Nutritional knowledge of primary school teachers in association with their participation in health education programs

Abstract

faced by traditional statistical methods when studying diet-disease associations, as they are robust to multicollinearity, do not make distributional assumptions and consider all possible interactions between explanatory variables. Aim: Το explore dietary-lifestyle factors and multilevel interactions of them as predictors of cognitive performance using CART. Materials and methods: Until September 2014, 1429 participants (41% male, median age 72 years) of the Hellenic Longitudinal Investigation of Aging and Diet (HELIAD) had completed a full baseline assessment via structured questionnaires of diet and lifestyle, medical history, memory and other cognitive problems. We performed CART analysis using percent Mini-Mental State Examination (MMSE %) score as the dependent variable in addition to 26 candidate explanatory variables. We tested 12 different combinations of minimum node size and minimum impurity improvement. Results: For 8 out of 12 configurations, age, education and Geriatric Depression Scale (GDS) were implicated in node splits of the corresponding trees. Age interacted with education in 2 configurations and with GDS in a third one. No dietary or physical activity variable emerged as a predictor in any regression tree. Conclusions: Using CART, we observed no association between cognitive function and dietary or other lifestyle factors in the elderly. However, advances in the field of decision trees (e.g. random forests) and machine learning (e.g. neural networks) require further investigation with respect to their application in the field of nutrition research.