Long-Term Physical Activity Participation and Subsequent Incident Type 2 Diabetes Mellitus: A Population-Based Cohort Study

Background Uncertainty remains concerning association between long-term physical activity and incident type 2 diabetes mellitus (DM). We intended to evaluate physical activity participation over a 6-year span and assess association with subsequent 10-year incident DM risk, as well as examine mediation role by obesity. Methods A total of 9757 community-dwelling adults aged ≥ 50 years in England were included in the population-based cohort. Physical activity participation, including trajectories and cumulative participation were assessed using weighted Z score over a 6-year span from wave 1 (2002–2003) to wave 4 (2008–2009). Incident DM recorded over a 10-year span from wave 4 (2008–2009) to wave 9 (2018–2019) was outcome. Results 5 distinct activity trajectories were identified, including persistently low (N=3037, incident DM=282), initially low then improving (1868, 90), initially high then declining (325, 20), persistently moderate (2489, 170), and persistently high (2038, 108). Compared with persistently low, participants of initially low then improving, persistently moderate and high were associated with lower incident DM risk, with multivariable-adjusted hazard ratios (HR) of 0.41 (95% confidence interval [CI]: 0.32 to 0.53, P<0.001), 0.70 (95% CI: 0.56 to 0.89, P=0.004) and 0.49 (95% CI: 0.37 to 0.65, P <0.001), respectively. Elevated cumulative activity was also associated with lower DM risk, with each quintile increment in cumulative weighted Z score corresponding to HR of 0.76 (95% CI: 0.71 to 0.82, P <0.001). Mediation analysis found that body mass index, waist circumference and change in body mass index mediate 10% (P <0.001), 17% (P <0.001) and 9% (P <0.001) of the observed association between activity and incident DM, respectively. Conclusions For middle aged and older adults, both gradually improved and persistently active participation in physical activity were associated with subsequent lower risk of incident DM, with obesity playing a potential mediator. Strategies focusing on improving and maintaining active participation in physical activity might be beneficial from DM prevention perspective.


BMI and waist circumference measurement
All participants were eligible to have their height and weight measured. If the nurse thought the measurement was likely to be more than 2 cm (3/4 inch) from the true figure for height or more than 1 kg (2 lbs.) from the true figure for weight, it was considered unreliable and they were asked to code it as such. Using confirmed reliable measured standing height and weight, BMI was calculated by dividing weight in kilograms by height in meters squared. For waist circumference, typically two measurements were conducted for each visit, unless the second measurement differed from the first by 3 cm or more, then the nurse was prompted to either amend one of the previous responses if a mistake had been made entering a measurement, or to take a third measurement. If the nurse believed that the measurements they took were 0.5 cm more or less than the true measurement because of problems encountered (e.g. clothing the respondent was wearing), this was considered unreliable. Only reliable measurements were used to calculate mean waist circumference.

Physical activity assessment
The ELSA used three questions to measure participation in physical activities of mild, moderate and vigorous intensity. Unified expression of these questions was "We would like to know the type and amount of physical activity involved in your daily life. Do you take part in sports or activities that are vigorous/moderately energetic/mildly energetic more than once a week, or once a week, or one to three times a month, or hardly ever or never ?". A card was presented for participants when being asked about these questions, with examples of different types of activities provided. Detailed examples included: 1) laundry, home repairs for mild intensity; 2) gardening, cleaning the car, walking at a moderate pace, dancing, floor or stretching exercises for moderate intensity; 3) running or jogging, swimming, cycling, aerobics or gym workout, tennis, digging with a spade or shovel for vigorous intensity.
Participants were also allowed to list additional examples and required to decide which of the three categories (vigorous, moderate and mild) could best match the activity.

GBTM modeling
The GBTM can fit non-monotonic trajectories and support multiple trajectory shapes including linear, quadratic and cubic. It also allows specification of number of trajectory groups before fitting the model. We selected number of groups from 3 to 7 and compared model fit statistics of the Bayesian information criterion (BIC) of different trajectory models to determine the most optimal number of trajectory groups.
Then, we determined that modeling 5 trajectory groups was appropriate for global, mild and moderate physical activity trajectories modeling, while 4 trajectory groups for vigorous physical activity trajectories modeling.
We further evaluated different trajectory shapes for each trajectory group by testing the null hypothesis that the shape parameter for the group equals zero. We also used graphics of trajectory group means to help determine which shape best fit each trajectory group. After the procedure, we determined that the best 5-group trajectory model for global, mild and moderate physical activity consisted of 2 cubic and 3 linear trajectories, while the best 4-group model for vigorous physical activity consisted of 3 cubic and 1 linear trajectories. Then the estimated trajectory groups membership was included as the independent variable for further multivariate analysis.

COX regression modeling
Association between physical activity trajectories, cumulative physical activity participation Z score and incident DM risk was evaluated using proportional hazard regression (Cox regression) model. We evaluated proportional hazard assumption for all included covariates using weighted Schoenfeld residuals, and addressed violation of assumption by including covariates as interaction term with time scale variable (years from wave 4 to first occurrence of event in interest or censoring). After assessment, the covariate age was identified of significant correlation with time scale variable, and we included the interaction term of age × time for all Cox regression models adjusted for age.