Development of predictive models for nutritional assessment in the elderly.

Objective: To propose malnutrition screening methods for the elderly population using predictive multivariate models. Due to the greater risk of nutrition deficiencies in ageing populations, nutritional assessment of the elderly is necessary in primary health care.
Design: This was a cross-sectional study. Multivariate models were obtained by means of discriminant analysis and binary logistic regression. The diagnostic accuracy of each multivariate model was determined and compared with the Chang method based on receiver operating characteristic curves. The optimal cut-point, sensitivity, specificity and Youden index were estimated for each of the models.
Setting: The province of Cordoba, Spain.
Participants: Two hundred fifty-five patients over the age of 65 years from three health centres and three nursing homes.
Results: Fourteen models for predicting risk of malnutrition were obtained, six by discriminant multivariate analysis and eight by binary logistic regression. Sensitivity ranged from 55·6 to 93·1 % and specificity from 64·9 to 94 %. The maximum and minimum Youden indexes were 0·77 and 0·49, respectively. We finally selected a model which does not require a blood test.
Conclusions: The proposed models simplify nutritional assessment in the elderly and, except for number 2 of those calculated by binary logistic regression, have better diagnostic accuracy than the Spanish version of the Mini Nutritional Assessment screening tool. The selected model, whose validation is necessary for the future with other different samples, provides good diagnostic accuracy, and it can be performed by non-medical personnel, making it an accessible, easy and rapid tool in daily clinical practice.