The present study was based on a cross-sectional multicenter study among three generations, consisting of dietetic students (freshmen), their mothers, and grandmothers. A total of 85 universities, colleges, and technical schools in 35 of 47 prefectures in Japan participated. The survey of institutions in northern and western Japan was conducted from April to May 2011 and that in eastern Japan from April to May 2012. All measurements at each institution were conducted according to the survey protocol. Briefly, a collaborator at each institution explained the general purpose and an outline of the survey to the total of 7016 participants (dietetic students) and distributed a dietary assessment questionnaire and lifestyle questionnaire during the orientation session or a first lecture designed for freshmen in April 2011 or 2012. The collaborators also requested those students able to directly distribute the questionnaires to their mothers and grandmothers to invite their mother and one grandmother to join the study. Recruitment priority was given first to the maternal grandmother; or if unavailable, to the paternal grandmother; or finally to a 65-89 year-old female acquaintance of the student. The student provided written and oral explanations of the general purpose and an outline of the survey to his/her mother and grandmother. Written informed consent was obtained from all participants, and also from a parent for participants aged <20 years. A total of 4933 students, including 4656 women and 277 men (response rate = 70.3%), 4044 women for the mother’s generation (57.6%), and 2332 women for the grandmother’s generation (33.2%) answered both questionnaires. The protocol of the study was approved by the Ethics Committee of the University of Tokyo Faculty of Medicine (No. 3249).
The subjects analyzed in the present study were the participants in the grandmothers’ generation (n = 2332). We excluded those subjects who lived in eastern Japan and answered questionnaires in 2011 (n = 47), because we assumed that they could not report their usual dietary habits and lifestyle due to the occurrence of the Great East Japan Earthquake in March 2011. We also excluded a woman who was in an institution where the response rate was extremely low (2%). Further, we excluded subjects whose age, height, weight, or residential area were missing (n = 20); those aged <65 years (n = 65); and those with a reported energy intake less than half of the energy requirement for the lowest physical activity category according to the Dietary Reference Intakes for Japanese, 2010 (<725 kcal/d; n = 14) , or more than 1.5 times of the energy requirement for the highest physical activity category (>3300 kcal/d; n = 32). We further excluded those with Parkinson’s disease (n = 8), chronic kidney disease (n = 13), those who were unable to walk (n = 20), and those with missing information on the variables used for the purpose of multivariate analysis (n = 4). The final sample thus comprised 2108 women aged 65-94 years.
Dietary habits during the preceding month were assessed using a previously validated, brief-type self-administered diet history questionnaire (BDHQ), which can assess the habitual dietary intake [15, 16]. Responses to the BDHQ as well as an accompanying lifestyle questionnaire were checked once by research staff at the study office. If any missing or erroneous responses were given to questions which were essential for the analysis, the subject was asked to complete those questions again. Details of the BDHQ’s structure, method of calculating dietary intake, and validity for commonly studied food and nutrient intakes have been published elsewhere [15, 16]. Briefly, the BDHQ is a four-page fixed-portion questionnaire used to estimate the dietary intake of 58 food items. To facilitate reading and completion, the present study used a large-print version which increased the size to 10 pages but contained no other changes to structure or content [15, 16]. The food items and portion sizes contained in the BDHQ were derived primarily from a food list used in the National Health and Nutrition Survey of Japan and from several recipe books for Japanese dishes [15, 16]. Estimates of the intake of the 58 food items and the intakes of energy, total protein, fat, and carbohydrate were calculated using an ad hoc computer algorithm for the BDHQ which was based on the Standard Tables of Food Composition in Japan . Protein from fish and shellfish, meat, eggs, and dairy products was included in animal protein. Protein from cereals, pulses, potatoes, confectionaries, fruits, vegetables, alcoholic beverages, and non-alcoholic beverages was included in plant protein. Intakes of eight selected amino acids, namely leucine, isoleucine, valine, methionine, cysteine, branched chain amino acids (sum of leucine, isoleucine, and valine), sulfur amino acids (sum of methionine and cysteine), and essential amino acids were estimated using the answers to the BDHQ and the amino acid composition database . Pearson’s correlation coefficients of protein intake between from the16-d dietary record and from the BDHQ in 92 women aged 31-69 was 0.35 , and those of selected amino acids were 0.36 for leucine, 0.34 for isoleucine, 0.34 for valine, 0.31 for methionine, and 0.37 for cysteine (unpublished observations, H. Suga, ). Although dietary supplement use was queried in the lifestyle questionnaire, intake from supplements was not included in the analysis due to the lack of a reliable composition table of dietary supplements in Japan. The percentage contribution of each food group to total protein was calculated by dividing daily protein from each food group by daily individual total protein.
Although frailty was operationally defined by Fried and colleagues  to include the measures of walking speed for slowness and grip strength for weakness, we did not obtain these measures in our study, but rather used the modified definition developed by Woods and colleagues . Frailty was assessed using the following four components: 1) slowness and weakness (score of the physical functioning scale of the Japanese version short-form 36-item health survey (SF-36) <75 [18–20]); 2) exhaustion (score of the vitality scale of SF-36 <55); 3) low physical activity (those in the lowest quartile); and 4) unintentional weight loss (weight loss in the previous one year >5%). Physical activity was calculated as the average metabolic equivalent-hours, on the basis of the self-reported duration of five activities (walking, bicycling, standing, running, and high-intensity activities) and sleeping and sitting hours over the preceding month, and the metabolic equivalent (MET) value assigned to each activity. These assigned MET values were 3.5 for walking, 7.5 for bicycling, 3.2 for standing, 7.0 for running, 8.0 for high-intensity activities, 1.0 for sleeping, and 1.3 for sitting . Weight loss was calculated from the self-reported weight at the time the BDHQ questionnaire was completed and that one year previously. Subjects with weight loss were asked the question, “Did you lose weight intentionally in the previous year?”, with an answer of no considered to indicate unintentional weight loss. All the questions required to assess frailty except for current weight were incorporated in the lifestyle questionnaires. Current weight was obtained from the response to the BDHQ.
Slowness and weakness was scored as two points, and the other components as one point each. Total frailty score was the sum of all available scores (0-5), with those subjects with a total score ≥3 defined as frail .
The subjects reported birth date and body height in the BDHQ. Body mass index (BMI) was calculated as current body weight (kg) divided by the square of body height (m). In the lifestyle questionnaire, the subject reported her residential area, which was grouped into six regions (Hokkaido and Tohoku, Kanto, Hokuriku and Tokai, Kinki, Chugoku and Shikoku, and Kyushu) and also into three categories according to population size (city with a population ≥1 million, city with a population <1 million, and town and village). The subject also reported in the lifestyle questionnaire if she was living alone, as well as her marital status (single, married, widowed, and separated), education (≤junior high school and others, high school, and ≥ college), current smoking status, and dietary supplement use. A history of chronic disease, including stroke, myocardial infarction, hypertension, diabetes, and chronic rheumatism, were considered to be factors which influenced the current state of frailty because the proportions of subjects with histories of these diseases significantly differed between the frail and non-frail group (data not shown). Alcohol drinking was assessed as part of the BDHQ. Depression symptoms were assessed using the Center for Epidemiologic Studies Depression (CES-D) scale [22, 23] incorporated in the lifestyle questionnaire, with subjects with a CES-D score ≥16 considered to have depression symptoms.
All nutrient intakes were adjusted for energy by the residual method using a linear regression model . The subjects were divided into quintiles according to each dietary intake. Odds ratios (ORs) and 95% confidence intervals (CIs) for frailty were calculated after adjusting for potential confounding factors. The initial logistic regression model was a crude model into which covariates were added using a forward selection method. The result from a multivariate adjusted model which included a variable of “history of chronic disease (yes or no)”, which indicated the presence of any of the diseases, did not differ to that of a model adjusted for each disease individually as separate variables (data not shown). We therefore treated these diseases as one variable. Final multivariate models were adjusted for age (y, continuous), BMI (kg/m2, continuous), residential region (six regions), size of residential area (three areas), living alone (yes or no), current smoking (yes or no), alcohol drinking (yes or no), dietary supplement use (yes or no), history of chronic disease (yes or no), depression symptoms (yes or no), and energy intake (kcal/d, continuous). Survey year (2011 or 2012), marital status (four categories), and education (three categories) were not included in the models, because these variables had no influence on the relationship between dietary variables and frailty (P >0.10). All statistical analyses were performed with SAS statistical software, version 9.3 (SAS Institute Inc., Cary, NC, USA). All reported P values were two-tailed, with a P value of <0.05 considered statistically significant.