Development of the Japanese version of the Council on Nutrition Appetite Questionnaire and its simplified versions, and evaluation of their reliability, validity, and reproducibility

Background Because few Japanese questionnaires assess the elderly's appetite, there is an urgent need to develop an appetite questionnaire with verified reliability, validity, and reproducibility. Methods We translated and back-translated the Council on Nutrition Appetite Questionnaire (CNAQ), which has eight items, into Japanese (CNAQ-J), as well as the Simplified Nutritional Appetite Questionnaire (SNAQ-J), which includes four CNAQ-J-derived items. Using structural equation modeling, we examined the CNAQ-J structure based on data of 649 Japanese elderly people in 2013, including individuals having a certain degree of cognitive impairment, and we developed the SNAQ for the Japanese elderly (SNAQ-JE) according to an exploratory factor analysis. Confirmatory factor analyses on the appetite questionnaires were conducted to probe fitting to the model. We computed Cronbach's α coefficients and criterion-referenced/-related validity figures examining associations of the three appetite battery scores with body mass index (BMI) values and with nutrition-related questionnaire values. Test–retest reproducibility of appetite tools was scrutinized over an approximately 2-week interval. Results An exploratory factor analysis demonstrated that the CNAQ-J was constructed of one factor (appetite), yielding the SNAQ-JE, which includes four questions derived from the CNAQ-J. The three appetite instruments showed almost equivalent fitting to the model and reproducibility. The CNAQ-J and SNAQ-JE demonstrated satisfactory reliability and significant criterion-referenced/-related validity values, including BMIs, but the SNAQ-J included a low factor-loading item, exhibited less satisfactory reliability and had a non-significant relationship to BMI. Conclusions The CNAQ-J and SNAQ-JE may be applied to assess the appetite of Japanese elderly, including persons with some cognitive impairment.


Introduction
In Japan, the prevalence of under-/malnutrition or body mass index (BMI) 20 was reported to be 16.8% among the community-dwelling elderly !65 years of age according to the National Health and Nutrition Survey, 2013. 1 The estimated prevalence in adults !85 years of age was reported to be 29.6%, and the figure among patients institutionalized in health-care facilities was 77.9%. 2 One of the major factors leading to deterioration of a healthy life among the elderly seems to be under-/malnutrition, which is triggered by loss of appetite (LOA; i.e., anorexia) caused not only by age-related physiological factors, including degraded/impaired oral health, sense of smell and taste, digestive functions, and physical activity, 3e7 but also by dysfunctions of clinical parameters, including mental/psychological disorders (such as dementia and depression), and effects/side effects of medicines. 8e11 Under-/ malnutrition is also associated with socioeconomic factors, including living alone and sparse or loss of family relationships and social/community communication. 12,13 Furthermore, LOA itself inevitably worsens activities of daily living (ADL) and quality of life (QOL) because meals are basic to as well as prerequisite for enjoyment of life, especially for the elderly. To adequately evaluate appetite seems critical; however, Japanese appetite questionnaires used currently are comprised of one or two yes/no questions, without evidence of reliability, validity, or reproducibility.
In 2005, using the Delphi method, Wilson et al 14 developed the Council on Nutrition Appetite Questionnaire (CNAQ), which includes eight items (eTable 1). Conducting studies on a long-term care group (mean [standard deviation {SD}] age 79.2 [9.0] years) and a community-dwelling group (mean [SD] age 53.5 [20.2] years, including an elderly group [age range >60e102 years] and a young and elderly group [age range >20e60 years]), they verified its reliability and concurrent validity using an external lengthy assessment tool (the Appetite, Hunger and Sensory Perception Questionnaire) 15 as the reference standard. The CNAQ and the Simplified Nutritional Appetite Questionnaire (SNAQ), which retains items #1, #2, #4, and #6 from the CNAQ, are now used for young and elderly people (including patients) worldwide in countries including Australia, Malaysia, Germany, and Korea. 16e19 In the present study, we translated and back-translated the original CNAQ and SNAQ into Japanese (the CNAQ-J and SNAQ-J, respectively) according to a standardized procedure and developed the SNAQ for the Japanese elderly (SNAQ-JE) using a series of exploratory factor analyses, and we studied fitting to the structural equation modeling (SEM), reliability, criterion-referenced/-related validity, and reproducibility to verify whether those appetite batteries can be applied to the Japanese elderly.

The original CNAQ and SNAQ
For convenience, we refer to CNAQ items as follows in this article: #1, Appetite; #2, Feeling full; #3, Feeling hunger; #4, Food tastes; #5, Food tastes compared to when younger; #6, Meal frequency per day; #7, Feel sick or nauseated when eating; and #8, Usual mood. The subjects were requested to reply using 1e5 ordinal scales (Likert scales) of each question.

Development of the Japanese versions of the CNAQ-J and SNAQ-J
We obtained permission from the original article's authors, 14 including the senior author (JEM), for development of the Japanese versions. Using a standardized translation and backtranslation method, a Japanese version (CNAQ-J) was developed by two nutrition researchers, a medical doctor, a professor of English, and a Japanese staff member majoring in English. Translation and back-translation were reiterated until equivalent expressions in English were attained. A pilot study of 15 people (aged 20 se90 s, including an elderly person attending a health-care facility) was conducted to examine whether the Japanese expressions adopted for the CNAQ-J were understandable or not. Younger people were also invited to participate in this study because the younger generation can play roles as proxy/surrogate interview-based responders. Minor revisions were added, and the questionnaire was reckoned as feasible and applicable for the main study. The Japanese versions (the CNAQ-J and SNAQ-J) were finalized (eTables 2 and 3).

Study subjects
From July to December 2013, we recruited 816 subjects, including 175 community-dwelling elderly (CE group) attending health promotion classes for a secondary prevention study to improve mild frailties in O City, Aichi Prefecture; 328 receiving meal delivery (MD group) services in N City, Aichi Prefecture; 163 attending day-care (DC group) facilities in O City, Fukuoka Prefecture, and T City, Toyama Prefecture; and 150 staying at group homes (GH group) in Y City, Kanagawa Prefecture.
For a testeretest reproducibility analysis, 54 elderly people (10 CE and 44 DC persons) were invited independently from the main study.

Data collection
We gathered data on subjects' basic characteristics (gender and age), their anthropometric measurements (height, weight, and BMI), the CNAQ-J, the Clinical Dementia Rating (CDR) 20,21 (from the MD, DC, and GH groups), the Mini Nutritional Assessment-Short Form (MNA-SF) 2,22 (from the CE, DC, and GH groups), and the Constipation Assessment Scale-Japanese Version (CAS-J) 23,24 (from the CE and GH groups).
All data, except for anthropometric measurements, were collected using relevant appetite and nutrition-related questionnaires. We obtained self-administered replies from the CE group, self-and/or interview-administered replies by proxy caregivers/ family members of the MD group, and self-and/or interviewadministered replies by surrogate facility staff of DC and GH groups.
Exploratory factor analyses of the appetite batteries, and development of the SNAQ-JE We performed exploratory factor analyses of the CNAQ-J and SNAQ-J under the SEM using the maximum likelihood method, and developed the SNAQ-JE, taking into account the distribution of the CNAQ-J item scores, deleting any item having a smaller factor loading, reducing Cronbach's a coefficient, or deploying a higher brain function.

Examination of fitting to the model
We conducted confirmatory factor analyses to scrutinize the goodness of fit index (GFI), adjusted GFI (AGFI), and a root mean square error of approximation (RMSEA) for the three appetite tools.

Appraisal of reliability/internal consistency
We examined reliability/internal consistency of the CNAQ-J, SNAQ-J, and SNAQ-JE.

Criterion-referenced/-related validity
Using the previously mentioned cutoff values of the CNAQ-J and SNAQ-J, along with cutoffs of 14 or !15 for the SNAQ-JE score (based on an average value 14.4), we categorized people into a lower-score group and a higher-score group. BMI values were contrasted between the two score groups in a cross-sectional manner, and correlations were probed between appetite questionnaire scores and nutrition-related questionnaire values.

Testeretest reproducibility
We studied testeretest reproducibility of appetite questionnaire scores setting an approximately two-week interval.

Statistical analyses
To compare baseline characteristics with the CNAQ-J item scores by subject group, we conducted t-tests for two-group comparisons, analysis of variance (ANOVA) with post hoc Bonferroni adjustment for multiple-group comparisons, and c 2 tests for comparisons of proportions.
Exploratory factor analyses of the appetite batteries were performed adopting the maximum likelihood method, and the SNAQ-JE was developed from these analyses.
Confirmatory factor analyses were then conducted to probe fitting to the model (including GFI, AGFI, and RMSEA) for the CNAQ-J, SNAQ-J, and SNAQ-JE.
Reliability/internal consistency was appraised calculating Cronbach's a coefficient.
For a criterion-referenced/-related validity study, the BMI difference between a higher-score group vs a lower-score group was examined by t-test. We also explored correlations of appetite questionnaire scores with nutrition-related instrument indices by age-and gender-adjusted Pearson correlation coefficients.
Testeretest reproducibility of appetite tool scores was verified adopting intraclass correlation coefficients (ICCs).
For statistical analyses, SPSS ver. 22 (IBM Corp., Armonk, NY, USA) and Amos were used, and P < 0.05 (two-tailed) was assumed as statistically significant.

Ethical issues
The present study protocol was submitted to the Committee of Ethics and Conflict of Interest at the National Center for Geriatrics and Gerontology (Number of receipt #648) and approved. All participants/proxies/surrogates were fully informed about the study and gave written consent.

Study subjects
After excluding 167 individuals (7 CE, 127 MD, 19 DC, and 14 GH group participants) having incomplete information from the 816 subjects recruited, we analyzed the data provided by 649 subjects (168 CE, 201 MD, 144 DC, and 136 GH group participants).
The CDR study, in which the CE group was excluded, showed that the percentage of participants with no problem (score 0) was 22.7%, with dementia suspected (score 0.5) was 19.6%, and with overt dementia (score 1e3) was 57.7% (34.4% light, 15.3% moderate, and 8.0% severe dementia).

The CNAQ-J scores
The average CNAQ-J score of all subjects was 29.3 (SD, 3.4) ( Table 2). The figure of the GH group was highest 30.9 (SD, 3.3), followed by the DC group at 29.8 (SD, 2.6), the CE group at 28.9 (SD, 2.8), and the MD group at 28.2 (SD, 3.8).

Examination of the CNAQ-J structure and development of the SNAQ-JE
An exploratory factor analysis demonstrated that Eigenvalues attenuated as follows: 2.921, 1.006, 0.862, and so forth. Thus, a onefactor (interpreted as appetite) solution appeared reasonable.
First, item #7 exhibited a ceiling effect and was excluded. An exploratory factor analysis of the remaining 7 items deleted item #6 because its factor loading was 0.248 (<0.4). The next analysis Table 1 Demographic and anthropometric characteristics, and CDR scores by gender and study group.  showed that all items had factor loading >0.4, but item #3 was omitted since it was a Cronbach's a coefficient reducer. Item #5 (Food tastes compared to when younger), which required the use of memory and judgment of the past, was considered to be inadequate for the elderly with higher brain dysfunction. The SNAQ-JE ultimately consisted of 4 items (#1, #2, #4, and #8) (eTable 4).

Comparison of the results of exploratory factor analyses
The factor loadings of item #6 were 0.257 and 0.279 for the CNAQ-J and SNAQ-J, respectively, and the factor loading values for the SNAQ-JE were all >0.5 ( Table 3). The explained variances for respective factor 1 were 28.22, 33.66, and 37.70 for the CNAQ-J, SNAQ-J, and SNAQ-JE, in that order.

Fitting to the model
The GFI and AGFI values for the CNAQ-J, SNAQ-J, and SNAQ-JE were all >0.9, exhibiting a good fit to the model using confirmatory factor analyses ( Table 4). The RMSEAs of the CNAQ-J, SNAQ-J, and SNAQ-JE were 0.063, <0.001, and 0.085, respectively.
Cronbach's a coefficients by sex and study group for the SNAQ-JE were uniformly greater than those of the SNAQ-J.

Discussion
We translated and back-translated the CNAQ into Japanese (the CNAQ-J), along with its simplified version (the SNAQ-J). An exploratory factor analysis demonstrated that the CNAQ-J was constructed of one factor (appetite). According to a step-by-step exploratory factor analysis, we developed the SNAQ-JE, which uses four questions derived from the CNAQ-J. The CNAQ-J, SNAQ-J, and SNAQ-JE showed similar fitting to the model and testeretest reproducibility. The CNAQ-J and SNAQ-JE demonstrated satisfactory reliability and significant criterion-referenced/-related validity values with BMIs, whereas the SNAQ-J manifested less satisfactory reliability, and non-significant criterion-referenced/-related validity with BMIs or MNA-SF values. Additionally, the SNAQ-J included item #6, which had a factor loading of only 0.279. Thus, the SNAQ-JE, instead of the SNAQ-J, may be better suited for use among Japanese elderly.
The SNAQ-JE reached a satisfactory level of Cronbach's a coefficient (0.700) in the present study, but the SNAQ-J did not (0.640).
The Cronbach's a coefficient of the SNAQ-J in the present study was comparable to figures for a long-term care group in the original SNAQ study (0.51), 14 geriatric patients in Malaysia (0.58), Brazilian participants in the Cardiopulmonary and Metabolic Rehabilitation Program (0.61), and community-dwelling Japanese elderly (0.55). 25e27 The lower internal consistency values may be due to the fact that most replies to item #6 were aggregated to "three meals a day" (>80% of responses in the present study) with smaller variance, reflecting in part its low factor loading value (<0.4); indeed, the reliability figure improved when the item was excluded. 26 LOA unduly drives down meal amount, causing weight loss and frailty in the elderly. Wilson et al 14 reported that a CNAQ score 28 (the SNAQ score 14) may predict being "at risk" of a 5%e10% weight reduction with approximately 80% sensitivity and specificity. Due to the limited observation period and research scheme, we were unable to assess sensitivity and specificity of the SNAQ-JE. Instead, using the SNAQ-JE average value (mean) of 14.4, we observed significantly reduced BMIs among the lower SNAQ-JE score ( 14) group compared with those of the higher-score (!15) group, but not for the SNAQ-J, in a cross-sectional analysis. Thus, the scores 14 and ! 15 could be used for the SNAQ-JE cutoff values.    As noted, the lower-score groups had significantly lower BMI values compared with respective higher-score groups on the CNAQ-J and SNAQ-JE, but not on the SNAQ-J. We observed that the three appetite battery scores were negatively associated with the CAS-J values, with statistical significance. The SNAQ-J and SNAQ-JE scores were significantly correlated with MNA-SF values (with age-and gender-adjusted Pearson correlation coefficients of 0.124 and 0.178, respectively), although both correlation coefficients were <0.3 observed in Japanese elderly people. 27 The SNAQ-JE acceptability as a screening instrument should be further investigated to verify the test performance (sensitivity, specificity, and receiver-operating characteristic [ROC] curve analyses) in comparison with nutrition-related questionnaires, including the Malnutrition Universal Screening Tool (MUST) and the Seniors in the Community: Risk Evaluation for Eating and Nutrition (SCREEN) 14,28,29 in addition to the MNA-SF.
Wilson et al 14 developed the SNAQ by deleting "reliability reducers" (supported by a principal components analysis) from the CNAQ, but the authors included item #6, which had a skewed distribution and a factor loading <0.4 in the present study. We developed the SNAQ-JE using a series of exploratory factor analyses, deleting the items showing a ceiling effect, having a smaller factor loading, being an internal consistency reducer, and requiring the use of a higher brain function. Ultimately, fitting to the model, associations with BMIs and CAS-J scores, and reproducibility values were almost equivalent between the SNAQ-J and SNAQ-JE, but the SNAQ-JE reliability and criterion-referenced/-related validity compared to MNA-SF values were uniformly and consistently (significantly/non-significantly) more favorable than those of the SNAQ-J, which included the low-factor-loading item #6. Thus, the evidence-based SNAQ-JE (including items #1, #2, #4, and #8) may be adopted to assess the appetite of other ethnic elderly people worldwide instead of the original SNAQ, although the instrument was primarily developed for the Japanese elderly.
In view of public health nutrition, it appears critical to manage the appetite of elderly people to detect LOA and to prevent LOArelated disorders/syndromes in the early phase using pertinent appetite questionnaires. Wilson et al 14 conducted surveys on a wide range of people (aged >20e102 years) but excluded persons having moderate-to-severe dementia (Mini-Mental State Examination [MMSE] score <18), because their instruments were selfadministered by the subjects. However, biases may still exist when obtaining information from the elderly with impaired cognitive function and obtaining replies with proxy/surrogate interview-based assistance. Question items should be a priori confined to basic physiologic perception related to an appetite, excluding items requiring deployment of a higher brain function. Meanwhile, reports have demonstrated that the proxy/surrogate respondent's information on dementia/Alzheimer's disease was satisfactorily valid and reliable, without systematic biases. 30e32 Such appeared also to be the case in the present study: the CNAQ-J scores of the MD, DC, and GH groups with proxy (caregivers/family members) or surrogate (facility staff) aids were not uniformly greater/lower than those of CE group self-administered without assistance, suggesting that relevant proxy/surrogate supports may serve to enhance validity and reproducibility.
There are some limitations in the present study. As discussed above, the participants were comprised of heterogeneous Japanese elderly with respect to cognitive function, as seen in the proportions of participants with no problems (22.7%), suspected dementia (19.6%), and apparent dementia (57.7%) in the CDR analysis. These observations reflect the real-world evidence that some proportion of elderly people have a certain degree of higher brain dysfunction. However, under the present study scheme, we were unable to incisively relate cognitive levels to appetite questionnaire scores by gender, age, and instrument administration method (selfadministered vs. interview-based with proxy/surrogate assistance). These issues of interest warrant further clarification using a specific research protocol.
Because we here investigated possible associations of appetite tool scores with BMIs and with nutrition-related questionnaire indices in a cross-sectional setting where causes and effects coexist at a certain point of time, we should be deliberate to draw causal inferences. Under a long-term prospective approach, with a !6month observation period and sensitivity, specificity, and ROC curve analyses, we could precisely determine cutoff values, verify possible factors related to LOA, quantify its long-term effect on weight changes, and detect weight loss-associated disorders/syndromes at an early stage.
In conclusion, we developed the Japanese versions of appetite instruments: the CNAQ-J, SNAQ-J, and SNAQ-JE. Of these, the CNAQ-J and SNAQ-JE could be applied to Japanese elderly people, including those who have some cognitive impairment. Because no appreciable discrepancies were noted in reliability, validity, or reproducibility values between the two appetite tools, the SNAQ-JE appeared more feasible and practicable than the CNAQ-J for assessing appetite among the elderly, predicting declining body weight, and screening for LOA-related diseases/syndromes in the premature phase. Thus, the present data-based appetite questionnaires may contribute to elderly people's quality of life and wellbeing.