Combined effects of body mass index and unhealthy behaviors on disability in older Japanese adults: the Okayama study

Study participants

The Okayama Study, a longitudinal retrospective cohort study, was conducted in Okayama city, Okayama prefecture, Japan, in 2018. Okayama is a city encompassing a wide socioeconomic and urban–rural distribution. The aim of the Okayama Study was to investigate the relationship between behaviors and certification for Long-term care insurance (LTCI). The present study was part of the Okayama Study. The requirement for informed consent was waived, as this was a retrospective study using anonymized data. The Ethics Committee of the Okayama University Graduate School of Medicine Dentistry and Pharmaceutical Sciences and Okayama University Hospital reviewed and approved the study (approval number K1703-037).

Figure 1 shows the flowchart of the study. From 2006 to 2007, the study enrolled 109,757 participants residing in Okayama city who completed the health checkup questionnaire (conducted by Healthy Service of the municipal government of Okayama) for the elderly at a hospital or in the participants’ homes under the assistance of doctors or nurses. We excluded the following persons: 41,979 persons who were still under 65 years in late 2014; 11,878 persons certified by LTCI before follow-up; 12 persons with unknown dates of death; and 2,828 persons with missing values for BMI, smoking, alcohol consumption, physical activity, or eating habits. We did not exclude persons with disease as was done in previous studies (Padwal et al., 2016; Yu et al., 2017; Zhang et al., 2016). Subsequently, 53,060 participants (available rate was 48.3%) were included in our analysis. During the 7-year period, 3,763 participants had not been certified by LTCI until they were lost to follow-up on account of death or migration from the study area. This resulted in a follow-up rate of 92.9%. Among the 309,336 person-years, 14,298 participants were certified by LTCI.

Assessment of exposure

Based on the measured height and weight, BMI was calculated as weight (kg) divided by the square of height (m²). Moreover, the WHO Asia criteria was applied to categorize BMI into four groups: underweight (<18.5 kg/m²), normal weight (18.5–23.0 kg/m²), overweight (23.0–27.5 kg/m²), and obesity (≥27.5 kg/m²) (WHO Expert Consultation, 2004). Information on behaviors were obtained from the completed health check-up questionnaire as follows: smoking status was classified as non-smoker (including former smoker) or current smoker (Kvaavik et al., 2010; Tamakoshi et al., 2009). Alcohol consumption was classified as non-drinker, light-to-moderate (equal or less than two units per day), or heavy (more than two units per day) (Minster of Health, Labour & Welfare, 2012). Physical activity was classified as inactivity or regular (more than half an hour of exercise per week for 1 year) (Minster of Health, Labour & Welfare, 2012). Eating habits were classified as healthy or unhealthy (four items, including not controlling salt and calorie intake and intake of few vegetables and calcium, which were associated with several chronic diseases) (Joint WHO/FAO Expert Consultation, 2003; Woo et al., 1998). We considered current smoker, physical inactivity, alcohol consumption other than light-to-moderate, and unhealthy eating habits to be unhealthy behaviors according to previous studies (Artaud et al., 2013; Byun et al., 2010; Chakravarty et al., 2012; Harrington et al., 2010; Sabia et al., 2009; Veronese et al., 2016) and the recommendation of Health Japan 21 (2nd series) from the Ministry of Health, Labor, and Welfare (Minster of Health, Labour & Welfare, 2012).

Ascertainment of disability

In the present study, certification of LTCI was used as a measure of incident functional disability in older adults—a process that is objectively, fairly, and nationally standardized in Japan (Tsutsui & Muramatsu, 2005). The LTCI is a mandatory national social insurance system, in which everyone aged above 40 years pays premiums, and everyone above 65 years is eligible to apply for certification from the municipal government, based on physical and mental disability as assessed by an objective test (Campbell & Ikegami, 2003). Briefly, a care manager will visit and interview the elderly person at home to assess the physical and mental health status by using several scales developed by the Ministry of Health, Labor, and Welfare. The care managers are licensed professionals with at least 5 years of experience, such as nurses, physicians, social workers, and physical therapists. They have also undertaken a few days of training about the interview. According to the results of the care manager’s investigation and physician’s opinion, municipal certification committee experts for LTCI (who are experienced or are of reputable academic standing in the fields of health, medical treatment, and welfare) determine the certification (eligible or not) and its level. Eligibility includes seven levels, ranging from support level 1 to level 2, and care level 1 to level 5. Support level 1 indicates a limitation in instrumental activities of daily living but independence in basic activities of daily living (Olivares-Tirado & Tamiya, 2013). In previous studies, the certification of LTCI was shown to be well correlated with other assessments of disability. Furthermore, LTCI certification has been applied as a measurement of incident functional disability in older adults by the Japanese government (Olivares-Tirado & Tamiya, 2013); several studies also used LTCI certification to define incident disability and analyzed the association between factors and incident disability in Japan (Akune et al., 2014; Ishii, Ogawa & Akishita, 2015; Nitta et al., 2010; Zhang et al., 2016). If an older adult was determined to be eligible for certification (support level 1 or higher), he/she was regarded as having incident functional disability. We obtained LTCI information from the Okayama City Public Health Center under privacy protection.

Assessment of covariates

Covariate assessment included age (continuous), sex, current employment (yes/no), self-rated health (good/other than good), and current diseases (yes/no, including heart disease, hypertension, kidney disease, diabetes mellitus, liver disease, anemia, and hyperlipidemia).

Statistical analysis

The participants contributed person-years from turned age ≥65 years, during the follow-up period, to the date of incident disability, loss to follow-up (death or emigration from Okayama City), or end of the follow-up period (December 31, 2014), whichever came first. The proportional hazards assumption was checked on both Kaplan–Meier curves and log–log curves. No violations of proportionality were observed. Cox proportional hazard models were used to calculate hazard ratio (HR) values for incidence of functional disability with their 95% confidence intervals (CIs) across BMI categories. Three models were used to calculate adjusted HR values as follows: age and sex were the first to be adjusted (model 1); followed by current employment, current diseases, and self-rated health (model 2); and finally, behaviors (model 3). Subgroup analysis by sex were performed to examine for possible heterogeneity.

Then, we evaluated the combined effects of BMI and unhealthy behaviors on the risk of incident functional disability, which was the primary purpose of our study. First, in adjusted models 1 and 2, we calculated the HR values for the incidence of functional disability across the categories of BMI after stratifying the participants by the number of unhealthy behaviors, which defined BMI in the range of 18.5–23.0, as the reference for each number of unhealthy behaviors. Second, in adjusted model 2, we calculated the HR values for incidence of functional disability according to the joint classification of BMI and unhealthy behaviors, which defined both BMI in the range of 18.5–23.0 and no unhealthy behaviors as the reference for all categories. In this step, to test for the linear trends in each BMI range, p-value for trend was estimated by entering ordinal numbers of unhealthy behaviors in each group in the Cox regression model. We also ran subgroup analysis by sex in these two steps.

We ran four sensitivity analyses to test the robustness of our results. First, individuals whose disability events occurred within the first 2 years of follow-up were excluded; second, individuals with diseases (including heart disease, hypertension, kidney disease, diabetes mellitus, liver disease, anemia, and hyperlipidemia) were excluded; third, the unhealthy group of eating habits was redefined (more than two items were unhealthy). Finally, weights were assigned to each unhealthy behavior based on the beta coefficients from the multivariable adjusted Cox model with incidence of functional disability.

All statistical analyses were performed by using SPSS version 21 (IBM Corp., Armonk, NY, USA). All p-values were two-sided and those less than 0.05 were considered statistically significant.

Baseline characteristics

Among 53,060 participants, the mean (SD) age was 71.4 (7.5) years according to the health check-up questionnaire. Table 1 shows the characteristics of the participants according to BMI categories. Participants with higher BMI values were more likely to have a disease. Participants with lower BMI values were more likely to smoke.

Association between BMI and incident functional disability

Table 2 shows the association between BMI and incident functional disability, separately. After multivariate adjustment for potential confounders, the results showed significantly higher HR values (95% CI) in the final model: 1.18 [1.11–1.25] overall, and in men (1.27 [1.14–1.40]) and women (1.14 [1.06–1.22]) among the underweight; as well as 1.26 [1.19–1.34] overall, and in men (1.15 [1.02–1.29]) and women (1.30 [1.22–1.40]) among obese participants. However, no significant association was observed in the overweight participants for both sexes.

Combined effects of BMI and unhealthy behaviors on the risk of incident functional disability by the number of unhealthy behaviors

Table 3 shows the combined effects of BMI and unhealthy behaviors on the risk of incident functional disability by the number of unhealthy behaviors, defined only by BMI range 18.5–23.0, as the reference for each number of unhealthy behaviors. After multivariate adjustment for potential confounders, among participants with no unhealthy behaviors, only the HR value (95% CI) for underweight men was significantly high at 1.47 [1.09–1.97]. Of participants with more than one unhealthy behavior, there were significantly higher HR values (95% CI): among the underweight, for one, 1.17 [1.05–1.30]; two, 1.16 [1.06–1.27]; and three or four, 1.24 [1.08–1.43] unhealthy behaviors. Among the obese, for one, 1.18 [1.63–1.31]; two, 1.29 [1.19–1.41]; and three or four, 1.33 [1.14–1.56] unhealthy behaviors. In the subgroup analyses, we found significantly higher HR values (95% CI) for underweight men for one, 1.36 [1.13–1.63] and for three or four, 1.31 [1.06–1.63] unhealthy behaviors. Significantly higher HR values (95% CI) were also observed in women for two unhealthy behaviors among the underweight (1.17 [1.06–1.30]). Among the obese, women with one, 1.27 [1.12–1.44]; two, 1.32 [1.20–1.45]; and three or four, 1.37 [1.13–1.66] unhealthy behaviors also had significantly higher HR values.

Combined effects of BMI and unhealthy behaviors on the risk of incident functional disability according to the joint classification of BMI and unhealthy behaviors

Figure 2 (Table S1) shows the combined effects of BMI and unhealthy behaviors on the risk of incident functional disability according to the joint classification of BMI and unhealthy behaviors, which defined both BMI in the range of 18.5–23.0 and no unhealthy behaviors as the reference. Even when the BMI value was within the normal-to-overweight range, the HR value (95% CI) was still significantly higher in the presence of unhealthy behaviors. These were 1.17 [1.01–1.21] overall, 1.18 [1.03–1.35] in men with one; 1.29 [1.18–1.41] overall, 1.45 [1.26–1.66] in men and 1.17 [1.03–1.33] in women with two; and 1.38 [1.24–1.54] overall, 1.67 [1.43–1.95] in men and 1.17 [1.01–1.36] in women with three or four unhealthy behaviors among normal weight participants. In overweight participants, these were 1.16 [1.05–1.27] overall, 1.16 [1.01–1.33] in men with one; 1.26 [1.15–1.38] overall, 1.34 [1.17–1.54] in men and 1.17 [1.02–1.33] in women with two; and 1.47 [1.31–1.64] overall, 1.64 [1.39–1.93] in men and 1.32 [1.13–1.54] in women with three or four unhealthy behaviors. In each BMI category, the risk of incident functional disability was graded to be increasing (p < 0.05 for linear trend) among participants with one, two, three, or four unhealthy behaviors. This linear trend was also significant by sex (p < 0.05 for linear trend), and the highest significant HR value (95% CI) in the entire study participants was 1.87 [1.58–2.20] overall, 2.20 [1.64–2.92] in men and 1.66 [1.35–2.04] in women, in combination with both at least three unhealthy behaviors and BMI ≥ 27.5.

Sensitivity analyses

The results of the sensitivity analyses excluding participants whose disability event occurred within the first 2 years of follow-up (Fig. S1); the results excluding participants with diseases (Fig. S2); the results redefining the unhealthy eating habits group (Fig. S3); and the results assigning weights to each unhealthy behavior (Fig. S4), were similar to those of the main analysis.