To construct a machine learning prediction model for postoperative liver injury in patients with non-liver surgery based on preoperative and intraoperative medication indicators.

A case-control study was conducted on 315 patients with liver injury after non-liver surgery selected from the databases developed by 3 large general hospitals from January 2014 to September 2022. With the positive/negative ratio of 1:3, 928 cases in corresponding period with non-liver surgery and without liver injury were randomly matched as negative control cases. These 1 243 patients were randomly divided into the modeling group (n=869) and the validation group (n=374) in a ratio of 7:3 using the R language setting code. Preoperative clinical indicators (basic information, medical history, relevant scale score, surgical information and results of laboratory tests) and intraoperative medication were used to construct the prediction model for liver injury after non-liver surgery based on 4 machine learning algorithms, k-nearest neighbor (KNN), support vector machine linear (SVM), logic regression (LR) and extreme gradient boosting (XGBoost). In the validation group, receiver operating characteristic (ROC) curve, precision-recall curve (P-R), decision curve analysis (DCA) curve, Kappa value, sensitivity, specificity, Brier score, and F₁ score were applied to evaluate the efficacy of model.

The model established by 4 machine learning algorithms to predict postoperative liver injury after non-liver surgery was optimal using the XGBoost algorithm. The area under the receiver operating characteristic curve (AUROC) was 0.916 (95%CI: 0.883~0.949), area under the precision-recall curve (AUPRC) was 0.841, Brier score was 0.097, and sensitivity and specificity was 78.95% and 87.10%, respectively.

The postoperative liver injury prediction model for non-liver surgery based on the XGBoost algorithm has effective prediction for the occurrence of postoperative liver injury.

To explore the protective effect of catalpol by enhancing TGR5 expression and reducing bile acid (BAs) level against liver and lung injury in rats with common bile duct ligation (CBDL).

Fifty male rats were randomly divided into sham-operation group (n=10), CBDL group (n=20) and catalpol group (n=20). The survival rate, arterial partial pressure of oxygen (PO₂) and alveolar and arterial PO₂ gradient difference [P(A-a)O₂], liver function, and pathological changes in the liver and lungs. The relationship between bile acid level and TGR5 was analyzed.

Catalpol ameliorated liver function damage, improved survival rate and hypoxemia induced by CBDL. It also reduced angiogenesis in the liver and lung. Meanwhile, catalpol mitigated lung injury caused by excessive BAs levels through enhancing TGR5 expression and reducing FXR expression.

The protective effects of catalpol on CBDL lung injury caused by excessive BAs levels through enhancing TGR5 expression and reducing FXR expression.

To establish a prediction model of risk for postoperative cognitive dysfunction（POCD）after total knee replacement（TKR）by Bayesian network（BN）algorithm and investigate its predictive performance.

A case-control trial was conducted on 1 260 inpatients who underwent TKR from January 2017 to December 2021 in the Department of Joint Surgery of our hospital. Their main diagnosis of inclusion was severe osteoarthritis of left/right knee joint. They were 240 cases of male（19.0%）and 1 020 cases of female（81.0%）, at an average age of 66.73±8.46（23~79）years and a mean body mass index（BMI）of 25.08±5.08 kg/m². The POCD patients（n=71）after surgery（from the end of surgery to discharge）were randomly divided into A1 group and B1 group at a ratio of 7∶3, and those without POCD（1 189 cases）were also randomly divided into A2 group and B2 group at a same ratio. The patients from A1 and A2 groups were combined together as training set, and those out of B1 and B2 groups as test set. 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 stay and promote the maximum recovery of patients.

The prediction model of risk for POCD after TKR was established based on BN algorithm. The area value under receiver operating characteristic curve（ROC-AUC）of the training set was 0.966 1（95% CI: 0.954 1~0.978 4）, and the ROC-AUC value of the test set was 0.897 4（95% CI: 0.867 2~0.928 5）, with an accuracy of 96.43%（95%CI: 0.951 1~0.976 4）and 93.44%（95% CI: 0.909 2~0.959 6）, respectively.

Our risk prediction model for POCD after TKR based on BN algorithm has good performance and high accuracy

To establish a risk model for predicting postoperative cognitive dysfunction（POCD）after non-cardiac surgery using preoperative indicators based on machine learning algorithm.

A case-control study was designed, and conducted on the POCD patients after non-cardiac surgery in the medical big data platform of our hospital from January 2014 to January 2019. Finally, 92 patients were included in the POCD group. According to surgical type and age matched of the POCD group, another 276 patients who did not develop POCD after surgery and discharged from the hospital during the same time period from the same big data platform were assigned into the non-POCD group at a ratio of 1∶3. At the same time, these 368 patients were randomly divided into modeling group（n=259）and validation group（n=109）at a ratio of 7∶3. After data preprocessing and feature selection of preoperative clinical indicators（general data, relevant scoring scales, surgical-related data and results of preoperative laboratory tests）, the risk prediction model of POCD for non-cardiac surgery was established based on 3 machine learning algorithms, that is, logistic regression（LR）, support vector machine（SVM）and Decision Tree. The model efficacy was evaluated by sensitivity, specificity, F1 score, and the area under the receiver operating characteristic curve（AUC）.

The SVM algorithm was the best model among the 3 machine learning algorithms to predict the risk of POCD after non-cardiac surgery. The AUC value of the model in the validation group was 0.957（95%CI: 0.905~1.000）, with a sensitivity and specificity of 92.6% and 98.8%, respectively.

A prediction model of POCD after non-cardiac surgery is successfully established based on machine learning algorithm. This model shows good predictive performance for POCD.

To explore the feasibility of constructing prediction models of postoperative sepsis mortality risk based on machine learning in patients undergoing abdominal surgery.

A case-control trial was designed and conducted on the patients diagnosed with sepsis after abdominal surgery from Medical Information Mart for Intensive Care Ⅳ（MIMIC-Ⅳ）database, and 90-day mortality was defined as the primary endpoint event after hospitalization. The dataset was ramdomly split into training（70%）and test（30%）datasets according to wether diagnosed with postopertive sepsis or not. On the training dataset, logistic regression（LR）, gradient boosting decision tree（GBDT）, random forest（RF）, support vector machine（SVM）and adaptive boosting（AdaBoost）were used to develop the prediction model for death. The area under the receiver operating characteristic curve（AUC）, sensitivity, specificity, positive predictive value, negative predictive value, accuracy and F1 score were used for model evaluation on the test dataset.

A total of 986 patients were finally analyzed, of whom 251 patients（25.5%）died within 90 d after hospitalization. The AUC values of LR, GBDT, RF, SVM and AdaBoost prediction models were 0.852, 0.903, 0.921, 0.940 and 0.906, respectively. The model based on SVM yielded the best AUC value, higher differentiation and better prediction performance, while LR performed the worst among them.

The performances of the prediction model of postoperative sepsis mortality based on GBDTT, RF, SVM and AdaBoost are all better than that of traditional LR model, which may help to assist clinical decision making and improve adverse outcomes.

To explore the application of deep learning in automatic classification of anemia with conjunctival images as input.

The conjunctival images of 284 patients undergoing elective surgery in the Department of Anesthesiology of the First Affiliated Hospital of Army Medical University from March 18 to April 26, 2021 were collected and analyzed prospectively. The images divided into 2 types: normal and anemia according to the corresponding hemoglobin concentration. Four deep learning algorithms, including InceptionV3, ResNet50V2, EfficientNetV2B0 and DenseNet121, were used to construct a prediction model for anemia. The performance of the model was evaluated by receiver operating characteristic（ROC）curve with accuracy, sensitivity, specificity, positive predictive value and negative predictive value.

The area under ROC curve（AUC）was 0.709（95%CI: 0.643~0.769）, 0.661（95%CI: 0.594~0.725）, 0.670（95%CI: 0.603~0.733）, and 0.695（95%CI: 0.628~0.756）, respectively for the 4 deep learning algorithms. The InceptionV3 model showed superior predictive performance on the test set, with an AUC value of 0.709（95%CI: 0.643~0.769）, an accuracy of 0.695, a sensitivity of 0.750, a specificity of 0.412, a positive predictive value of 0.707 and a negative predictive value of 0.629. Based on the optimal algorithm, a network service application which can be used for online prediction of anemia was developed（http://150.158.58.4）.

Our model, which is established based on deep learning algorithm with conjunctiva image as input, has a good performance on fast and automatic prediction for anemia. The InceptionV3model has better comprehensive prediction performance.

To explore the application of support vector machine (SVM) in predicting the mortality risk after intensive care unit (ICU) admission.

A total of 18 094 ICU inpatients from MIMIC Ⅲ dataset were enrolled in the study. The total data set (n=18 094) was randomly divided into training data set (n=12 666, 70%) and test data set (n=5 428, 30%). Based on the Python, the machine learning algorithm, SVM, was used to establish a prediction model of the mortality risk after ICU admission with the results of LASSO feature selection. The efficacy of model was evaluated using the test data set.

The areas under the receiver operating characteristic (AUCROC) curves of the SVM-based model for predicting the mortality risk in 24 h and 48 h after ICU admission were 0.805 1 (0.793 6~0.816 6) and 0.811 7 (0.799 9~0.824), with sensitivities of 0.751 3 and 0.737 2, and specificities of 0.713 0 and 0.742 9, respectively.

The SVM-based model for predicting the mortality risk after ICU admission has a satisfactory result and high accuracy.