Study population

The detailed study designs, survey methods, and population characteristics of the UK Biobank have been described previously [13]. Briefly, the UK Biobank is a national, prospective cohort study that enrolled over 500,000 participants aged 38–73 years from 2006 to 2010, across 22 assessment centers in the UK. Participants provided extensive data on sociodemographics, lifestyle, dietary habits, and health status by completing touchscreen questionnaires, oral interviews, and a series of anthropometric and physiological measures during the assessment visit. Blood samples collected were genotyped. The study received approval from the North West Research Ethics Committee (06/MRE08/65), and all participants provided written informed consent.

In this study, we excluded participants who had a diagnosis of RA (n = 3266) and those with missing values for five sleep behaviors (n = 89,431), PA (n = 29,674), genetic risk score (GRS) (n = 9178), and other covariate information (n = 7651) at baseline, leaving a total of 363,211 participants for the primary analysis. The flowchart describing the exclusion criteria was presented in Additional file 1: Fig. S1.

Assessment of sleep behaviors

Information on sleep characteristics was collected via the touchscreen questionnaire at baseline visit between 2006 and 2010. Questions regarding five sleep characteristics were derived from standardized questionnaires [14]. Sleep duration was recorded by asking, “About how many hours sleep do you get in every 24 h? (please include naps).” Chronotype was assessed by asking, “do you consider yourself to be?,” with responses options of “definitely a ‘morning’ type,” “a ‘morning’ more than ‘evening’ type,” “an ‘evening’ more than ‘morning’ type,” or “definitely an ‘evening’ type.” Insomnia symptoms were assessed based on the question: “do you have trouble falling asleep at night or do you wake up in the middle of the night?,” with responses of “never/rarely,” “sometimes,” or “usually.” Snoring information was recorded based on the question, “does your partner or a close relative or friend complain about your snoring?,” with responses of “yes” or “no.” To assess subjective daytime sleepiness symptoms, participants were asked: “how likely are you to doze off or fall asleep during the daytime when you don’t mean to?” with responses of “never” or “rarely,” “sometimes,” “often,” or “always.”

Definition of healthy sleep score and sleep pattern

A healthy sleep score (range 0–5) was constructed based on five identified sleep components: sleep duration, chronotype, insomnia, snoring, and excessive daytime sleepiness. Healthy sleep behavior is defined as sleep 7–8 h per day, early chronotype (“morning” or “morning than evening”), reported never or rarely insomnia symptoms, no snoring, and no frequent daytime sleepiness (“never/rarely” or “sometimes”). For each sleep characteristic, a participant was scored 1 if she or he exhibited low-risk sleep behavior, and 0 otherwise. The sum of all five individual sleep behaviors was regarded as the healthy sleep score. A higher score indicates a healthier sleep pattern. We categorized the healthy sleep score into three categories: healthy sleep pattern (healthy sleep score ≥ 4), intermediate sleep pattern (2 ≤ healthy sleep score ≤ 3), and poor sleep pattern (healthy sleep score ≤ 1).

In the sensitivity analysis, we further generated a weighted sleep score based on the five sleep behaviors using the following equation: weighted sleep score = (β1 × sleep behavior 1 + β2 × sleep behavior 2 + … + β5 × sleep behavior 5) × (5/sum of the β coefficients). The weighted score takes into account the relative risk of each sleep behavior and calculates a weighted average of the five sleep behaviors, resulting in a weighted sleep score ranging from 0 to 5.

Assessment of physical activity

PA information was collected using the well-validated long International Physical Activity Questionnaire (IPAQ), which includes the frequency and duration of three levels of activities: walking, moderate-intensity, and vigorous activity. Participants were asked how many days per week they engaged in each category of activity and the number of minutes they spent each day on these activities. The answer “unable to walk” was coded as 0, and both “unwilling to answer” and “don't know” were set as missing. IPAQ has demonstrated excellent reliability and acceptable validity [15], and its robust validity among older adults in UK has also been verified [16]. Metabolic equivalents (METs) quantify self-reported physical activity, where each MET represents the energy expended sitting quietly for 1 h. The MET value reflects the ratio of energy expended per kilogram of body weight per hour to the energy expended while sitting quietly. The number of MET minutes per day was calculated as follows: the number of minutes per day involved in each activity level was multiplied by the MET score for the corresponding activity level. Weekly MET minutes were then obtained based on the number of MET minutes per day. The METs for walking, moderate, and vigorous activity levels were summed as total METs. According to the IPAQ guidelines, the MET for walking is 3.3, the MET for moderate PA is 4.0, and the MET for vigorous PA is 8.0, and 0 for PA less than 10 min per day in each category [17]. Participants were divided into three groups based on the standard scoring criteria of IPAQ: low (< 600 MET-mins/week), moderate (≥ 600 and < 3000 MET-mins/week), and high (≥ 3000 MET-mins/week) [18, 19]. PA groups can also be grouped according to quintiles of total METs.

Assessment of genetic risks of RA

Detailed information about genotyping process and quality control in the UK Biobank study has been described elsewhere [20]. We obtained the released standard GRS for RA from UK Biobank [21]. GRS algorithms were built from trait-specific meta-analyses using a Bayesian approach, which combined data across multiple ancestries and related traits when appropriate. Unlike general GRS, which were built based on reported top SNPs, per-individual GRS value was calculated as the genome-wide sum of the per-variant posterior effect size multiplied by allele dosage. In this study, GRS was classified as low (lowest quintile), intermediate (2–4 quintiles), and high (highest quintile) genetic risk.

Ascertainment of outcomes

In the UK Biobank, data on incident RA were identified through linkage with NHS hospital inpatient data from hospital event statistics in England, the Scottish Morbidity Records, and the Patient Episode Database for Wales. Diagnostic results were defined using the ICD-10 (International Classification of Diseases, 10th Revision) coding system. Participants with primary or secondary RA coded M05 and M06 were defined as endpoint events. The follow-up time was from recruitment until the data of first diagnosis, loss to follow-up, death or censoring data, whichever occurred first.

Statistical analyses

A multivariate-adjusted Cox proportional hazard model was used to compute hazard ratio (HR) and 95% confidence intervals (CIs) for the associations of sleep patterns and PA with incident RA risk. Models were adjusted for sex (male, female), age at recruitment (continuous, years), ethnicity (White, and other), the Townsend Deprivation Index (continuous), smoking status (current, past, never), alcohol consumption (current, past, never), body mass index (BMI, < 18.5, 18.5–25, 25–30, ≥ 30, kg/m2), waist hip ratio (WHR, low: < 0.91 for males and < 0.79 for females; medium: 0.91–0.96 for males and 0.79–0.85 for females; high: ≥ 0.96 for males and ≥ 0.85 for females), PA (MET-min/week), history of diabetes (yes, no), history of cancer (yes, no), history of bone fracture (yes, no), vitamin D supplementation (yes, no) and genetic risk score. The proportional hazards assumption model was tested using the Schoenfeld residuals method and no violation was detected. Restricted cubic spline (RCS) regression model was used to analyze the nonlinear relationship between PA and the occurrence of RA. Further stratified analyses were performed to assess the association of sleep patterns with RA incidence across different PA levels or GRS groups as well as the association of PA levels with RA incidence across different sleep patterns or GRS groups. In addition, we evaluated the combined effects of sleep patterns, PA, and GRS on the risk of new-onset RA.

To investigate the potential impact of the relationship between sleep patterns, PA, and GRS on RA, we used multiplicative and additive interaction to evaluate the interaction between each of the two exposures in the Cox proportional hazard models. In terms of multiplicative interaction, we derived the HR 95% CI and Pinteraction by using a likelihood ratio test to compare Cox models with and without a product term (exposure 1 × exposure 2), while the additive interaction can be calculated by relative excess risk due to interaction (RERI) and attributable proportion due to interaction (AP). The 95% CI for both the RERI and the AP was calculated by drawing 5000 bootstrap from the estimation dataset [22, 23]. If the 95% CI of RERI and AP contains 0, there is no additive interaction.

In order to test the reliability and robustness of the primary associations, we performed a series of sensitivity analyses: (1) participants who were diagnosed with RA within the first 1, 2, and 3 years of follow-up were excluded respectively; (2) participants with shift work; (3) participants who were non-White; (4) the missing values of covariates were filled by multiple inference through chained equations (the R package of “mice” had less than 3% missing values for all covariates) [24]; (5) in the analysis of the relationship between PA and RA, air pollution data (NO2, NOX, PM2.5, PM10, PM2.5–10) related to PA were additionally adjusted; (6) re-calculating the relationship between weighted sleep score and risk of RA. We also conducted a stratified subgroup analysis to evaluate the modification effect of covariates on the association of sleep pattern with RA risk.

All statistical analyses were performed using R software (version 4.2.2). A two-sided test p < 0.05 was defined as statistically significant.