RESUMEN
Preventing lifestyle-related diseases requires understanding and managing the intake of total fats and specific types of fatty acids, especially trans fatty acids. There are several methods for measuring fat intake, each with its own strengths and limitations. Guidelines for nutritional epidemiology studies recommend employing objective biomarkers. This study aimed to estimate fatty acid intake based on serum fatty acid levels using multiple regression analysis and a machine learning technique, and to compare their accuracy. The subjects were healthy women aged 18 to 64 living in Toyama, Japan. A dietary survey to determine fatty acid intake was conducted using a 3-day dietary record completed by the participant. Blood samples were collected after an overnight fast, and serum was obtained through centrifugation. A total of 300 women participated in the study. The fatty acid levels in serum were determined using gas chromatography with a capillary column. Using multiple regression analysis and neural networks, the intakes of saturated, monounsaturated, n-6 polyunsaturated, n-3 polyunsaturated, and trans fatty acids from serum fatty acid levels were predicted. Significant correlations were observed between the intakes of the five classified fatty acids and the predicted intakes obtained from the multiple regression analysis (r = 0.39 - 0.49, p < 0.01). Significant correlations were also observed between the five classified fatty acid intakes and the intakes predicted by the neural network (r = 0.52 - 0.79, p < 0.01), and the correlation coefficient showed a significantly higher value than that predicted by the multiple regression analysis. These results suggest that serum fatty acid levels may be used as biomarkers to estimate the intake of fatty acids, including that of trans fatty acids, and that machine learning may be able to predict fatty acid intake with higher accuracy than multiple regression analysis.
