the results are consistent with proposed mechanisms and the reduced risk of severe COVID-19 in participants who self-reported consistently meeting physical activity guidelines in the two years prior to the pandemic [13] .

Conclusion
Physical activity appeared to be associated with lower odds of severe-COVID-19, with stronger associations for intensity of movement in women than men. Odds of severe COVID-19 were lower by 37% in women and 16% in men per 30 minutes of daily MVPA. Higher total physical activity appeared to increase the odds of non-severe infection, but a greater proportion of high intensity activity was associated with 8-10% lower odds. Non-severe infections reflect community transmission, thus the greater odds associated with higher total physical activity levels, accumulated at lower intensity, likely reflect greater exposure to the virus, e.g., through occupational activity.
Results from this study are consistent with self-reported data and provide further evidence for the role of physical activity in reducing the odds of a severe infection, and the potential for vigorous activity to play a role in reducing the odds of non-severe infection, possibly through more robust immunity. This is important as a reduction in non-severe infections has potential for reducing community transmission. This study provides further support to calls for public health messaging to highlight the potential of physical activity, particularly of moderate-to-vigorous intensity, in reducing the risk of severe COVID-19. Where 'adj' follows the variable name, it indicates the two variables were mutually adjusted.   J o u r n a l P r e -p r o o f       Figure S5. Adjusted odds ratios for risk factors (including health-related co-variates potentially on the causal pathway from physical activity to COVID-19 risk) entered into logistic regression analyses for Models 1, 2 and 3 with MVPA as the exposure. Continuous variables (MVPA, age, BMI and deprivation) are standardized, thus odds ratios expressed per standard deviation. The adjusted odds ratios for co-variates were similar across all physical activity outcomes. Open circles = reference category.