Prediction Model Of Risk For Postoperative Cognitive Dysfunction After Total Knee Replacement Based On Machine Learning Algorithm

Objective:postoperative cognitive dysfunction（POCD）refers to the changes in cognitive function such as inattention and reduced memory ability after anesthesia or surgery.In severe cases,there is an acute,reversible psychotic disorder with personality changes and a decline in social behavior.The specific clinical manifestations of POCD include memory decline,impairment of understanding ability and language expression ability,difficulty in concentrating attention,personality and behavior change,etc.As one of the common complications of the central nervous system after surgery,POCD can prolong the hospital stay of patients,increase hospital costs and mortality,which has a great impact on the postoperative recovery of patients,but also seriously threatens the life safety of patients.At the same time,it causes great economic burden and mental pressure to patients and their families.Studies have shown that the incidence of POCD in elderly patients undergoing surgery is as high as 45%～60%.However,the pathogenesis of POCD remains unclear.In this study,a Bayesian network（BN）algorithm was used to establish a risk prediction model for POCD after total knee replacement（TKR）,and its predictive performance was investigated.At the same time,several machine learning algorithms commonly used in the medical field,including extreme gradient Boost（XGBoost）,Random forest（RF）,support vector machine（SVM）and multi-layer perceptron（MLP）,were used to model the data.The prediction effect of other machine learning algorithms in the data set of this study is verified.In the real world research and application process,BN can be complementary to provide more reference for clinicians and nursing.Methods:1.A case-control study design was used to select 1 260 inpatients who underwent TKR from January 2017 to December 2021 in the Department of Joint Surgery of the First Affiliated Hospital of the Army Medical University.The main diagnosis of participants was severe osteoarthritis of left/right knee joint,of which 240 cases were male（19.0%）and 1020cases were female（81.0%）.The average age was（66.73±8.46）years,ranging from 23 to 79years.The mean body mass index（BMI）was（24.75±5.75）kg/m².Patients with POCD（71cases）after surgery（from the end of surgery to discharge）were randomly divided into A1group（70%）and B1 group（30%）according to 7︰3,and patients without POCD（1 189 cases）were randomly divided into A2 group（70%）and B2 group（30%）according to 7︰3.Group A1 and A2 together form group A（training data set）,group B1 and B2 form group B（test data set）,in which group A is used for model training and group B is used for model testing.Thirty-six indexes related to perioperative anesthesia decision,disease outcome and length of stay in TKR were selected as nodes,and the probability distribution model diagram of each node was established by using BN algorithm to predict the probability of risk for POCD,so as to minimize the length of hospital stay and promote the maximum recovery of patients.2.The final data included in the study of the risk prediction model of cognitive dysfunction after total knee replacement based on MLP included the medical data of 1260patients with TKR.Patients with POCD after TKR（71 cases）were randomly divided into group A1（70%）and group B1（30%）according to 7:3,and patients without POCD（1 189cases）were randomly divided into group A2（70%）and group B2（30%）according to 7:3.Group A1 and A2 together form group A（training data set）,group B1 and B2 are group B（test data set）,in which group A is used for model training and group B is used for model testing.SPSS 23.0 was used to statistically analyze the gender distribution,age,ASA grading,BMI and other patient data of the training set and test set.The normal distribution measurement data were represented by mean and standard deviation,the non-normal distribution measurement data were represented by M(P₂₅,P₇₅),and the count data were represented by cases（%）.χ²test,t test,Fisher exact test and Kruskal-Wallis test were used for comparison among groups.All statistical tests were bilateral,and P<0.05 was considered statistically significant.The prediction model of POCD risk after TKR operation was established based on MLP algorithm,and the probability of POCD risk was predicted.At the same time,several commonly used machine learning algorithms in the medical field,including extreme gradient Boost（XGBoost）,Random forest（RF）,and support vector machine（SVM）,were used to model the data,and the prediction effect of other machine learning algorithms in the data set of this study was verified.Results:1.The final number of patients included in the trial was 1260,including 240 males（19.0%）and 1020 females（81.0%）.The average age was（66.73±8.46）years,ranging from 23to 79 years.The average BMI was（25.08±5.09）kg/m2;108 cases（8.6%）were classified as ASA GradeⅠ,and 1152 cases（91.4%）were classified as ASA gradeⅡ.There was no significant difference in gender distribution,ASA classification,BMI and other data between the POCD group（71 cases）and the non-POCD group（1189 cases）（P>0.05）,but there was significant difference in age between the two groups（P<0.05）.2.the risk model for predicting POCD occurrence after TKR was established based on BN algorithm,and the area under the subject curve（AUC）value of the training set was0.9661（95%CI:0.9541,0.9784）,The sensitivity and specificity were 86.01%and 97.02%,respectively.The accuracy was 96.43%（95%CI:0.9511,0.9764）.The AUC value of the test set was 0.8974（95%CI:0.8672,0.9285）,the sensitivity and specificity were 77.14%and95.53%,respectively,and the accuracy was 93.44%（95%CI:0.9092,0.9596）.3.Based on the MLP algorithm,the AUC value of the training set was 0.8715（95%CI:0.8189,0.9242）,the sensitivity and specificity were 74.01%and 89.52%respectively,and the accuracy was 96.15%（95%CI:0.9466,0.9732）.The AUC value of the test set was 0.6819（95%CI:0.5673,0.7965）,the sensitivity and specificity were 70.76%and 88.59%,respectively,and the accuracy was 83.44%（95%CI:0.8037,0.8556）.Conclusion:1.It is feasible to establish a model for predicting the risk of cognitive dysfunction after total knee replacement based on Bayesian network algorithm,and this model has good predictive performance and high accuracy.2.Although the performance and accuracy of the model based on multi-layer perceptron algorithm in predicting the risk of cognitive dysfunction after total knee replacement is inferior to that of Bayesian network model,it can be used as a supplementary method of Bayesian network model in the process of clinical research and application,and provide more references for clinicians and caregivers in the early detection of POCD.