Research On The Establishment Of Refined Nutritional Support Strategies For Major Abdominal Surgery Based On Machine Learning

Objective:Malnutrition could increase the rate of postoperative complications,while the status and outcomes of can be improved by the reasonable nutritional support.However,the current nutritional prescription is based on clinical experience,and it is difficult to achieve individual treatment.So we are planning to establish a refined nutritional support strategy for surgical abdominal surgery through machine learning.Materials and Methods:We planed to perform a observational study among 120 persons who received abdominal surgeries.Nutritional assessments were performed before surgeries,and blood samples were obtained for the testing of metabolites and inflammatory factors.The postoperative individual nutritional prescription were prescribed by the trained doctors according to the current guidelines.Then we used machine learning to establish the support vector machine(SVM)model of the first,the second,and the third day after surgery based on the clinical outcomes,and the inverse algorithm were used to find the everyone' s individual optimal nutrition strategy.Finally,we tried to establish a specification and a standard of nutrition therapeutic efficacy by using the changes of metabolites in patients at different times obtained by the metabolomics technology.Results:Finally,there were 89 patients who met the included criterion put into statistical analysis.In Logistic regression analysis,age,disease classification,and whether received an open surgery were main risk factors for complications.The hospitalization cost and time in patients of malnutrition are both higher than those who are well-nourished;the patients with complication,compared with those without complication,have a higher concentration of MDA,and there is on significant difference in IL-1a?MDA and IL-6 in the two groups after the surgery as well as the difference of the two groups between the first day and the fifth after surgery.We can find out the range of energy and protein for the best clinical outcome at different time after surgery by machine learning in the future;The best energy range for patients in the first day of postoperation is 5-10kcal/kg/d,and the optimal protein range for patients is 0.2-0.5g/kg/d.While in the second day of postoperation,the best energy range is 25-30kca1/day/d,and the optimal range of protein is 0.5-1g/kg/d.In the third day of postoperation,the best range of energy is 27-32kca1/kg/d,and the optimal range of protein is 0.8-1.lg/kg/d.Based on 1H-NMR metabolomics technology,we can clearly identify the difference in metabolites at different time points before and after surgery.In the end,we screened 26 biomarkers to identify preoperative and postoperative day 1 as well as 43 biomarkers to identify postoperative day 1 and postoperative day 3.And the results show that there were mainly differences in relative produces of glucose metabolism on postoperative day 1,which indicates that there may be an increase of gluconeogenesis on postoperative day 1,meanwhile the differences in produces of protein metabolism imply that there may be an increase of proteolysis on postoperative day 3.Conclusion:Establishing model by machine learning can help to determine the optimal range of energy and protein at different time after surgery,which could provide the methodological assurance for the appropriate individual treatment.