This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology-nutritional epidemiology (STROBE-nut) checklist (Additional file 1) [20].
Study design and participants
The present study analyzed cross-sectional data previously collected for the Hatoyama Cohort Study, Kusatsu Longitudinal Study, and Itabashi Cohort Study, which are population-based cohort studies of community-dwelling adults aged 65 years or older. Data were obtained from comprehensive health examinations conducted in the same manner. The details of the study designs and participants have been reported elsewhere [21, 22]. We used data from the year in which brief diet history questionnaires (BDHQ) were distributed, all of which were collected during the period from 2012 through 2014. These studies were approved by the relevant institutional review board, and written informed consent was obtained from all participants.
Of the 1928 participants who agreed to participate in the study (n=576 in the Hatoyama Cohort Study; n=608 in the Kusatsu Longitudinal Study; n=759 in the Itabashi Cohort Study), we excluded those with missing information on dietary intake (n =81), those with under-reported and over-reported energy intakes (energy intakes less than half the requirement for the lowest physical activity category, according to the Dietary Reference Intakes for Japanese, 2015 [< 1050 kcal/day for men aged 65–69 years: n=1; < 925 kcal/day for men aged > 70 years: n=3; < 750 kcal/day for women aged > 70 years: n=2], or more than 1.5 times the energy requirement for the highest physical activity category [> 3750 kcal/day for men aged > 70 years: n=16; > 3000 kcal/day for women aged > 70 years: n=19]) [23, 24], those with severe cognitive impairment, defined as a Mini Mental State Examination Score (MMSE) of < 18 (n=5) [25], and those with missing data for the MMSE (n=79), the present outcome variables (n=81), or covariates (n=50). Ultimately, data from 1606 participants were analyzed (Fig. 1).
Definition of sarcopenia
Sarcopenia was defined by using the algorithm of the Asian Working Group for Sarcopenia 2019 [26]. In accordance with the algorithm, we used the criteria low muscle mass, defined as an appendicular lean mass (ALM)/height2 of < 7.0 kg/m2 for men and < 5.7 kg/m2 for women; low muscle strength, defined as a grip strength of < 28 kg for men and < 18 kg for women; and low physical performance, defined as a gait speed of < 1.0 m/s for men and women. Participants with low muscle mass and either low muscle strength or low physical performance were categorized as having sarcopenia. The procedures for measuring muscle mass, grip strength, and gait speed have been described in detail previously [21, 27].
Dietary assessment
Dietary habits during the preceding 1-month period were assessed with a validated brief self-administered diet history questionnaire (BDHQ) [28, 29]. The BDHQ is a four-page fixed-portion questionnaire and consists of five sections: 1) frequency of intake of 46 foods and non-alcoholic beverages; 2) daily frequency of rice and miso soup intake; 3) frequency of alcoholic drinking and amount of consumption for five alcoholic beverages per typical drinking occasion; 4) usual cooking methods; and 5) general dietary behavior. The validity of this questionnaire has been reported previously [28, 29]. To facilitate reading and completion for older adults, the present study used a large-print version that increased the number of pages to 10. Responses to the BDHQ were checked by research staff for completeness and, when necessary, were reviewed with participants to ensure answer clarity. Dietary intakes of 58 food and beverage items, energy, and specific nutrients were calculated by using an ad hoc computer algorithm based on the Standard Tables of Food Composition in Japan [30].
Dietary pattern analysis
Dietary patterns were assessed by using RRR analysis, which identifies patterns in a set of food groups that explain as much variation as possible in response variables (e.g., nutrients or biomarkers) [16]. Unlike principal components analysis, which derives dietary patterns based on the covariance structure of foods, RRR allows us to more directly link the exploratory identification of dietary pattern to the outcome of interest by the choice of informative outcome-related response variables [31]. The RRR method has been described in detail elsewhere [16]. As predictors, 52 food and beverage items were energy-adjusted with the density method and used [32]. Response variables were selected for nine nutrients— protein, vitamin D, vitamin C, vitamin E, folate, vitamin K, magnesium, iron, and calcium—because these nutrients were reported to protect against sarcopenia or related outcomes in previous studies [5, 33,34,35,36,37] and were variables with a P value of < 0.2 [38] in a preliminary analysis of their intakes and sarcopenia in our study population. Although omega-3 polyunsaturated fatty acids (PUFAs) are of interest because of their demonstrated effects on skeletal muscle health [9], the P value for PUFAs was > 0.2 in our preliminary analysis. Therefore, we decided not to include PUFAs as a response variable. The extracted dietary pattern score was classified by tertile (T1–T3) in all participants.
Covariates
The covariates used in the analyses were sex, age, study site, education, living arrangement (single, with spouse only, or other), smoking habit (never, former, or current smoker), drinking habit (never/rarely, sometimes, or every day), self-perceived chewing ability (can chew anything/almost anything, cannot chew much), frequency of going outdoors, self-reported medical history (hypertension, diabetes, heart disease, stroke, cancer, chronic obstructive pulmonary disease), body mass index (calculated as weight in kilograms divided by height in meters squared), and energy intake.
Statistical analysis
Complete case analysis was used to address missing data (details of missing data are shown in Fig. 1). The characteristics of the study population, by category of dietary pattern score, were compared by using weighted one-way analysis of variance for continuous variables or the Mantel–Haenszel chi-square test for categorical variables. Associations between the dietary pattern extracted by RRR and nutrients as response variables were evaluated by using Spearman rank correlation coefficients. Multiple logistic regression and linear regression analyses were used to examine the associations of the first dietary pattern with sarcopenia and its components. In the logistic regression analysis, odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for sarcopenia in relation to first dietary pattern scores, with the lowest tertile category defined as the reference. Tests for trend associations were based on assigning the ordinal numbers 0–3 to the three categories of first dietary pattern score. In the multiple linear regression analysis, we calculated mean (SE) values of ALM, grip strength, and usual gait speed in relation to tertile of dietary pattern. We also calculated the unstandardized partial coefficient, which reflects change in ALM, grip strength, and usual gait speed per one-tertile increase in conformity to the first dietary pattern.
The multivariate model was adjusted for the following potential confounding variables. The first model was adjusted for age (years, continuous), sex (men or women), and study site (Hatoyama, Kusatsu, or Itabashi) and further adjusted, in model 2, for education (years, continuous), living arrangement (single, with spouse only, or other), smoking habit (current, former, or never), drinking habit (every day, sometimes, or none/rarely), self-perceived chewing ability (can chew anything/almost anything or cannot chew much), frequency of going outdoors (more than once a day or less than once a day), medical history (hypertension, diabetes, cancer, stroke, heart disease, chronic obstructive pulmonary disease; yes or no), body mass index (kg/m2, continuous), total energy intake and MMSE score (score, continuous). These potential confounders were chosen after reviewing previous findings suggesting relations with both the exposure and outcome of interest.
A two-sided P-value of < 0.05 was considered to indicate statistical significance. Dietary pattern analyses (RRR) were performed with SAS version 9.4 (SAS Institute, Inc., Cary, NC, USA), and all other analyses were performed with IBM SPSS Statistics version 23 (IBM Corp, Armonk, NY, USA).