Study population
A 3-year longitudinal observational survey on health and daily life (baseline: 2011; follow-up: 2014), using a self-administered questionnaire with elderly people, was conducted in Kurihara city, Miyagi Prefecture, Japan. Kurihara city, which is in northeast Japan and has the largest area in Miyagi Prefecture, was established in 2005 by merging 10 municipalities (As of October 2010, 24,383 of its population of 76,851 were aged 65 years or older.). In this study, we extracted six of the 10 regions (former municipalities) as a study area with the cooperation of Kurihara City Hall. To obtain the representative sample of Kurihara city, these regions were extracted based on regional characteristics (classified as densely populated area, plains area, intermediary area between plains and mountains, or mountainous area based on population density and geographical conditions) and population size (classified as higher or lower than 7500 residents). The eligible population was all residents aged 65 years or older and lived in the extracted six regions at the time of each survey. Those who were facility residents and hospital inpatients were, however, excluded from the survey. At the baseline survey, we sent the questionnaire to 14,097 residents in February 2011 and received responses from 11,821 (a response rate of 83.9%). Of those, 8375 residents responded to a follow-up survey, which was conducted in January 2014. Both surveys were conducted using anonymized IDs, and the results of both surveys were linked. These processes were also conducted with the cooperation of Kurihara City Hall.
Of the 11,821 individuals who responded to the baseline survey, 3419 were selected as the target population for this study, according to the following criteria: those who scored 13 points (perfect score) on the Tokyo Metropolitan Institute of Gerontology Index of Competence (TMIG-IC) and had valid responses to the items on living arrangement and covariates at the baseline. TMIG-IC measures higher-level competence in elderly people using 13 items. It comprises three subcategories, namely, instrumental self-maintenance (five items, e.g., “are you able to shop for daily necessities” or “are you able to prepare meals by yourself”), intellectual activity (four items, e.g., “are you able to fill out forms for your pension” or “do you read newspapers”), and social role (four items, e.g., “do you visit the homes of friends” or “are you sometimes called on for advice”) [20]. All questions are answerable by “yes” or “no.” Points are calculated by adding up the number of “yes” answers. The higher scores indicate higher functional capacity. In order to avoid reverse causality that low functional capacity had preceded a specific living arrangement at baseline, we selected well-performing respondents who had the highest scores. Of the 3419 respondents, 2814 (82.3%) responded to the follow-up survey; 2627 had a valid response to TMIG-IC at the follow-up as outcome measurement (final tracking rate of 76.8%). We analyzed these 2627 respondents in this study. Figure 1 illustrates the conceptual framework for the sampling process described above. The study protocol was approved by the Research Ethics Committee in Keio University Shonan Fujisawa Campus (No. 44) and the Ethics Committee of Faculty of Medicine, Toho University (No. 25104).
Exposure measurement
The exposure in this study was the living arrangement at baseline. From the responses about persons with whom they were living, we classified the living arrangement into following five categories: “with spouse only,” “living alone,” “with child and his/her spouse,” “with child without his/her spouse,” “with other family/person.” These categories were set with reference to the reports of the national survey, Comprehensive Survey of Living Conditions, conducted by Japan’s Ministry of Health, Labour and Welfare [21]. With child and his/her spouse means that respondents were living with at least a child and his/her spouse. With child without his/her spouse refers to living with at least a child. These two categories included cases in which the spouse of respondent or other family is present. With other family/person involved the cases not applicable to the other four categories.
Outcome measurement
The outcome measurement was the decline in functional capacity at the follow-up survey (about 3 years later). We defined “decline” as a decrease in TMIG-IC score to 10 points or less from 13 points at the baseline. This cutoff value was set based on the average value of Japanese people aged over 65 years [22] and used in the previous study [23].
Covariates
In addition to age and sex, educational attainment, health behavior and condition (current drinking, current smoking, history of major diseases, and depressive symptoms), and activities inside and outside the home (housework, social participation, and relationship with neighbors) at the baseline survey were considered as covariates that might be related to the functional capacity of the elderly. Age was a continuous variable. The other covariates were dichotomized as follows: educational attainment (“≥10 years” and “<10 years”), current drinking (“no” and “yes”), current smoking (“no” and “yes”), history of major diseases (“no” and “yes”), depressive symptoms (“normal: 1 point or less out of 5 measured by the 5-item Geriatric Depression Scale [GDS5]” and “have depressive symptoms: 2 points or more on the GDS5”), housework (“mainly do” and “not mainly do”), social participation (“yes” and “no”), and relationship with neighbors (“frequent” and “not frequent”). History of major diseases was defined as having any one of the following diseases known to be causes of death or disability in older adults, referring to the previous study [24]: stroke, myocardial infarction/angina, diabetes, Parkinson’s disease, femoral neck fracture, and cancer. GDS5 was developed as a short form of the 30- or 15-item GDS and composed of 5 items such as life satisfaction and feeling of helplessness [25]. The validity of the GDS5 and the cutoff value (2 points or more) were verified [26]. In addition, GDS5 was showed to be associated with a future decline in ADL in the elderly [27]. Social participation was measured by the active members of a group/organization in the following four categories: local community groups; sports, hobby, or leisure group; voluntary organization or nonprofit organization; or other organizations. These categories were used in the previous study and associated with a future decline in ADL and death [28]. Though TMIG-IC includes a social role as a subcategory, we put activities inside and outside the home (housework, social participation, and relationship with neighbors) into covariates as indicators related to more independent activity at the home and social capital in the community [29].
Statistical analysis
After stratification by gender, odds ratios (ORs) of the presence on outcome were estimated using a multivariable logistic regression analysis, to analyze whether the living arrangement at baseline was associated with the decline in functional capacity after 3 years. The reference category was “with spouse only,” based on previous studies [17, 30]. The reason for stratification by gender in the analyses is that there was a statistically significant interaction between living arrangement and sex on the outcome (p for interaction = 0.014 in the crude model). First, the association of each exposure variable with outcomes was assessed in the model adjusting for age (Model 1). Next, educational attainment, current drinking, current smoking, history of major diseases, and depressive symptoms were added to Model 1 (Model 2). Finally, activities inside and outside the home (housework, social participation, and relationship with neighbors) were included in the model (Model 3). In addition, the following three sensitivity analyses were conducted on Model 2. First, to examine whether ORs were changed when each covariate of activities inside and outside the home was added separately (sensitivity analysis 1). Second, to confirm whether the results did not depend on the cutoff value, analyses were conducted in which the cutoff value was changed from 10 points to 9 points and 11 points, respectively (sensitivity analysis 2). Third, the analyzed population was expanded from those who scored 13 points on the TMIG-IC to those who scored 11 points or more (sensitivity analysis 3). The statistical significance level was set at p < 0.05. All analyses were performed using STATA, version 14.0 (STATA Corporation, College Station, Texas).