Development Of A Clinical Prediction Model Combining Traditional Chinese And Western Medicine For In-hospital Death In Heart Failur

Purpose:This study describes the Chinese and Western medical characteristics of heart failure inpatients from the Affiliated Hospital of Liaoning University of Chinese Medicine;and then the risk factors of in-hospital mortality in heart failure patients were explored and a risk score scale was developed for in-hospital mortality in heart failure patients by applying logistic regression methods based on the above risk factors;finally we use machine learning methods of RUST Boost Trees algorithm to construct a prediction model to explore the the influence of TCM factors on the predictive ability of the model.With the risk score scale and mechine learning prediction model,we could assist physicians to identify patients with a heightened risk of mortality.Material and method:This study is carried out layer by layer through three parts.Firstly,baseline data,physical,physical and chemical inspection,medical history,tongue and pulse,Chinese medical evidence,Chinese season,and birth date were statistically compiled for all heart failure patients hospitalized in the Department of Cardiology in the Affiliated Hospital of Liaoning University of Chinese Medicine from January 1,2019 to December 31,2020 to derive the Chinese and Western medical characteristics of heart failure hospitalized patients.Secondly,1,068 patients discharged from January 1,2019 to December 31,2019 were used as the test set.Logistic regression and Jorden index were applied to develop a risk predictive score of in-hospital death and the cut-off points to distinguish low-risk and high-risk patients was 13.And then 508 patients discharged from January 1,2020 to December 31,2020 were used as the validation set to evaluate the predictive and calibration ability of the score table and column line graphs.Thirdly,we tried to apply machine learning method to detect a predictive model.Patients from January 1,2019 to December 31,2020 were divided as the test set and validation set by appling the five-fold cross-validation method.And 25 machine learning algorithms were used to derive a model of best predictive power,and after deriving the optimal algorithm and apply the optimal algorithm to construct the model.Results:In this study of 1676 heart failure inpatient records,the median age was 75(66,84),and 51.3% were male.Phlegm and blood stasis evidence of phlegm and blood deficiency confirmed the most common TCM evidence of heart failure inpatients in our hospital.Blood stasis,qi deficiency,phlegm,water-drinking,yang deficiency and yin deficiency were the most interval TCM pathology-related evidence in hospitalized heart failure patients.Red tongue,dark tongue,white and greasy coating are common tongue signs in heart failure patients;thin pulse,string pulse,sunken pulse and slippery pulse are common veins in hospitalized heart failure patients.There were significant differences in gender composition,age,Cl ion,albumin,creatinine,urea nitrogen,total bilirubin,and leukocyte levels among patients hospitalized with heart failure of different evidence types.The highest incidence of heart failure patients in Shenyang area was in the order of cold dew > summer > winter >snow.By logistic regression method,admission heart rate,admission diastolic blood pressure,carbon dioxide binding capacity,albumin level,total leukocyte count,NT-pro BNP,BNP,and sunken pulse were found to be independent risk factors for in-hospital mortality of patients,and a risk score scale was derived based on the results of multiple logistic regression with AUC=0.871(95% CI 0.823,0.920)for the score scale,applying the Jorden index to derive a score scale yielding 13 points to distinguish between low and high risk of in-hospital mortality,and the score scale was found to have excellent predictive and calibration power when validated in the validation set.Finally,among 25 machine learning algorithms,model conducted by RUST Boost Trees algorithm has the best predictive power.Conclusion:The age of our inpatients with heart failure was higher than the average in China.The proportion of males in the patient population was higher than that of females,and there were significant differences in certain tests among patients with different types of disease.The independent risk of in-hospital death in patients with heart failure is very different from those in foreign countries.The scoring scale developed in this study has good predictive ability for whether in-hospital death occurs in our patients with heart failure,and a score of 13 points can significantly distinguish between low-risk and high-risk patients.The machine learning algorithm found that the prediction model could improve the model of great predictive ability,sensitivity,specificity and distinguishing ability.