Research And Development Of Intelligent Prediction System For Red Blood Cell Demand In Trauma Patients

OBJECTIVE To study the pre-transfusion needs assessment of emergency trauma patients and the post-transfusion efficacy evaluation auxiliary diagnosis and treatment software system(the mathematical model of red blood cell transfusion),combined with non-invasive monitoring technology,to develop a portable and intelligent blood transfusion diagnosis and treatment auxiliary device to provide accurate treatment for traumatic patients with bleeding.Quick blood transfusion decision-making opinions.Significance In the emergency rescue of trauma patients,due to the lack of testing aids,doctors do not have enough time to formulate blood transfusion treatment programs,and only rely on descriptive examination results,blood transfusion protocols based on personal experience alone are not accurate,and it is impossible to timely evaluate the blood transfusion effect and affect the success rate of transfusion.The application of the blood transfusion aided decision-making system can accurately formulate individualized blood transfusion treatment programs,effectively reduce the mortality rate of trauma patients and improve the success rate of trauma treatmentPart 1 Abstract Background The paucity of accurate quantitative standards for determining the quantity of red blood cell(RBC)needed for perioperative patients and the predominant application of the empirical decision-making model of "hemoglobin+clinical symptoms"have led to the widespread problem of inadequate or excessive RBC preparation.This not only increases patient risk and medical costs,but also intensifies the problem of blood shortage.Objective In this artificial intelligent(A1)-based study,we analyzed multiple parameters that may affect the volume of RBC transfusion using big data,and established a mathematical model for forecasting the quantity of RBC needed during the perioperative period.Methods The clinical data of 130,996 patients who underwent surgery in our hospital between January 2011 through June 2017 were reviewed.The A1 prediction model of RBC demand was established using the boosting trees algorithm,in which general physiological indices,laboratory tests and surgical information of the patients were used as the variables.We used the model to predict "whether the above patients need RBC transfusion" and "quantity of RBC needed" and compared the results with those predicted by the clinicians.Results The A1 model generated a significantly more accurate(85.6%vs.45.4%)and precise prediction of the need for blood transfusion than the clinicians(85%vs.77%).Furthermore,the A1 model gave more accurate prediction of the quantity of blood transfusion for patients than the clinicians(RMSE 1.758 vs.1.97).Conclusions The boosting trees algorithm-based model is more accurate than the clinician experience-based model in the prediction of perioperative RBC transfusion,and can thereby provide better support to surgeons on the decision of RBC preparation.Part 2 Objective This study conducted a comprehensive and systematic evaluation of previous traumatic emergency cases,and analyzed the correlation between red blood cell transfusion volume and type of trauma,vital signs of patients,and laboratory test indicators.According to the calculation of big data,a multi-parameter mathematical model for pre-transfusion needs assessment and post-transfusion efficacy evaluation of trauma patients was constructed.Quickly draw decision-making opinions on blood transfusion of recipients and provide reference for evaluation of efficacy after transfusion.Methods A retrospective analysis of 1667 patients with "injury" in the emergency department of our hospital during the period of 2015.4-2018.3 was performed.Using the patient's vital signs,laboratory tests,admissions and outpatient conditions,blood transfusions,etc.as variables,the boosting trees algorithm was used to construct an intelligent prediction model of red blood cell demand,and the model was used to "require red blood cell infusion" for the above patients.And "red blood cell infusion demand" is predicted and the data of the test set is compared with the actual blood transfusion value.Results The area under the ROC curve was 0.80 for whether or not the blood transfusion was predicted.The demand for blood transfusion patients is predicted to be MAE=1.60.Conclusion The red blood cell transfusion prediction model based on boosting trees algorithm provides technical support for doctors to make rapid and accurate blood transfusion decision in emergency rescue environment.Part 3 Objective To develop a non-invasive,portable multi-wavelength human body hemoglobin detector and verify the practicability and accuracy of the instrument.Methods The photoelectric volume pulse wave signal collected by photoelectric sensor was used to analyze the change of human physiological parameter-hemoglobin by the change of absorbance,and 234 cases were collected from blood donors and patients in our hospital.Compared with routine laboratory invasive method and this study,Non-invasive instruments developed to detect the accuracy of hemoglobin.Results The routine laboratory invasive method and the non-invasive instrument developed by this study detected the hemoglobin correlation coefficient R of 0.935,which has obvious correlation.The root mean square error RMSE is 0.8563.Conclusion The human body non-invasive and portable multi-wavelength physiological hemoglobin detector developed in this study can quickly and non-invasively detect hemoglobin with good practicability.Compared with conventional laboratory invasive methods,it has high accuracy and good consistency.