Modifiable lifestyle factors and the risk of post-COVID-19 multisystem sequelae, hospitalization, and death – Nature.com
							July 29, 2024
							Data sources and study cohorts    
    UK Biobank is a large-scale population-based prospective cohort    study with deep phenotyping and genomic data, as detailed    elsewhere51. Briefly,    between 2006 and 2010, over 500,000 individuals aged 4069    years were recruited from 22 assessment centers across the    United Kingdom at baseline, with collection of    socio-demographic, lifestyle and health-related factors, a    range of physical measures, and blood    samples51. Follow-up    information is obtained by linking health and medical records,    including national primary and secondary care, disease and    mortality registries52, with validated    reliability, accuracy and completeness53. To identify    cases of SARS-CoV-2 infection, polymerase chain reaction    (PCR)-based test results were obtained by linking all    participants to the Public Health Englands Second Generation    Surveillance System, with dates of specimen collection and    healthcare settings of testing54. Outbreak    dynamics were validated to be broadly similar between UK    Biobank participants and the general population of    England54.  
    In this study, we included participants who were alive by March    1, 2020 and had a positive SARS-CoV-2 PCR test result between    March 1, 2020 (date of the first recorded case in the UK    Biobank), and March 1, 2022, with the date of first infection    considered as index date (T0). For    those diagnosed with COVID-19 in hospital, we defined    T0 as the date of hospital admission    minus a random number of 7 days. The major prevalent variants    during the study period included wildtype, Alpha (B.1.1.7),    Delta (B.1.617.2), and Omicron (B.1.1.529 BA.1). The calendar    periods of dominant variants in the UK were based on pandemic    data from the Office for National Statistics26. Participants    with missing data on study exposures at baseline were excluded.    We addressed missing data on covariates using the following    approaches: (1) participants with missing values in age and sex    (<0.1%) were excluded. (2) participants with missing    values in ethnicity were classified as other ethnic groups.    (3) participants with missing values in education level (0.9%)    were classified as category I, which includes none of the    above and prefer not to answer. (4) missing values in IMD    (13.8%) were imputed with the mean value of the entire UK    Biobank cohort. All participants included in this study    provided written informed consent at recruitment. This study    followed the Strengthening the Reporting of Observational    Studies in Epidemiology (STROBE) reporting guidelines and    received ethical approval from the UKBB ethics advisory    committee. Study design, cohort construction, and timeline are    provided in Supplementary Fig.1. All participants    provided written informed consent at the UK Biobank cohort    recruitment. This study received ethical approval from UK    Biobank Ethics Advisory Committee (EAC) and was performed under    the application of 65397.  
    Ten prespecified potentially modifiable lifestyle factors were    assessed, including smoking, alcohol consumption, body mass    index (BMI), physical activity, sedentary time, sleep duration,    intake of fruit and vegetable, intake of oily fish, intake of    red meat, and intake of processed meat. Selection and    categorization of lifestyle factors was based on literature    review, previous knowledge, and UK national health service    guidelines55,56. Multiple    lifestyle factors were measured by validated questionnaire for    all participants at baseline recruitment. Detailed definitions    on measurement and classification of lifestyle factors are    provided in Supplementary Table1. Briefly, healthy    lifestyle components including past or never smoker, moderate    alcohol intake (4 times week), BMI<30kg/m2,    at least 150min of moderate or 75min of vigorous physical    activity per week, less sedentary time (<4h per day),    healthy sleep duration (79h per day), adequate intake of    fruit and vegetables (400g/day), adequate oily fish intake    (1 portion/week), moderate intake of red meat (4 portion    week) and processed meat (4 portion week) were defined, in    accordance with previous evidence or UK national health service    guidelines55,56.  
    A binary variable was created for each of the 10 factors, with    1 point assigned for those meeting the healthy criteria and 0    otherwise. A composite lifestyle score was then calculated for    each participant by summing the total number of healthy    lifestyle factors, ranging from 0 to 10. Based on the composite    score, participants were classified into three lifestyle    categories: unfavorable (05), intermediate (67), and    favorable (810). The lifestyle score was also used as a    continuous variable of number of healthy lifestyle factors.    Similar methods of defining lifestyle score have been used in    the same UK Biobank cohort57 as well as    external cohorts16,28. Distributions    of lifestyle score and categories are provided in Supplementary    Table2.  
    The median [IQR] duration between baseline assessment of    lifestyle factors and the date of infection was 12.5    [11.813.3] years. Part of participants took part in up to two    further touchscreen interviews with lifestyle and    health-related factors similarly measured. There were generally    stable responses to lifestyle factors between baseline    assessment and the latest repeat assessment (median time    difference from baseline, 8 years) as shown in Supplementary    Fig.2. 34.9% of    participants with an unfavorable lifestyle, 48.6% with an    intermediate lifestyle, and 73.7% with a favorable lifestyle at    baseline remained in the same corresponding lifestyle category    at the latest repeat assessment following a median of 8 years.    Overall, the proportion of stable lifestyle categories is    60.6%.  
    The outcomes after COVID-19 were prespecified, including a set    of multisystem sequelae, death, and hospital admission    following the SARS-CoV-2 infection. The multisystem sequelae    were selected and defined based on previous evidence of the    long COVID, including 75 systemic diseases or symptoms in 10    organ systems: cardiovascular46, coagulation    and hematologic46, metabolic and    endocrine44,    gastrointestinal48,    kidney43, mental    health45,    musculoskeletal47,    neurologic47, and    respiratory disorders10,13,14, and general    symptoms of fatigue and malaise3,4,42,49. Detailed    definitions of multisystem sequelae are listed in the    Supplementary Table3. Outcomes were    identified as follows: individual sequela from the hospital    inpatient ICD-10 (International Classification of Diseases    10th Revision) diagnosis codes, deaths from the records of    national death registry, and hospital admission from hospital    inpatient data from the Hospital Episode Statistics. Incident    outcomes were assessed in participants with no history of the    related outcome within one year before the date of the first    infection.  
    As SARS-CoV-2 infection has been associated with both    multisystem manifestations during its acute phase and with    sequelae during its post-acute phase7,49, we conducted    analyses stratified by phase of infection. We reported risk of    each outcome during the acute phase    (T0 to    T0+30d), post-acute phase    (T0+30d to    T0+210d), and overall period    following infection (T0 to    T0+210d) to reflect the full    spectrum of post-COVID conditions. The end of follow-up for the    overall cohort was September 30, 2022, with the maximum    follow-up period censored to 210 days.  
    We prespecified a list of covariates for adjustment or    stratification based on literature review and prior knowledge:    socio-demographic characteristics including age, sex, education    level (mapped to the international standard for classification    of education), index of multiple deprivation (IMD, a summary    measure of crime, education, employment, health, housing,    income, and living environment)58, and race and    ethnicity; and infection related factors including healthcare    settings of the testing (community/outpatient vs inpatient    setting as proxy of severity of infection), COVID-19    vaccination status, and SARS-CoV-2 variants.  
    Baseline characteristics of the overall cohort of participants    with SARS-CoV-2 infection and by composite healthy lifestyle    categories were reported as mean and standard deviation or    frequency and percentage, when appropriate. Multivariable cox    proportional hazard (PH) model was used to assess the    association between composite healthy lifestyle and risk of    multisystem sequelae (composite or by organ systems), death,    and hospital admission, with adjustment for age, sex,    ethnicity, education level, and IMD. PH assumption across    lifestyle categories was tested by Schoenfeld residuals with no    violations observed for outcomes. Hazard ratio (HR) and    absolute risk reduction (ARR, difference in incidence rate    between lifestyle groups per 100 persons during the    corresponding follow-up period) were estimated from the Cox    model. We also assessed the association between individual    lifestyle factor instead of composite categories (each    component as a categorical variable with or without mutual    adjustment for others, or the number of factors as continuous    variables) and risk of outcomes.  
    We conducted causal mediation analysis59,60 to quantify the    extent to which the habitual healthy lifestyle may affect    COVID-19 sequelae through the potential pathway of relevant    pre-infection medical conditions (mediator), with the    proportion of direct and indirect effects estimated by    quasi-Bayesian Monte Carlo methods with 1000 simulations for    each. Detailed modeling procedures and a directed acyclic graph    are provided in Supplementary Methods.  
    We examined the association between composite healthy lifestyle    and the overall risk of multisystem sequelae in prespecified    clinical subgroups by demographic and infection-related    factors. The demographic factors included age (65 and >65    years), sex (male and female), and ethnicity (White and other    ethnic groups). As the risk profile of COVID sequelae was    related to vaccination and severity of infection, and may    change with the evolving pandemic, infection-related factors    including vaccine status (no or one-dose partial vaccination    and two-dose full vaccination), test setting (inpatient and    outpatient or community), dominant variants during the study    period (wildtype, Alpha, Delta, and Omicron BA.1) were    assessed. Multiplicative interactions between the composite    healthy lifestyle and the stratification variables were tested,    with P-value reported.  
    We conducted multiple sensitivity analyses to assess the    robustness of primary findings. First, to reflect the    multisystem and potentially comorbid nature of COVID sequelae,    accounting for both the number of sequelae by an individual and    the relative health impact of each sequela. Weights based on    Global Burden of Disease study data and methodologies for    general diseases and long COVID were assigned to each sequela    (Supplementary Table1)61,62. The weighted    score was calculated for each participant by summing the    weights of all incident sequelae during the follow-up period.    Zero inflated Poisson regression was then used to calculate    relative risk (RR), with follow-up time set as the offset of    the model and adjustment for covariates. Second, to further    account for potential reverse causality and more accurately    define incident cases, extending the washout period for    outcomes from one year to two years. Third, defining events of    post-acute sequelae 90 days after infection (follow-up period    T0+90d to    T0+210d), instead of 30 days in the    main analyses. The adjustment was made as there is no uniform    definition for long COVID, which is currently described as    conditions occurring 3090 days after infection in existing    guidelines27. Fourth,    restricting the identification of outcomes to the first three    ICD diagnoses, which are the main causes for each hospital    admission. Fifth, reconstructing a composite lifestyle index    without BMI and assessed its association with outcomes.    Finally, we conducted quantitative sensitivity analysis to    adjust for changes in lifestyle factors over time since the    baseline assessment. We used odds ratios to quantify    associations and assumed a sensitivity and specificity of 90%    for each lifestyle component (Supplementary Methods).  
    As a healthy lifestyle is associated with a lower risk of    chronic diseases and mortality among the general population    predated pandemic, we conduct exploratory analysis to compare    the effects of healthy lifestyle on adverse outcomes following    COVID-19 with the effects among participants without infection.    A random index date was assigned to the participants without    infection based on the distribution of    T0 among those with infection, and we    repeated the main analyses with the maximum follow-up period    censored to 210 days.  
    Statistical significance was determined by a 95% confidence    interval (CI) that excluded 1 for ratios and 0 for rate    differences. All analyses and data visualizations were    conducted using R statistical software (version 4.2.2).  
    Further information on research design is available in    theNature Portfolio    Reporting Summary linked to this article.  
Visit link:
Modifiable lifestyle factors and the risk of post-COVID-19 multisystem sequelae, hospitalization, and death - Nature.com