COMPARATIVE STRATEGIES FOR USING CLUSTER ANALYSIS TO ASSESS DIETARY PATTERNS

Authors: BAILEY, REGAN - PENN STATE UNVI; DAVIS, MELISSA - PENN STATE UNIV; MITCHELL, DIANE - PENN STATE UNIV; MILLER, CARLA - PENN STATE UNIV; LAWRENCE, FRANK - PENN STATE UNIV; SMICIKLAS-WRIGHT, HELEN - PENN STATE UNIV

Submitted to: Journal Of The American Dietetic Association
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/1/2005
Publication Date: 8/1/2006
Citation: Bailey, R.K., Davis, M.S., Mitchell, D.C., Miller, C.K., Lawrence, F.R., Smiciklas-Wright, H. 2006. Comparative strategies for using cluster analysis to assess dietary patterns. Journal Of The American Dietetic Association. 106(8):1194-1200.

Interpretive Summary: Dietary patterns reflect whole foods and/or combinations of consumption, temporal distribution of intake and habitual patterns (e.g. snacking and food preparation methods). Examination of the totality of dietary patterns provides a more accurate description of actual dietary exposure. Dietary pattern analysis is an ideal tool to identify those who may be at nutritional risk for appropriate intervention. We examined dietary patterns of older adults using 2 different cluster analysis strategies: by number of servings (SVGS) and % energy contribution (KCAL) from food subgroups. Data were part of a cohort from the Geisinger Rural Aging Study (GRAS), a longitudinal study of older adults in rural Pennsylvania. Demographic, health, and anthropometric data were collected via home visit; dietary intake was assessed by five 24-hour recalls collected over 10 months. All foods were classified into 24 food subgroups. The methods differed in the food subgroups that “clustered” together. Both methods produced clusters that had significant differences in Healthy Eating Index (HEI) scores, a measure of diet quality. The clusters with higher HEI scores contained significantly greater amounts of most micronutrients. Waist circumference was significantly lower in the cluster with higher HEI scores using the SVGS method but not with the KCAL method. Both methods were able to cluster food subgroups with high energy contribution (e.g., fats and oils, processed meat and dairy desserts). KCAL was not sensitive to food subgroups with minimal energy contribution (e.g., fruits and vegetables). Clusters resulting from the percent energy method were less likely to differentiate fruit and vegetable subgroups. Older adults in the healthier dietary pattern derived from the number of servings method had more favorable weight status The two methods employed yielded different solutions; the commonality of the two methods provides specific food targets for nutrition intervention.

Technical Abstract: The objective of this study was to characterize dietary patterns using two different cluster analysis strategies. In this cross-sectional study, diet information was assessed by five 24-hour recalls collected over 10 months. All foods were classified into 24 food subgroups. Demographic, health, and anthropometric data were collected via home visit. One hundred seventy-nine community-dwelling adults, aged 66 to 87 years, in rural Pennsylvania. Cluster analysis was performed. The methods differed in the food subgroups that clustered together. Both methods produced clusters that had significant differences in overall diet quality as assessed by Healthy Eating Index (HEI) scores. The clusters with higher HEI scores contained significantly higher amounts of most micronutrients. Both methods consistently clustered subgroups with high energy contribution (eg, fats and oils and dairy desserts) with a lower HEI score. Clusters resulting from the percent energy method were less likely to differentiate fruit and vegetable subgroups. The higher diet quality dietary pattern derived from the number of servings method resulted in more favorable weight status. Cluster analysis of food subgroups using two different methods on the same data yielded similarities and dissimilarities in dietary patterns. Dietary patterns characterized by the number of servings method of analysis provided stronger association with weight status and was more sensitive to fruit and vegetable intake with regard to a more healthful dietary pattern within this sample. Public health recommendations should evaluate the methodology used to derive dietary patterns.