This study was approved by the National Ethical Review Agency, Stockholm, Sweden (DNR 2022-05751-01). Informed consent was obtained from all participants on inclusion into the cohorts. The Helsinki Declaration was followed, as was the STROBE statement [12].

Due to the sensitive nature of the collected data, data cannot be publicly shared. However, researchers with a valid ethical approval may contact the steering committee of the Swedish Infrastructure for Medical Population-based Life-course and Environmental Research (SIMPLER) [13] to access data.

Participants were drawn from the population-based Swedish Mammography Cohort (SMC) and Cohort of Swedish Men (COSM). The SMC was started in 1987 by inviting all women born 1914–1948 living in Uppsala county and all women born 1917–1948 living in Västmanland county to answer self-administered lifestyle questionnaires. This first questionnaire had a participation rate of 74%. A second more detailed questionnaire was administered in 1997 and was answered by 70% (n = 39,277) of the original cohort. COSM was started in 1997 by sending a questionnaire similar to the one used in SMC to all men born 1918–1952 living in Västmanland and Örebro counties. The response rate was 49% (n = 48,850). The SMC and COSM are representative of the Swedish population [14, 15]. Participants have been linked to the National Patient Register, the Swedish Intensive Care Registry (SIR) and the Cause of Death Register, using the unique Personal Identification Number issued to all residents in Sweden.

All participants having entered an incorrect personal identification number, that died before January 1st, 1998, or were diagnosed with cancer (except nonmelanoma skin cancer) were excluded, leaving 38,984 women aged 48 to 83 years and 45,906 men aged 45 to 79 at baseline in the final cohort (Additional file 1).

Physical activity was ascertained through self-administered lifestyle questionnaires administered in 1997. Questions inquired about time spent on leisure exercise (< 1 h/week, 1 h/week, 2–3 h/week, 4–5 h/week or > 5 h/week), time spent walking or bicycling (hardly ever, < 20 min/day, 20–40 min/day, 40–60 min/day, 60–90 min/day or > 90 min/day, the highest two levels were collapsed for analysis), physical activity at work (mostly sitting, sitting half of the time, most standing, mostly walking with little carrying, mostly walking with considerable carrying, heavy manual labor, levels were collapsed into: sitting, sitting half of the time, standing and walking, lifting), reading or watching television (< 1 h/day, 1–2 h/day, 3–4 h/day, 5–6 h/day, > 6 h/day, the highest two levels were combined) and household work (< 1 h/day, 1–2 h/day, 3–4 h/day, 5–6 h/day, 7–8 h/day, > 8 h/day, the highest three levels were combined). The measures of physical activity have been validated against 7-day activity records with correlations of circa 0.4 [16, 17] and against accelerometer data with correlation of 0.38 [17], suggesting acceptable validity.

Covariates were selected using a directed acyclic graph (DAG)-based method [18]. All covariates were collected from the questionnaires, except comorbidities which were collected from the National Swedish Patient Register, which has near-complete coverage and high validity [19].

Incident cases of infection or sepsis were collected from the Swedish National Patient Register until 31th December 2021. First occurrence of infection or sepsis was grouped and defined according to the International Statistical Classification of Diseases and Related Health Problems (ICD-10) diagnosis codes other sepsis (A40, A41, A32, A48, A49, B95, B96 D65, T802, R65.1, R57.2), abdominal infection (K35, K57.0, K57.2, K57.4, K57.8, K63.0, K63.1, K65, K80.0, K80.1, K80.3, K80.4, K81, K83.0, K85), urogenital infection (N10, N12, N13.6, N39.0, N70), soft tissue infection (M00, M01, M72.6, A46), pneumonia (J13, J14, J15, J16, J18, J85, J86), endocarditis (I33, I39), tuberculosis (A15, A16, A17, A18, A19) and central nervous system infection (A39, G00, G01, G02, G03.9, G05.0, G06, G07.9), see Additional file 2. ICU admission, physiological derangement at admission (classified using the Simplified Acute Physiology Score, SAPS3 [20]) and organ-replacement therapy were acquired from the Swedish Intensive Care Registry [21]. Death due to infection or sepsis was drawn from the Cause of Death Register, using the same ICD-10 codes.

Using Cox proportional hazards regression with attained age as timescale, hazard ratio (HR) and 95% confidence intervals (CI) were calculated for first occurrence of infection or sepsis, ICU admission, and death due to infection or sepsis. Participants contributed person-time at risk from baseline (January 1st 1998) until infection or sepsis, death or end of follow-up (December 31st, 2021), whichever occurred first. Exposures of interest were exercise; walking; physical activity at work; reading or watching television and household physical activity. The lowest level was used as reference. Estimates were first calculated in a crude model adjusted only for age (as timescale) and sex, and subsequently in a fully adjusted model, including age (as timescale), sex (man/woman), marital status (cohabiting/living alone), education (≤ 9 years/9–12 years/ > 12 years/other, such as vocational), smoking status (current/former/never), alcohol consumption (g/day, continuous) and Charlson’s weighted comorbidity index [22] (continuous). To test whether potential associations are linear, adjusted models using exposures as continuous variables and as restricted cubic splines with the median as reference and knots placed at the 10th and 90th percentile [23] were compared using likelihood ratio (LR) tests. Log–log plots were used to test the proportional hazards assumption. The population-attributable fraction (PAF) for contracting and dying in infection or sepsis, respectively, was calculated as:

where Pc is the proportion of exposure among cases and HR is the adjusted hazard ratio [24]. For this analysis, the lowest level of exercise (to approximate inactivity) was chosen as exposure, and all other levels of exercise were collapsed and treated as unexposed.

The main analysis was performed using complete-case analysis. The proportion of missing was 13.8% for physical activity at work, 11.0% for exercise and 10.0% for household work. The proportion of missing was < 10% for all other variables. For the main analysis of exercise, 64,850 participants had all information necessary for inclusion (see Additional file 1: Cohort flowchart).

Sensitivity analyses were performed to test the robustness of results. Two sensitivity analyses were performed to attempt to further adjust for confounding by previous health (i.e., if previous illness or poor health leads to an individual being less physically active): first the main analysis was further adjusted for self-rated health, and second, an analysis with a washout period of three years was performed, where participants became at risk January 1, 2001. To account for the possibility of nonlinear effects in confounding by comorbidity, a sensitivity analysis was performed where Charlson’s weighted comorbidity index was included as a categorical variable (0/1/ ≥ 2). The main analysis of death included infection and sepsis both as underlying and contributing causes of death in case of multiple causes of death. A sensitivity analysis was performed where only the underlying cause of death was considered. The main analysis was repeated, but with December 31st, 2019, as the end of follow-up, to exclude Covid-19 pandemic years. To assess whether observed associations were mediated by body mass index (BMI, weight in kilograms divided by length in meters squared), a sensitivity analysis was performed adjusting for BMI divided into categories (< 20/ ≥ 20 to < 25/ ≥ 25 to < 30/ ≥ 30). Analyses were performed for each diagnosis group separately. To check whether associations were similar in smokers and nonsmokers, the main analysis was repeated stratified for smoking status.

Descriptive statistics are presented as median (interquartile range, IQR). All analyses were performed in Stata 15.1 (Stata Corp., College station, Texas, USA).