ObjectiveTo explore the clinical value of three early predictive scale of lung injury (ALI) in patients with high risk of acute lung injury (ALI) after lung cancer surgery.MethodsA convenient sampling method was used in this study. A retrospective analysis was performed on patients with lung cancer underwent lung surgery. The patients were divided into an ALI group and a non-ALI group according to ALI diagnostic criteria. Three kinds of lung injury predictive scoring methods were used, including lung injury prediction score (LIPS), surgical lung injury prediction (SLIP) and SLIP-2. The differences in the scores of the two groups were compared. The correlation between the three scoring methods was also analyzed. The diagnostic value was analyzed by drawing receiver operating characteristic (ROC) curves.ResultsA total of 400 patients underwent lung cancer surgery, and 38 patients (9.5%) developed ALI after operation. Among them, 2 cases progressed to acute respiratory distress syndrome and were treated in intensive care unit. There were no deaths. The predictive scores of the patients in the ALI group were higher than those in the non-ALI group, and the difference was statistically significant (all P<0.001). There was a good correlation between the three scoring methods (allP<0.001). The three scoring methods had better diagnostic value for early prediction of high risk ALI patients after lung cancer surgery and their area under ROC curve (AUC) were larger than 0.8. LIPS score performed better than others, with an AUC of 0.833, 95%CI (0.79, 0.87).ConclusionThree predictive scoring methods may be applied to early prediction of high risk ALI patients after lung cancer surgery, in which LIPS performs better than others.
Objective To systematically review risk prediction models of in-hospital cardiac arrest in patients with cardiovascular disease, and to provide references for related clinical practice and scientific research for medical professionals in China. Methods Databases including CBM, CNKI, WanFang Data, PubMed, ScienceDirect, Web of Science, The Cochrane Library, Wiley Online Journals and Scopus were searched to collect studies on risk prediction models for in-hospital cardiac arrest in patients with cardiovascular disease from January 2010 to July 2022. Two researchers independently screened the literature, extracted data, and evaluated the risk of bias of the included studies. Results A total of 5 studies (4 of which were retrospective studies) were included. Study populations encompassed mainly patients with acute coronary syndrome. Two models were modeled using decision trees. The area under the receiver operating characteristic curve or C statistic of the five models ranged from 0.720 to 0.896, and only one model was verified externally and for time. The most common risk factors and immediate onset factors of in-hospital cardiac arrest in patients with cardiovascular disease included in the prediction model were age, diabetes, Killip class, and cardiac troponin. There were many problems in analysis fields, such as insufficient sample size (n=4), improper handling of variables (n=4), no methodology for dealing with missing data (n=3), and incomplete evaluation of model performance (n=5). Conclusion The prediction efficiency of risk prediction models for in-hospital cardiac arrest in patients with cardiovascular disease was good; however, the model quality could be improved. Additionally, the methodology needs to be improved in terms of data sources, selection and measurement of predictors, handling of missing data, and model evaluations. External validation of existing models is required to better guide clinical practice.
Objective To investigate the nutritional status of hospitalized patients with chronic kidney disease (CKD), analyze the influencing factors, and construct a predictive model to provide a localized theoretical basis and more convenient risk prediction indicators and models for clinical nutrition support and intervention treatment of CKD patients in China. Methods Convenience sampling was used to select hospitalized CKD patients from Department of Nephrology, West China Hospital, Sichuan University, from January to October 2019. General information questionnaires, the Nutritional Risk Screening 2002 scale, and the Huaxi Emotional-distress Index questionnaire were used for data collection. Single factor analyses and multiple logistic regression analysis were conducted to explore the risk factors for malnutrition in CKD hospitalized patients. A predictive model was established and evaluated using receiver operating characteristic (ROC) curve analysis and bootstrap resampling. Results A total of 1059 valid copies of questionnaires were collected out of 1118 distributed. Among the 1059 CKD hospitalized patients, 207 cases (19.5%) were identified as having nutritional risk. The multiple logistic regression analysis showed that CKD stage [odds ratio (OR)=1.874, 95% confidence interval (CI) (1.631, 2.152), P<0.001], age [OR=1.015, 95%CI (1.003, 1.028), P=0.018], and the Huaxi Emotional-distress Index [OR=1.024, 95%CI (1.002, 1.048), P=0.033] were independent risk factors for malnutrition in CKD hospitalized patients, while serum albumin [OR=0.880, 95%CI (0.854, 0.907), P<0.001] was an independent protective factor. The evaluation of the multiple logistic regression analysis predictive model showed a concordance index of 0.977, standard deviation of 0.021, and P<0.05. The area under the ROC curve was 0.977. Conclusions The prevalence of malnutrition is relatively high among CKD hospitalized patients. CKD stage, age, psychological status, and serum albumin are influencing factors for malnutrition in CKD hospitalized patients. The multiple logistic regression model based on the above indicators demonstrates good predictive performance and is expected to provide assistance for early nutritional intervention to improve the clinical outcomes and quality of life for CKD patients with malnutrition in China.
The drug-target protein interaction prediction can be used for the discovery of new drug effects. Recent studies often focus on the prediction of an independent matrix filling algorithm, which apply a single algorithm to predict the drug-target protein interaction. The single-model matrix-filling algorithms have low accuracy, so it is difficult to obtain satisfactory results in the prediction of drug-target protein interaction. AdaBoost algorithm is a strong multiple classifier combination framework, which is proved by the past researches in classification applications. The drug-target interaction prediction is a matrix filling problem. Therefore, we need to adjust the matrix filling problem to a classification problem before predicting the interaction among drug-target protein. We make full use of the AdaBoost algorithm framework to integrate several weak classifiers to improve performance and make accurate prediction of drug-target protein interaction. Experimental results based on the metric datasets show that our algorithm outperforms the other state-of-the-art approaches and classical methods in accuracy. Our algorithm can overcome the limitations of the single algorithm based on machine learning method, exploit the hidden factors better and improve the accuracy of prediction effectively.
Objective To evaluate the predictive effect of three machine learning methods, namely support vector machine (SVM), K-nearest neighbor (KNN) and decision tree, on the daily number of new patients with ischemic stroke in Chengdu. Methods The numbers of daily new ischemic stroke patients from January 1st, 2019 to March 28th, 2021 were extracted from the Third People’s Hospital of Chengdu. The weather and meteorological data and air quality data of Chengdu came from China Weather Network in the same period. Correlation analyses, multinominal logistic regression, and principal component analysis were used to explore the influencing factors for the level of daily number of new ischemic stroke patients in this hospital. Then, using R 4.1.2 software, the data were randomly divided in a ratio of 7∶3 (70% into train set and 30% into validation set), and were respectively used to train and certify the three machine learning methods, SVM, KNN and decision tree, and logistic regression model was used as the benchmark model. F1 score, the area under the receiver operating characteristic curve (AUC) and accuracy of each model were calculated. The data dividing, training and validation were repeated for three times, and the average F1 scores, AUCs and accuracies of the three times were used to compare the prediction effects of the four models. Results According to the accuracies from high to low, the prediction effects of the four models were ranked as SVM (88.9%), logistic regression model (87.5%), decision tree (85.9%), and KNN (85.1%); according to the F1 scores, the models were ranked as SVM (66.9%), KNN (62.7%), decision tree (59.1%), and logistic regression model (57.7%); according to the AUCs, the order from high to low was SVM (88.5%), logistic regression model (87.7%), KNN (84.7%), and decision tree (71.5%). Conclusion The prediction result of SVM is better than the traditional logistic regression model and the other two machine learning models.
ObjectiveTo analyze prognostic ability of inflammation-based Glasgow prognostic score (GPS) in patients with ST-segment elevation myocardial infarction (STEMI).MethodsWe retrospectively analyzed the clinical data of 289 patients with STEMI admitted to the Department of Emergency in West China Hospital from April 2015 to January 2016. All study subjects were divided into three groups: a group of GPS 0 (190 patients including 150 males and 40 females aged 62.63±12.98 years), a group of GPS 1 (78 patients including 58 males and 20 females aged 66.57±15.25 years), and a group of GPS 2 (21 patients including 16 males and 5 females aged 70.95±9.58 years). Cox regression analysis was conducted to analyze the independent risk factors of predicting long-term mortality of patients with STEMI.ResultsThere was a statistical difference in long-term mortality (9.5% vs. 23.1% vs. 61.9%, P<0.001) and in-hospital mortality (3.7% vs. 7.7% vs. 23.8%, P<0.001) among the three groups. The Global Registry of Acute Coronary Events (GRACE) scores and Gensini scores increased in patients with higher GPS scores, and the differences were statistically different (P<0.001). Multivariable Cox regression analysis showed that the GPS was independently associated with STEMI long-term all-cause mortality (1 vs. 0, HR: 2.212, P=0.037; 2 vs. 0, HR: 8.286, P<0.001).ConclusionGPS score is helpful in predicting the long-term and in-hospital prognosis of STEMI patients, and thus may guide clinical precise intervention by early risk stratification.
Abstract： Objective To establish a risk prediction model and risk score for inhospital mortality in heart valve surgery patients, in order to promote its perioperative safety. Methods We collected records of 4 032 consecutive patients who underwent aortic valve replacement, mitral valve repair, mitral valve replacement, or aortic and mitral combination procedure in Changhai hospital from January 1,1998 to December 31,2008. Their average age was 45.90±13.60 years and included 1 876 (46.53%) males and 2 156 (53.57%) females. Based on the valve operated on, we divided the patients into three groups including mitral valve surgery group (n=1 910), aortic valve surgery group (n=724), and mitral plus aortic valve surgery group (n=1 398). The population was divided a 60% development sample (n=2 418) and a 40% validation sample (n=1 614). We identified potential risk factors, conducted univariate analysis and multifactor logistic regression to determine the independent risk factors and set up a risk model. The calibration and discrimination of the model were assessed by the HosmerLemeshow (H-L) test and [CM(159mm]the area under the receiver operating characteristic (ROC) curve,respectively. We finally produced a risk score according to the coefficient β and rank of variables in the logistic regression model. Results The general inhospital mortality of the whole group was 4.74% (191/4 032). The results of multifactor logistic regression analysis showed that eight variables including tricuspid valve incompetence with OR=1.33 and 95%CI 1.071 to 1.648, arotic valve stenosis with OR=1.34 and 95%CI 1.082 to 1.659, chronic lung disease with OR=2.11 and 95%CI 1.292 to 3.455, left ventricular ejection fraction with OR=1.55 and 95%CI 1.081 to 2.234, critical preoperative status with OR=2.69 and 95%CI 1.499 to 4.821, NYHA ⅢⅣ (New York Heart Association) with OR=2.75 and 95%CI 1.343 to 5641, concomitant coronary artery bypass graft surgery (CABG) with OR=3.02 and 95%CI 1.405 to 6.483, and serum creatinine just before surgery with OR=4.16 and 95%CI 1.979 to 8.766 were independently correlated with inhospital mortality. Our risk model showed good calibration and discriminative power for all the groups. P values of H-L test were all higher than 0.05 (development sample: χ2=1.615, P=0.830, validation sample: χ2=2.218, P=0.200, mitral valve surgery sample: χ2=5.175,P=0.470, aortic valve surgery sample: χ2=12.708, P=0.090, mitral plus aortic valve surgery sample: χ2=3.875, P=0.380), and the areas under the ROC curve were all larger than 0.70 (development sample: 0.757 with 95%CI 0.712 to 0.802, validation sample: 0.754 and 95%CI 0.701 to 0806; mitral valve surgery sample: 0.760 and 95%CI 0.706 to 0.813, aortic valve surgery sample: 0.803 and 95%CI 0.738 to 0.868, mitral plus aortic valve surgery sample: 0.727 and 95%CI 0.668 to 0.785). The risk score was successfully established: tricuspid valve regurgitation (mild:1 point, moderate: 2 points, severe:3 points), arotic valve stenosis (mild: 1 point, moderate: 2 points, severe: 3 points), chronic lung disease (3 points), left ventricular ejection fraction (40% to 50%: 2 points, 30% to 40%: 4 points, ＜30%: 6 points), critical preoperative status (3 points), NYHA IIIIV (4 points), concomitant CABG (4 points), and serum creatinine (＞110 μmol/L: 5 points).Conclusion Eight risk factors including tricuspid valve regurgitation are independent risk factors associated with inhospital mortality of heart valve surgery patients in China. The established risk model and risk score have good calibration and discrimination in predicting inhospital mortality of heart valve surgery patients.
Pathological images of gastric cancer serve as the gold standard for diagnosing this malignancy. However, the recurrence prediction task often encounters challenges such as insignificant morphological features of the lesions, insufficient fusion of multi-resolution features, and inability to leverage contextual information effectively. To address these issues, a three-stage recurrence prediction method based on pathological images of gastric cancer is proposed. In the first stage, the self-supervised learning framework SimCLR was adopted to train low-resolution patch images, aiming to diminish the interdependence among diverse tissue images and yield decoupled enhanced features. In the second stage, the obtained low-resolution enhanced features were fused with the corresponding high-resolution unenhanced features to achieve feature complementation across multiple resolutions. In the third stage, to address the position encoding difficulty caused by the large difference in the number of patch images, we performed position encoding based on multi-scale local neighborhoods and employed self-attention mechanism to obtain features with contextual information. The resulting contextual features were further combined with the local features extracted by the convolutional neural network. The evaluation results on clinically collected data showed that, compared with the best performance of traditional methods, the proposed network provided the best accuracy and area under curve (AUC), which were improved by 7.63% and 4.51%, respectively. These results have effectively validated the usefulness of this method in predicting gastric cancer recurrence.
ObjectiveTo systematically review the prediction models of blood-based biomarkers for non-small cell lung cancer (NSCLC). MethodsThe PubMed, Embase, Cochrane Library, Web of Science, VIP, WanFang Data and CNKI databases were electronically searched to collect studies related to the objectives from inception to June, 2023. Two reviewers independently screened literature, extracted data and assessed the risk of bias of the included studies. Meta-analysis was then performed by using RevMan 5.4.1 software. ResultsA total of 8 studies were included and all of them were retrospective cohort studies. The models were internally validated in 2 studies and externally validated in 4 studies. The performances of the eight predictive models were stable, which was measured by the area under the curve of receiver operating characteristic curve lying between 0.664 and 0.783. However, the risk of bias was high, which may mainly be reflected in data processing, model validation and performance adjustment. Meta-analysis showed that LDH (HR=1.86, 95%CI 41.32 to 2.63, P<0.01), dNLR (HR=2.15, 95%CI 1.56 to 2.96, P<0.01) and NLR (HR=1.71, 95%CI 1.08 to 2.69, P=0.02) were independent factors of prognosis for NSCLC patients. Conclusion?Current evidence shows that the NSCLC prediction models based on peripheral blood biomarkers are still in the development stage, and the models have a high risk of bias.
ObjectiveTo analyze the predictive value of ensemble classification algorithm of random forest for delirium risk in ICU patients with cardiothoracic surgery. MethodsA total of 360 patients hospitalized in cardiothoracic ICU of our hospital from June 2019 to December 2020 were retrospectively analyzed. There were 193 males and 167 females, aged 18-80 (56.45±9.33) years. The patients were divided into a delirium group and a control group according to whether delirium occurred during hospitalization or not. The clinical data of the two groups were compared, and the related factors affecting the occurrence of delirium in cardiothoracic ICU patients were predicted by the multivariate logistic regression analysis and the ensemble classification algorithm of random forest respectively, and the difference of the prediction efficiency between the two groups was compared.ResultsOf the included patients, 19 patients fell out, 165 patients developed ICU delirium and were enrolled into the delirium group, with an incidence of 48.39% in ICU, and the remaining 176 patients without ICU delirium were enrolled into the control group. There was no statistical significance in gender, educational level, or other general data between the two groups (P>0.05). But compared with the control group, the patients of the delirium group were older, length of hospital stay was longer, and acute physiology and chronic health evaluationⅡ(APACHEⅡ) score, proportion of mechanical assisted ventilation, physical constraints, sedative drug use in the delirium group were higher (P<0.05). Multivariate logistic regression analysis showed that age (OR=1.162), length of hospital stay (OR=1.238), APACHEⅡ score (OR=1.057), mechanical ventilation (OR=1.329), physical constraints (OR=1.345) and sedative drug use (OR=1.630) were independent risk factors for delirium of cardiothoracic ICU patients. The variables in the random forest model for sorting, on top of important predictor variable were: age, length of hospital stay, APACHEⅡ score, mechanical ventilation, physical constraints and sedative drug use. The diagnostic efficiency of ensemble classification algorithm of random forest was obviously higher than that of multivariate logistic regression analysis. The area under receiver operating characteristic curve of ensemble classification algorithm of random forest was 0.87, and the one of multivariate logistic regression analysis model was 0.79.ConclusionThe ensemble classification algorithm of random forest is more effective in predicting the occurrence of delirium in cardiothoracic ICU patients, which can be popularized and applied in clinical practice and contribute to early identification and strengthening nursing of high-risk patients.