2.1 Study population

The UK Biobank is a large-scale and prospective cohort study with over 500 000 participants aged 40–69 years recruited from 2006 to 2010 from across the United Kingdom (www.ukbiobank.ac.uk).²² Participants attended one of 22 assessment centers across England, Scotland, and Wales, where they completed touchscreen and face-to-face questionnaires, underwent physical measurements, provided biological samples and were longitudinally followed up for a wide range of health-related outcomes. All participants gave written consent, and ethical approval was obtained from the Northwest Multicenter Research Ethics Committee (London, UK)

In the current study, we excluded individuals with diabetes at baseline (n = 28 410) and those with missing data on daytime napping (n = 1634), CRP (n = 30 141), or BFP (n = 6980) at baseline, leaving a total of 435 342 participants in the main analysis (Figure 1).

2.2 Exposure assessment

At the baseline assessment, all study participants reported their daytime napping frequency through a touch-screen questionnaire by answering the following question: “Do you have a nap during the day?” with the responses being (a) never or rarely (< once per week), (b) sometimes (1–3 times per week), or (c) usually (≥4 times per week), where every category increase reflected more frequent napping. Prefer not to answer responses were set to missing.

Blood samples were collected from the participants at the initial assessment center visit (2006–2010). CRP (mg/L) was measured by immunoturbidimetric high-sensitivity analysis on a Beckman Coulter AU5800, glucose (mmol/L) was measured by hexokinase analysis on a Beckman Coulter AU5800, and glycosylated hemoglobin (HbA1c; mmol/mol) was measured by high-performance liquid chromatography analysis on a Bio-Rad VARIANT II Turbo. Details on quality control and sample preparation have been described previously.²³ CRP was log-transformed in all analyses due to its skewed distribution.

Body composition measures of all participants were assessed by trained personnel using a standard protocol at enrollment. BFP was estimated using a Tanita BC418MA segmental body composition analyzer (Tanita, Tokyo, Japan) according to the manufacturer's instructions. This device estimates body composition by conducting bioimpedance analysis. Both log (CRP) and BFP levels were treated as continuous variables. In addition, participants were grouped based on quartiles of log (CRP) and BFP to evaluate the effects of these variables on the association between daytime napping and incident T2DM.

2.3 Ascertainment of outcomes

The UK Biobank algorithms²⁴ used to define incident and prevalent outcomes have been reported (Supplementary Table S1 in Appendix S1). In brief, prevalent diabetes was ascertained according to medical history and medication use self-reported and queried by trained health professionals at baseline; incident T2D was ascertained using admissions and diagnoses data of hospital inpatient records obtained through linkage to Hospital Episode Statistics for England, Scottish Morbidity Record data for Scotland, and the Patient Episode Database for Wales, with the International Classification of Diseases, Tenth Revision code E11.

2.4 Statistical analysis

Baseline characteristics of the participants are described as the means (SDs) or n (percentage) according to the frequency of daytime napping (nerve/rarely, sometimes, or usually). The relationship between daytime napping and CRP levels, BFP levels, glucose levels, and HbA1c levels was evaluated using a general linear model. The associations of CRP, BFP, and daytime napping with incident T2D were investigated using Cox proportional hazard models. The proportional hazards assumption was tested using Schoenfeld residuals. Follow-up time was calculated from the baseline date to diagnosis of T2D, death, loss to follow-up, or censoring date (30 June 2020, which was the end of follow-up for the current data release), whichever came first.

Two approaches were used. First, separate associations of daytime napping and quartiles of CRP and BFP with incident T2D were investigated. CRP and BFP were also treated as continuous variables, and hazard ratios (HRs) per one SD increase in inflammation and adiposity were estimated. Second, joint associations between daytime napping, CRP, and incident T2D and between daytime napping, BFP, and incident T2D were evaluated. Daytime napping frequencies and quartiles for CRP and BFP were derived, and HRs were calculated, with the referent category comprising individuals who were in both the lowest frequency group of daytime napping (never/rarely) and the highest quartile of CRP or BFP. To investigate whether adiposity and/or inflammation moderated the association between daytime napping and the risk of incident T2D, we assessed the relative excess risk due to interaction (RERI) as an index of additive interaction. In our study, RERIs were defined for dichotomized exposures^{25, 26} and were calculated by comparing frequent nappers to nonnappers using the formula $\left({\mathit{HR}}_{\text{Frequent nappers},\text{highest}\mathrm{BFP}/\mathrm{CRP}\text{quartile}}-{\mathit{HR}}_{\text{Frequent nappers},\text{lowest}\mathrm{BFP}/\mathrm{CRP}\text{quartile}}-{\mathit{HR}}_{\text{Nonnappers},\text{highest}\mathrm{BFP}/\mathrm{CRP}\text{quartile}}\right)+1$. The confidence intervals (CIs) for RERI were estimated using the standard delta method.²⁷ Multiplicative interaction was also tested by including a product term in the models.

Several potential confounders were adjusted in the models described: age (continuous), sex (male/female), race (white and others), UK Biobank assessment center, Townsend deprivation index (continuous), average total annual household income (<£18 000, £18 000–30 999, £31 000–51 999, £52 000–100 000, ≥£100 000, and “do not know” or missing), BMI (<18.5, 18.5–24.9 or 25–29.9, ≥30 kg/m²), smoking status (current, former, or never), alcohol consumption (current, former, or never), physical activity (metabolic equivalent task [MET]-minutes per week, continuous), healthy diet (yes/no), family history of diabetes (yes/no), antihypertensive medication use (yes/no), cholesterol medication use (yes/no), aspirin use (yes/no), and nonaspirin nonsteroidal anti-inflammatory drug (NSAID) use (yes/no). A healthy diet was defined as an adequate intake of at least four of seven commonly eaten food groups recommended as dietary priorities for cardiometabolic health²⁸: fruit intake ≥3 servings/day; vegetable intake ≥3 servings/day; fish intake ≥2 servings/week; processed meat intake ≤1 servings/week; unprocessed red meat intake ≤1.5 servings/week; whole grain intake ≥3 servings/day; and refined grain intake ≤1.5 servings/day. Missing data were imputed with mean values for continuous variables or coded as a missing indicator category for categorical variables.

We performed a stratified analysis to evaluate potential modification effects according to the following factors: sex (male or female), race (White, Asian, Black, or mixed/other), age (<55 or ≥ 55 years), BMI (<30 or ≥ 30 kg/m²), physical activity (<600 MET min/week or ≥ 600 MET min/week), healthy diet (yes or no), sleep duration (short [<6 h/night], normal [6–9 h/night], long [>9 h/night]), smoking status (never, past, current), alcohol consumption status (never, past, current), Townsend deprivation index (<median or ≥ median), family history of diabetes (yes or no), aspirin use (yes or no), and nonaspirin NSAID use (yes or no). Sleep duration was based on the following standardized question: “Approximately how many hours do you sleep every 24 h?”, for which responses were given in hours (https://bbams.ndph.ox.ac.uk). Sleep duration information was considered to be missing if participants failed to answer or had a sleep duration <4 or > 18 h to reduce the possibility of an implausible sleep duration. Based on the classification proposed by the National Sleep Foundation,^{29, 30} sleep duration was categorized as short (<6 h/night), normal (6–9 h/night), and long (>9 h/night). The interaction test between daytime napping (usually versus never/rarely) and each category was performed by adding interaction terms to the Cox models. Several sensitivity analyses were also performed to confirm the robustness of our results¹: excluding participants with glucose levels ≥7.0 mmol/L (n = 6242) or HbA1c levels ≥48 mmol/mol (6.5%, n = 2933) to address the issues of potential undiagnosed diabetes at baseline²; excluding participants with T2D incidence occurring in the first 2 years to minimize the influence of reverse causation (n = 18 254)³; excluding participants who regularly took an anti-inflammatory drug (aspirin or nonaspirin NSAID) to minimize the influence of other regular inflammatory drug use (n = 116 095); and⁴ adjusting for glucose levels, HbA1c levels, CRP levels, or sleep duration (hours) in the multivariable models.

All analyses were conducted using Stata (version 15; StataCorp, College Station, Texas). All statistical tests were two sided, and p < .05 was considered statistically significant.

3.1 Baseline characteristics of participants by daytime napping frequency

Table 1 shows the baseline characteristics of the study participants stratified by the frequency of daytime napping. Of the 435 342 participants (mean [SD] age, 56.3 (8.1) years), 240 665 (55.3%) were women. Overall, 249 813 (57.4%) participants reported never or rarely napping during the day (nonnappers), 163 973 (37.7%) reported napping sometimes (occasional nappers), and 21 556 (5.0%) reported napping often (habitual nappers) at baseline.

After adjustment for a set of covariates, daytime napping frequencies showed a significant positive association with glucose, HbA1c, BFP and CRP levels at baseline (Supplementary Table S2 in Appendix S1).

3.2 Association between daytime napping and risk of incident T2D

During a median of 9.2 years of follow-up (4 013 849 person-years), 17 592 incident cases of T2D were identified. Kaplan–Meier curves of incident T2D demonstrated a significant, incremental increase in rates of incident T2D associated with increasing frequency of daytime napping (log-rank p < .001, Supplementary Figure S1 in Appendix S1). Higher daytime napping frequency was significantly associated with an increased risk of incident T2D in the age-, sex-, and multivariate-adjusted models (Table 2). In the age- and sex-adjusted model, the HRs associated with daytime napping were 1.62 (95% CI: 1.57–1.67) and 2.15 (95% CI: 2.03–2.27) among occasional nappers and habitual nappers, respectively (p < .001). After further adjustment, the HRs were 1.28 (95% CI: 1.24–1.32) and 1.49 (95% CI: 1.41–1.57) among occasional nappers and habitual nappers, respectively.

Similar associations of daytime napping with the risk of incident T2D were observed in several sensitivity analyses, including further removing participants with a high level of glucose (≥7.0 mmol/L), participants with a high level of HbA1c (≥48 mmol/mol [6.5%]), participants who developed T2D within 2 years of follow-up or participants who took any anti-inflammatory drugs or after adjusting for glucose levels, fasting time, HbA1c levels, or CRP levels (Supplementary Tables S3–S7 in Appendix S1).

We also conducted stratified analyses according to potential T2D risk factors to evaluate whether these factors modified the association between daytime napping and the risk of T2D (Figure 2). We found that the associations between daytime napping and the risk of incident T2D were stronger among men (p for interaction <.001), White and Black individuals (p for interaction = .011), younger participants (<55 years, p for interaction = .006) and obese participants (≥ 30 kg/m², p for interaction = .032).

3.3 Interaction between daytime napping and CRP levels and BFP levels on the risk of T2D

We first evaluated the separate associations between inflammation, adiposity, and risk of incident T2D. Table 3 shows the HRs for the risk of T2D for quartiles of CRP and BFP levels. Higher levels of CRP and BFP were associated with an increased risk of incident T2D (all p for trend <.001), a one-SD increase in CRP levels (1.1 mg/L) was associated with a 40% increase in the risk of incident T2D (95% CI: 37%–42%), and a one-SD increase in BFP (8.5%) was associated with a 69% increase in the risk of incident T2D (95% CI: 64%–75%).

Next, we examined the additive and multiplicative interactions between daytime napping and CRP levels and between daytime napping and BFP on the risk of T2D. Figure 3 shows the associations between daytime napping and risk of T2D, stratified by CRP and BFP quartiles, with those in both nonnappers and the lowest CRP/BFP quartile as the reference group. We found an additive and multiplicative effect for daytime napping and adiposity on the risk of incident T2D (RERI = 0.490, 95% CI: 0.307–0.673; p for multiplicative interaction = .009; Figure 3A). Within each daytime napping frequency stratum, there was an increase in the strength of the association with increasing BFP, and the highest risk of incident T2D was observed in habitual nappers in the highest quartile of BFP (HR = 4.45, 95% CI: 3.92–5.06). A similar pattern was observed for CRP, and the highest risk of incident T2D was observed in habitual nappers in the highest quartile of CRP (HR = 3.66, 95% CI: 3.30–4.05). Although the multiplicative interaction between daytime napping and CRP was not statistically significant (p for multiplicative interaction = .089), a similar additive interaction (RERI = 0.266, 95% CI: 0.094–0.439) was observed (Figure 3B). These results remained materially unchanged after performing several sensitivity analyses (Supplementary Figures S2–S5 in Appendix S1).