Study population
This longitudinal prospective study utilized data from the cohort study, which has been described previously [16]. This cohort study had been managed by the National Institute of Health and Nutrition (NIHN) since 2007 and aimed to evaluate the association between lifestyle-related diseases and modifiable risk factors, such as dietary intake and physical activity. This cohort study included 760 adults, who agreed to participate in this study, aged 26–85 years, and lived in the Tokyo metropolitan area (n = 504) and Okayama prefecture (n = 256) in Japan from 2007 to 2018 (population density: 6168.7 people/km2 in Tokyo and 270.1 people/km2 in Okayama). These participants were recruited when they participated in a specific health examination conducted at the Okayama Southern Institute of Health and the NIHN. All participants were requested to participate in a health checkup conducted through an annual face-to-face meeting and a mail survey. The investigations were conducted annually using the same survey content and methodology, and the participants were followed up for a maximum of 12 years (Table S1).
Among the participants that were initially included at baseline (n = 760), we excluded individuals with missing data on age and sex (n = 1), those whose physical activity could not be measured using an accelerometer (n = 5), and those who completed the assessment of physical activity only once (n = 65). The final dataset included 689 participants (3914 measurements) whose physical activity information from at least two accelerometer surveys was available. Of all participants in this study, 175 individuals (25.4% of 689 individuals) under 65 years old with a lower amount of MVPA (less than 3.3 metabolic equivalents [METs]-h/day) received an intervention composed of five-time brief counseling sessions to increase MVPA.
Evaluation of physical activity
Objective physical activity was measured using a validated triaxial accelerometer (EW4800, Panasonic Co., Ltd, Osaka, Japan) against total energy expenditure (TEE), which measured movement using both the doubly labeled water (DLW) and metabolic chamber methods [17]. Research staffs were educated on how to use and handle the accelerometer, using the manual. Participants were instructed to wear an accelerometer around the waist upon waking up till bedtime except when swimming, sleeping, and bathing. All participants were asked to wear a triaxial accelerometer for 28 days. The intensity for every minute, basal metabolic rate, step count, and physical activity level (PAL) were determined using the maker’s algorithm. Wearing time was defined as 24 hours minus non-wearing and no signal time and it excluded these data with wearing time less than 6 hours per day [18]. To calculate the mean physical activity time, the sum of all physical activities surveyed over at least 7 days (including weekdays and weekends) was divided by the number of survey days.
We obtained the daily physical activity times corresponding to <1.5 METs (sedentary), 1.5 to 2.9 METs (light intensity physical activity: LPA), and ≥3.0 METs (MVPA). The inactive times were calculated using the sum of sedentary (<1.5 METs) and non-wearing periods, which was calculated as 1440 minutes − wearing periods (daily time spent in sedentary, light, moderate, and vigorous physical activity times) [16]. We included inactive time, LPA, MVPA, TEE, PAL, and step count as objective physical activity-related variables.
Self-reported covariate
Health information, such as medical history and smoking status, was obtained using self-reported structured questionnaires. Dietary intake was evaluated using the Brief-type self-administered Diet History Questionnaire (BDHQ), which consists of 58 food and beverage items that were validated against dietary records [19]. The diet quality was assessed using a previously validated Nutrition Rich Food (NRF) index 9.3 score [20]. This score ranges from 0 (worst diet quality) to 900 (best diet quality) [20]. The research staff checked all questionnaires and interviewed respondents with unanswered questions, unclear responses, or to confirm answers. Based on the data obtained regarding the comorbidity status of each individual, the comorbidity score was added to obtain a total score (including hypertension, dyslipidemia, diabetes, ischemic heart disease, other heart diseases, cerebrovascular diseases, renal failure, cancer, osteoporosis, and depression) ranging from 0 (no comorbidity) to 10 (poor status).
Measured covariate
Each participant’s body weight was measured while in light clothing (BC-600, TANITA Corp., Tokyo, Japan). The body mass index (BMI) was calculated by dividing the measured body weight (kg) by the square of the height (m2). The waist/hip ratio was calculated as by the circumferences of the waist (at the level of the navel) to hip (the greatest posterior protuberance, perpendicular to the long axis of the trunk). Trunk flexibility was measured using a sit-and-reach digital instrument (T.K.K.5112; Takei Scientific Instruments Co., Ltd, Japan). Resting heart rate (HR) was measured using an electrocardiogram mounted on a pulse wave examination device (form PWV/ABI BP-203RPEⅡ, Omron Colin, Kyoto, Japan). Hemoglobin was measured using a colorimetric method that utilizes sodium lauryl sulfate from fasting blood samples (≥12 hours). Leg press power was measured using a leg muscle strength measuring device (Anaeropress 3500, COMBI, Tokyo, Japan). This device measured the unidirectional power production of the leg extensors. Grip strength was evaluated using a Smedley Hand Dynamometer (Grip-D TKK5101, Takei Scientific Instruments, Niigata, Japan). Measurements were taken twice from each hand, and the mean of the highest value of each hand was used.
Statistical analysis
The participants’ characteristics were expressed as a number and a percentage for categorical variables and mean and standard deviation for continuous variables. We performed imputation to missing values of covariates from five data sets, which were created using the multiple imputation method that utilizes multivariate imputation through a chained equation (MICE) by R software [21]. The details are shown in Table S1. These missing values were assumed as missing at random.
To identify the longitudinal trajectory from repeat measures of physical activity, we estimated a single mean physical activity trajectory across the group using a sex-stratified model that uses the latent growth curve models (LGCM). In addition, the latent class growth models (LCGM) were applied to assess whether study participants could be classified into multiple trajectory groups through the maximum likelihood method. These analyses were used by the STATA macro TRAJ [22], and to construct trajectory shape with a cubic specification. The best-fitting model for LCGM was identified by estimating models with 2 to 8 latent clusters and comparing them using the sample size of the clusters (≥ 5%) and the Bayesian Information Criterion as the primary fit index [23].
To calculate the correlation coefficients by repeated measurement and cross-sectional analysis between chronological age and physical activity-related variables, we performed Repeated Measures Correlation by the R software [24] and Pearson’s correlation analysis, respectively. We evaluated the accuracy and precision of the mean of the group's physical activity-related variables trajectory using previously reported equations [25]. These equations were used to estimate the required sample size and periods from within‐person variance, between‐person variance, and the ratio of within‐person to between‐person variance.
To evaluate the factors associated with the physical activity trajectory, we used the multivariate regression analysis of random-effect panel data, which is a method that evaluates related factors from the longitudinal changes of the dependent variable and the explanatory variable to adjust the between-individual characteristics [26]. To evaluate factors associated with the physical activity trajectory, the multivariate analysis included the age (continuous), sex (female or male), region (urban (Tokyo) or local (Okayama)), BMI (continuous), waist/hip ratio (continuous), comorbidity score (continuous), smoking status (never smoker or past and current smoker), alcohol intake (continuous), energy intake (continuous), NRF 9.3 score (continuous), hemoglobin (continuous), HR (continuous), hand grips (continuous), leg power (continuous), and trunk flexibility (continuous). These variables were selected in reference to covariates used in previous studies [10-12, 27-29]. The results of these analyses were shown in regression coefficients (RC) and 95% confidence interval (CI) for each variable per unit increment. To conduct the sensitivity analysis for results, we conducted a similar analysis using the complete cases dataset.
A two-sided p-value < 0.05 was considered significant. All analyses were performed using STATA MP version 15.0 (StataCorp LP, College Station, TX, USA) and/or R software 3.4.3 (R Core Team, Vienna, Austria).
Ethical considerations
This study was conducted according to the guidelines laid down in the 1964 Declaration of Helsinki and all procedures involving research study participants were approved by the Research Ethics Committee of the National Institute of Health and Nutrition (approval no. kenei102-01). Written informed consent was obtained from all participants before data acquisition.