Optimization The Nutritional Administration For Critical Illness Patients

Objective:based on real-world clinical research methods,accurately determine the optimal energy and protein intake of critically ill patients by combining classical statistics and machine learning techniques so as to reduce the adverse outcomes caused by too little or too much energy and protein intake.Methods:We conducted an observational study involving critically ill patients who had received parenteral and enteral nutrition support in the emergency intensive care unit of Sichuan Provincial People’s Hospital since September 2018.General information,assessment,diagnosis,clinical outcome and vital signs,laboratory examination and nutritional support treatment plan for 14 consecutive days were recorded for the critically ill patients.The variables of energy,protein intake,diagnosis and treatment of patients in the survival cohort and the death cohort were compared respectively.Firstly,exploratory statistical analysis was conducted to find the factors that had significant influence on clinical outcomes.Thus representative date sets was constructed in order to assessing variation of the patients.Interpolation method is applied to supply missing data,and modeling for pattern recognition based on machine learning methods.The variables of the model include daily energy intake,protein intake and factors may affect the clinical prognostic.Then,we found that under line of different illness condition(critical)makes the lowest fatality rate of energy and protein intake.This study was reviewed by the medical ethics committee of Sichuan Provincial People’s Hospital,and the clinical trial was registered in the Chinese Clinical Trial Registry(clinical registration number:ChiCTR1900024746).This study is an observational study,which will not interfere with the patient’s treatment plan,thus the ethics committee agreed to exempt informed consent.Results:A total of 105 patients were enrolled in the first round which carried out until December 2019.44 was assigned to the death cohort(group D)and 61 to the survival cohort(group S).The ratio of male patients in the two groups was similar,but patients in the survival cohort were younger than those in the death cohort.Fewer patients were on ventilators in survival cohort in comparison with the death cohort,and the nutritional risk and severity scores of patients in the death cohort were significantly higher than those in the survival cohort.For patients with relatively high scores,energy intake and protein intake in the survival cohort was higher than in the death cohort.In the machine learning database,each patient included 869 influencing factors,so a data matrix of 105*869 was obtained.In selected factors after received the greatest influence on prognosis of patients with 105*77 data matrix,by using partial least squares discriminant analysis method to classify,based on the weight of the first principal component load diagram affect the prognosis of patients with model:PC1＝pulse pressureD 13*0.244+blood albuminD8*0.171+blood albuminD2*0.146+blood albuminD14*0.049+energy intakeD11*0.147+energy intakeD8*0.073+energy intakeD10*0.061+energy intakeD7*0.024(formula 3);PCl=pulse pressureD13*0.346+pulse pressureD9*0.214+pulse pressureD6*0.084+blood albuminD7*0.218+portein intakeD13*0.026+portein intakeD14*0.018+portein intake D12*0.007(formula 4)。Conclusions:Patients with higher nutritional risk,higher organ failure index,and higher severity of the disease were able to benefit from higher energy(25kcal/kg·d)and protein(1.2g/kg·d)intake.Real world clinical study involved a large quantity of factors which may affect the state of disease progression or the therapeutic effect of critically ill patients.Thus we screened out various important factors by statistical analysis to explore the best nutritional intake,combined with previous contrast is different from the critically ill patients after the gap between energy and protein intake.This project is an early application of the combination of real-world clinical research methods and machine learning model to explore the optimal nutrition support treatment scheme for critically ill patients.Based on this,an auxiliary decision making system that can guide the selection of clinical nutrition treatment scheme is established.Due to the limitation of sample size,we have only reached a preliminary conclusion as yet.Next,as the sample size increases,the data matrix will be richer.Meanwhile,we will continue to optimize the algorithm to iterate the model,and finally make the energy and protein intake program of critically ill patients individual and precise.