Cluster Analysis As An Approach For Evaluating Nutritional Risk - Brief Article

Nutrition Research Newsletter, April, 2001

Since diet has been associated with the prevention of six of the nation's ten leading causes of death, more effective intervention strategies to promote optimal food and nutrient intake is imperative. The multiple patterns of food consumption in this country suggest that population-specific variations in dietary patterns be considered during policy development and nutrition interventions. Despite this recognition, the conventional approach in nutrition epidemiology has not emphasized defining distinct patterns of dietary behavior within a risk or health status. Therefore, a recent article in JADA validated the use of cluster analysis for characterizing population dietary patterns. Cluster analysis can be used to compare relationships among patterns of food consumption, nutrient intake, and health risk profiles.

Subjects were taken from the Framingham Offspring-Spouse (FOS) study. The data reported here was collected among women at Exam 3 of the FOS cohort during 1984-1988. The women were asked to complete a food frequency questionnaire, where the food items were classified into 42 categories based on similarities in nutrient content. Next, the 42 food categories were clustered into 13 food groups according to similarities in frequency of consumption. Independent estimates of nutrient intake were obtained from three-day food diaries. Heart disease risk factors were assessed. BMI was calculated and total cholesterol levels and blood pressure was determined.

The cluster analysis determined five distinct dietary patterns among the FOS women. The patterns were Heart Healthy, Light Eating, Wine and Moderate Eating, High Fat, and Empty Calorie. The Heart Healthy cluster was characterized by higher consumption of foods that are typically recommended for health promotion, while the Light Eating cluster was distinguished from the others by lower intakes of certain food groups, because of their overall low food intake. Aside from a relatively higher consumption of wine, women in the Wine and Moderate Eating cluster practiced moderation by consuming less hidden fats and sweets. Women in the High Fat cluster chose more servings of foods high in fat, while women in the Empty Calorie cluster consumed more desserts and sweets. The Heart Healthy women were significantly older and most likely to be non-smokers. Among women in the Light Eating cluster, only half or fewer had desirable lipid levels. Women in the Wine and Moderate Eating group were most likely to be normal weight and have desirable LDL cholesterol levels, but were least likely to be normotensive.

Millen et al. state that that dietary patterns based upon differences in food consumption were associated with distinct and consistent differences in nutrient intake profiles. It does appear that cluster analysis is a valid tool for evaluating nutrition risk. However, this approach should not be considered fool proof and the limitations should be carefully thought about. It appears that the women in the Heart Healthy cluster had previous risk factors and probably adopted more healthful eating habits as a result of certain recommendations due to these risk factors. Therefore, these cross-sectional analyses may not depict the true relationship between improved eating habits and better risk factor profiles. Longitudinal studies should be used to clarify these relationships.

Barbara E. Millen, Paula A. Quatromoni, Donna L. Copenhafer, et al., Validation of a Dietary Pattern Approach for Evaluating Nutritional Risk: The Framingham Nutrition Studies, JADA 101(2): 187-194 (February 2001) [Address correspondence to: Barbara A. Millen, DrPH, RD, Department of Social and Behavioral Sciences, Boston University School of Public Health, 715 Albany St, Boston, MA 02118].