Page 103 - 《运动与健康科学》(英文)2024年第2期

TaggedAPTARAEndSleep, physical activity, and type 2 diabetes 225
year), sex (male/female), ethnicity (white/others), season of wearing, recruitment center, Townsend Deprivation Index,
accelerometer wearing (spring/summer/autumn/winter: spring for education level, smoking status, alcohol consumption, healthy
March to May, summer for June to August, autumn for diet score, obesity status, TV watching time, grip strength, and
September to November, and winter for December to February; HbA 1c . Model 3 was further adjusted for hypertension, high
UK Meteorological Office definitions), recruitment center cholesterol levels, depression, and a family history of diabetes.
(England/Wales/Scotland), Townsend Deprivation Index (contin- Sleep duration was included as a covariate in the models for PA
uous, a score representing the deprivation of the participant’s and vice versa. We also used a restricted cubic spline with 4
neighborhood as a reflection of their socioeconomic position), knots selected to smoothen the curve and examine the potential
education level (degree or above/any other qualification/no quali- linear or nonlinear pattern of the estimated association.TaggedAPTARAEnd
fication), smoking status (never/previous/current), alcohol TaggedAPTARAPTo test the joint association between sleep duration and PA,
consumption (not current/less than 3 times a week/3 or more we first examined the interactions between sleep duration and
times a week), healthy diet score (continuous), TV watching time PA in association with the risk of type 2 diabetes on either an
(continuous), obesity status (grouped using body mass index, additive or multiplicative scale. Second, we repeated multivar-
2
2
normal or underweight (<25 kg/m )/overweight (25<30 kg/m )/ iable Cox regression and restricted cubic spline analyses to
2
obese (30 kg/m )), grip strength (continuous), glycated investigate the association between sleep duration and incident
hemoglobin (HbA 1c , continuous), hypertension (yes/no), high type 2 diabetes in each subset of participants with different
cholesterol (yes/no), depression (yes/no), and family history of levels of PA. Third, we subdivided the overall sample into 6
diabetes (yes/no). Age, sex, recruitment center, and Townsend groups according to PA volume and sleep duration. We then
Deprivation Index were known before arrival at the assessment used multivariable Cox models to estimate the joint associa-
center. The Townsend Deprivation Index, a composite measure tion between PA and sleep duration with incident type 2
of deprivation, was based on a participant’s postcode. Informa- diabetes, with the group having higher PA volume and normal
tion on ethnicity, education level, smoking status, alcohol sleep duration, concurrently, used as a reference. Fourth, in the
consumption, healthy diet score, TV watching time, family subgroups stratified by sleep duration associated with an
history of diabetes, and depression was obtained using increased risk of type 2 diabetes, we repeated multivariable
touchscreen questionnaires or verbal interviews. Physical Cox regression and restricted cubic spline analyses to compare
measurements, including height, weight, and grip strength, the risks of incident type 2 diabetes associated with higher PA
were obtained by training nurses. Body mass index was calcu- volume with those associated with lower PA volume.TaggedAPTARAEnd
lated as weight in kilograms divided by the square of height in TaggedAPTARAPSeveral sensitivity analyses were conducted to examine the
meters. HbA 1c levels were measured using high-performance robustness of the current findings, including the exclusion of
liquid chromatography on a Bio-Rad VARIANT II TURBO participants with any missing covariate data, the exclusion
analyzer (Bio-Rad, Des Plaines, IL, USA). The prevalence of of type 2 diabetes events occurring in the first 2 years of
hypertension and high cholesterol levels was obtained from follow-up, the restriction of participants who wore the acceler-
self-reported questionnaires, hospital records, and death regis- ometer for the whole 7 days, and the exclusion of participants
tries. The initial assessments using touchscreen questionnaires with a history of night shift work. To minimize the possible
or verbal interviews were carried out between 2006 and 2010. bias exerted by the coronavirus disease 2019 pandemic, we
Some of the covariates, including education level, smoking repeated the analyses by censoring up to December 31, 2019,
status, alcohol consumption, healthy diet score, obesity status, which was regarded as the commencement of the pandemic.TaggedAPTARAEnd
TV watching time, grip strength, and hypertension, were TaggedAPTARAPInspection of Schoenfeld residual plots for covariates
obtained again during 2012 and 2013 and since 2014. For the confirmed the proportional hazards assumption for all varia-
purposes of this study, these were determined by the time- bles. Statistical analyses were performed using R software
point closest to the accelerometry. Detailed information is Version 4.1.2 (R Development Core Team, Vienna, Austria).
provided in Supplementary Table 2.TaggedAPTARAEnd Statistical significance was defined as p < 0.05 (2-sided).TaggedAPTARAEnd

TaggedAPTARAH13. ResultsTaggedAPTARAEnd
TaggedAPTARAH22.5. Statistical analysesTaggedAPTARAEnd
TaggedAPTARAH23.1. Baseline characteristicsTaggedAPTARAEnd
TaggedAPTARAPDescriptive characteristics were presented as frequencies and
percentages if categorical, and as mean or median (interquartile TaggedAPTARAPTable 1 shows the baseline characteristics of the study
range) if continuous. Missing covariate data were imputed with participants stratified by sleep duration. A total of 88,000
multiple imputations using chained equations to minimize the participants were included (mean age, 62.2 years; 57.2%
potential inferential. The percentage of missing data for each females). The percentage of participants with normal, short,
and long sleep durations was 73.4%, 6.9%, and 19.7%, respec-
covariate was less than 10% (Supplementary Table 3).TaggedAPTARAEnd
TaggedAPTARAPWe conducted 3 multivariable Cox proportional hazards tively. Fewer than half (40.0%) of the participants met recom-
regression models, with the time from accelerometer wearing as mended MVPA levels. Compared with participants with
follow-up time, to estimate the associations of PA (total volume normal sleep duration, those with short sleep duration were
of PA, MVPA, and LPA) and sleep duration with incident type 2 more likely to have socioeconomic deprivation, smoking,
diabetes. Model 1 was adjusted for age and sex. Model 2 was obesity, family history of diabetes, and history of depression,
additionally adjusted for ethnicity, season of accelerometer hypertension, and dyslipidemia, whereas long sleepers tended