Analysis of Quantitative Relationship between Dietary Characteristics and Health Indicators of Long-lived Population by Random Forest Regression Algorithm

To explore the potential quantitative relationship between dietary characteristics and physical health indicators of the long-lived population in Guangxi, the longevous in Guangxi (Fengshan County, Donglan County, Shanglin County, Dahua County, and Cenxi City) and some of their descendants were recruited from 2016 to 2018. A semi-quantitative food frequency questionnaire (FFQ) was used to investigate their diet, routine physical examination indicators and blood sample analysis data of the long-lived population were collected, and the weight of the corresponding indicators obtained was calculated by the entropy weight method. Health composite index (HCI) was further introduced to conduct a comprehensive evaluation of health indicators, and stochastic forest regression algorithm was used to model the relationship between volunteers' food and nutrient intake and HCI, to explore the quantitative relationship between each parameter. The results of the model were evaluated by ‘Accuracy’. The results showed that BMI, hemoglobin, triglyceride and C-reactive protein were significantly different in different age groups. There were significant differences in food component intake and nutrient intake among different age groups (P<0.001). The centenarian group had the lowest energy intake, with a mean of 1451.43 kcal. The centenarian group had the largest median HCI (0.676) and distribution were relatively concentrated. Among the 8 models established by random forest regression algorithm, the regression accuracy of food and nutrient intake to HCI of the centenarian group was the highest (93.37% and 93.40%, respectively). Among them, high fish and shrimp food intake had a greater impact on HCI in the centenarian group (0.41). Low red meat food intake had a higher impact intensity of 0.20 on HCI in the 90s group, which also quantitatively reflected the impact of aquatic products and red meat products on the health status of the elderly. Among the samples involved in this study, the effect intensity of energy and protein intake on HCI was low (effect intensity<0.12). Vitamin A intake had a higher impact intensity on HCI (effect intensity≥0.17). This study constructed a random forest regression model between the dietary characteristics and health indicators of the long-lived population. The model found that food intake of more fish and shrimp, less red meat, high vitamin A and dietary fiber intake were unique dietary characteristics of the centenarian group and 90s group. Indicating the two groups' HCI were influenced positively, which would provide a reference for improving human health through diet. Simultaneously, the quantitative relationship between diet and health indicators clarified by the model makes it possible to predict the health status of volunteers according to dietary characteristics.