Ten-fold Cross-validation Research Articles

HFpEF as systemic disease, insight from a diagnostic prediction model reminiscent of systemic inflammation and organ interaction in HFpEF patients

Systemic inflammation and reciprocal organ interactions are associated with the pathophysiology of heart failure with preserved ejection fraction (HFpEF). However, the clinical value, especially the diagnositc prediction power of inflammation and extra-cardiac organ dysfunction for HfpEF is not explored. In this cross-sectional study, 1808 hospitalized patients from January 2014 to June 2022 in ChiHFpEF cohort were totally enrolled according to inclusion and exclusion criteria. A diagnostic model with markers from routine blood test as well as liver and renal dysfunction for HFpEF was developed using data from ChiHFpEF-cohort by logistic regression and assessed by receiver operating characteristic curve (ROC) and Brier score. Then, the model was validated by the tenfold cross-validation and presented as nomogram and a web-based online risk calculator as well. Multivariate and LASSO regression analysis revealed that age, hemoglobin, neutrophil to lymphocyte ratio, AST/ALT ratio, creatinine, uric acid, atrial fibrillation, and pulmonary hypertension were associated with HFpEF. The predictive model exhibited reasonably accurate discrimination (ROC, 0.753, 95% CI 0.732–0.772) and calibration (Brier score was 0.200). Subsequent internal validation showed good discrimination and calibration (AUC = 0.750, Brier score was 0.202). In additoin to participating in pathophysiology of HFpEF, inflammation and multi-organ interactions have diagnostic prediction value for HFpEF. Screening and optimizing biomarkers of inflammation and multi-organ interactions stand for a new field to improve noninvasive diagnostic tool for HFpEF.

Using machine learning techniques to predict the risk of osteoporosis based on nationwide chronic disease data

Osteoporosis is a major public health concern that significantly increases the risk of fractures. The aim of this study was to develop a Machine Learning based predictive model to screen individuals at high risk of osteoporosis based on chronic disease data, thus facilitating early detection and personalized management. A total of 10,000 complete patient records of primary healthcare data in the German Disease Analyzer database (IMS HEALTH) were included, of which 1293 diagnosed with osteoporosis and 8707 without the condition. The demographic characteristics and chronic disease data, including age, gender, lipid disorder, cancer, COPD, hypertension, heart failure, CHD, diabetes, chronic kidney disease, and stroke were collected from electronic health records. Ten different machine learning algorithms were employed to construct the predictive mode. The performance of the model was further validated and the relative importance of features in the model was analyzed. Out of the ten machine learning algorithms, the Stacker model based on Logistic Regression, AdaBoost Classifier, and Gradient Boosting Classifier demonstrated superior performance. The Stacker model demonstrated excellent performance through ten-fold cross-validation on the training set and ROC curve analysis on the test set. The confusion matrix, lift curve and calibration curves indicated that the Stacker model had optimal clinical utility. Further analysis on feature importance highlighted age, gender, lipid metabolism disorders, cancer, and COPD as the top five influential variables. In this study, a predictive model for osteoporosis based on chronic disease data was developed using machine learning. The model shows great potential in early detection and risk stratification of osteoporosis, ultimately facilitating personalized prevention and management strategies.

Novel Functional Radiomics for Prediction of Cardiac Positron Emission Tomography Avidity in Lung Cancer Radiotherapy.

Traditional methods of evaluating cardiotoxicity focus on radiation doses to the heart. Functional imaging has the potential to provide improved prediction for cardiotoxicity for patients with lung cancer. Fluorine-18 (18F) fluorodeoxyglucose (FDG)-positron emission tomography (PET)/computed tomography (CT) imaging is routinely obtained in a standard cancer staging workup. This work aimed to develop a radiomics model predicting clinical cardiac assessment using 18F-FDG PET/CT scans before thoracic radiation therapy. Pretreatment 18F-FDG PET/CT scans from three study populations (N = 100, N = 39, N = 70) were used, comprising two single-institutional protocols and one publicly available data set. A clinician (V.J.) classified the PET/CT scans per clinical cardiac guidelines as no uptake, diffuse uptake, or focal uptake. The heart was delineated, and 210 novel functional radiomics features were selected to classify cardiac FDG uptake patterns. Training data were divided into training (80%)/validation (20%) sets. Feature reduction was performed using the Wilcoxon test, hierarchical clustering, and recursive feature elimination. Ten-fold cross-validation was carried out for training, and the accuracy of the models to predict clinical cardiac assessment was reported. From 202 of 209 scans, cardiac FDG uptake was scored as no uptake (39.6%), diffuse uptake (25.3%), and focal uptake (35.1%), respectively. Sixty-two independent radiomics features were reduced to nine clinically pertinent features. The best model showed 93% predictive accuracy in the training data set and 80% and 92% predictive accuracy in two external validation data sets. This work used an extensive patient data set to develop a functional cardiac radiomic model from standard-of-care 18F-FDG PET/CT scans, showing good predictive accuracy. The radiomics model has the potential to provide an automated method to predict existing cardiac conditions and provide an early functional biomarker to identify patients at risk of developing cardiac complications after radiotherapy.

The value of speckle-tracking stratified strain combined with myocardial work measurement in evaluating left ventricular function in patients with heart failure with preserved ejection fraction.

Heart failure with preserved ejection fraction (HFpEF) is a highly prevalent progressive disease accompanied by poor quality of life, high utilization of medical resources, morbidity, and mortality. However, the role of left ventricular (LV) systolic dysfunction has yet to be well elaborated despite the preservation of the LV ejection fraction. This study aimed to explore the diagnostic value of speckle-tracking stratified strain combined with myocardial work (MW) measurement in evaluating LV systolic dysfunction in patients with HFpEF. A total of 125 study consecutive individuals, 64 HFpEF patients, and 61 controls were prospectively enrolled in the Fourth Affiliated Hospital of Harbin Medical University. In addition to the conventional echocardiographic parameters, LV stratified strain and MW parameters were statistically compared between the HFpEF and control groups. The global longitudinal strain (GLS) of the subendocardium, myocardium, and subepicardium (GLSendo, GLSmyo, and GLSepi); the transmural gradient (ΔGLS); the global myocardial work index (GWI), global myocardial work efficiency (GWE), global myocardial constructive work (GCW), and the global myocardial wasted work (GWW) were included. Area under the receiver operating characteristic curve analysis was used to evaluate the diagnostic performance of these univariate and multivariable logistic models in detecting impaired LV systolic function in HFpEF. Ten-fold cross-validation was used to evaluate the generalizability of the predictive model. Stratified strains values showed a gradient decline from GLSendo to GLSepi in both control and HFpEF patients. Compared with the control group, HFpEF patients had a significantly reduced GLSepi, GLSmyo, GLSendo, ΔGLS, GWI, GWE, and GCW and a significantly increased GWW (all P<0.001). In the derivation set, the optimal logistic model (combined stratified strain and MW variables) demonstrated the highest performance in predicting LV systolic function impairment in HFpEF patients. The best-performing model with a mean area under the curve (AUC) of 0.966 [95% confidence interval (CI): 0.88 to 1] accessed by 10-fold cross-validation. In the validation set, the AUC of the optimal logistic model was 0.933 (95% CI: 0.85 to 1), the sensitivity was 87%, and the specificity was 93%. Both speck-tracking stratified strain and MW measurement may sensitively detect impairment of LV myocardial function at an early stage for patients with HFpEF. Combining the two techniques may improve the quality of HFpEF diagnosis and may provide a reference value for the early diagnosis of HFpEF in the future.

Risk factor analysis and prediction model construction for surgical patients with venous thromboembolism: A prospective study

Abstract Objective: Patients undergoing surgery are at high risk of developing venous thromboembolism (VTE). This study aimed to determine the predictive value of risk factors for VTE in surgical patients and to develop a prediction model by integrating independent predictors. Methods: A total of 1111 patients who underwent surgery at clinical departments in a tertiary general hospital were recruited between May and July 2021. Clinical data, including patient-related, surgery-related, and laboratory parameters, were extracted from the hospital information system and electronic medical records. A VTE prediction model incorporating ten risk variables was constructed using artificial neural networks (ANNs). Results: Ten independent factors (X1: age, X2: alcohol consumption, X3: hypertension, X4: bleeding, X5: blood transfusions, X6: general anesthesia, X7: intrathecal anesthesia, X8: D-dimer, X9: C-reactive protein, and X10: lymphocyte percentage) were identified as associated with an increased risk of VTE. Ten-fold cross-validation results showed that the ANN model was capable of predicting VTE in surgical patients, with an area under the curve (AUC) of 0.89, a Brier score of 0.01, an accuracy of 0.96, and a F1 score of 0.92. The ANN model slightly outperformed the logistic regression model and the Caprini model, but a DeLong test showed that the statistical difference in the AUCs of the ANN and logistic regression models was insignificant (P&gt;0.05). Conclusion: Ten statistical indicators relevant to VTE risk prediction for surgical patients were identified, and ANN and logistic regression both showed promising results as decision-supporting tools for VTE prediction.

Prediction of solid and micropapillary components in lung invasive adenocarcinoma: radiomics analysis from high-spatial-resolution CT data with 1024 matrix

PurposeTo predict solid and micropapillary components in lung invasive adenocarcinoma using radiomic analyses based on high-spatial-resolution CT (HSR-CT).Materials and methodsFor this retrospective study, 64 patients with lung invasive adenocarcinoma were enrolled. All patients were scanned by HSR-CT with 1024 matrix. A pathologist evaluated subtypes (lepidic, acinar, solid, micropapillary, or others). Total 61 radiomic features in the CT images were calculated using our modified texture analysis software, then filtered and minimized by least absolute shrinkage and selection operator (LASSO) regression to select optimal radiomic features for predicting solid and micropapillary components in lung invasive adenocarcinoma. Final data were obtained by repeating tenfold cross-validation 10 times. Two independent radiologists visually predicted solid or micropapillary components on each image of the 64 nodules with and without using the radiomics results. The quantitative values were analyzed with logistic regression models. The receiver operating characteristic curves were generated to predict of solid and micropapillary components. P values < 0.05 were considered significant.ResultsTwo features (Coefficient Variation and Entropy) were independent indicators associated with solid and micropapillary components (odds ratio, 30.5 and 11.4; 95% confidence interval, 5.1–180.5 and 1.9–66.6; and P = 0.0002 and 0.0071, respectively). The area under the curve for predicting solid and micropapillary components was 0.902 (95% confidence interval, 0.802 to 0.962). The radiomics results significantly improved the accuracy and specificity of the prediction of the two radiologists.ConclusionTwo texture features (Coefficient Variation and Entropy) were significant indicators to predict solid and micropapillary components in lung invasive adenocarcinoma.

Advanced gastrointestinal stromal tumor: reliable classification of imatinib plasma trough concentration via machine learning

AimPatients with advanced gastrointestinal stromal tumors (GISTs) exhibiting an imatinib plasma trough concentration (IM Cmin) under 1100 ng/ml may show a reduced drug response rate, leading to the suggestion of monitoring for IM Cmin. Consequently, the objective of this research was to create a customized IM Cmin classification model for patients with advanced GISTs from China.MethodsInitial data and laboratory indicators from patients with advanced GISTs were gathered, and the above information was segmented into a training set, validation set, and testing set in a 6:2:2 ratio. Key variables associated with IM Cmin were identified to construct the classification model using the least absolute shrinkage and selection operator (LASSO) regression and forward stepwise binary logistic regression. Within the training and validation sets, nine ML classification models were constructed via the resampling method and underwent comparison through the Brier scores, the areas under the receiver-operating characteristic curve (AUROC), the decision curve, and the precision-recall (AUPR) curve to determine the most suitable model for this dataset. Two methods of internal validation were used to assess the most suitable model's classification performance: tenfold cross-validation and random split-sample validation (test set), and the value of the test set AUROC was used to evaluate the model's classification performance.ResultsSix key variables (gender, daily IM dose, metastatic site, red blood cell count, platelet count, and percentage of neutrophils) were ultimately selected to construct the classification model. In the validation set, it is found by comparison that the Extreme Gradient Boosting (XGBoost) model has the largest AUROC, the lowest Brier score, the largest area under the decision curve, and the largest AUPR value. Furthermore, as evaluated via internal verification, it also performed well in the test set (AUROC = 0.725).ConclusionFor patients with advanced GISTs who receive IM, initial data and laboratory indicators could be used to accurately estimate whether the IM Cmin is below 1100 ng/ml. The XGBoost model may stand a chance to assist clinicians in directing the administration of IM.

Characterizing hub biomarkers for post-transplant renal fibrosis and unveiling their immunological functions through RNA sequencing and advanced machine learning techniques

BackgroundKidney transplantation stands out as the most effective renal replacement therapy for patients grappling with end-stage renal disease. However, post-transplant renal fibrosis is a prevalent and irreversible consequence, imposing a substantial clinical burden. Unfortunately, the clinical landscape remains devoid of reliable biological markers for diagnosing post-transplant renal interstitial fibrosis.MethodsWe obtained transcriptome and single-cell sequencing datasets of patients with renal fibrosis from NCBI Gene Expression Omnibus (GEO). Subsequently, we employed Weighted Gene Co-Expression Network Analysis (WGCNA) to identify potential genes by integrating core modules and differential genes. Functional enrichment analysis was conducted to unveil the involvement of potential pathways. To identify key biomarkers for renal fibrosis, we utilized logistic analysis, a LASSO-based tenfold cross-validation approach, and gene topological analysis within Cytoscape. Furthermore, histological staining, Western blotting (WB), and quantitative PCR (qPCR) experiments were performed in a murine model of renal fibrosis to verify the identified hub genes. Moreover, molecular docking and molecular dynamics simulations were conducted to explore possible effective drugs.ResultsThrough WGCNA, the intersection of core modules and differential genes yielded a compendium of 92 potential genes. Logistic analysis, LASSO-based tenfold cross-validation, and gene topological analysis within Cytoscape identified four core genes (CD3G, CORO1A, FCGR2A, and GZMH) associated with renal fibrosis. The expression of these core genes was confirmed through single-cell data analysis and validated using various machine learning methods. Wet experiments also verified the upregulation of these core genes in the murine model of renal fibrosis. A positive correlation was observed between the core genes and immune cells, suggesting their potential role in bolstering immune system activity. Moreover, four potentially effective small molecules (ZINC000003830276-Tessalon, ZINC000003944422-Norvir, ZINC000008214629-Nonoxynol-9, and ZINC000085537014-Cobicistat) were identified through molecular docking and molecular dynamics simulations.ConclusionFour potential hub biomarkers most associated with post-transplant renal fibrosis, as well as four potentially effective small molecules, were identified, providing valuable insights for studying the molecular mechanisms underlying post-transplant renal fibrosis and exploring new targets.

Cancer screening in hospitalized ischemic stroke patients: a multicenter study focused on multiparametric analysis to improve management of occult cancers.

The reciprocal promotion of cancer and stroke occurs due to changes in shared risk factors, such as metabolic pathways and molecular targets, creating a "vicious cycle." Cancer plays a direct or indirect role in the pathogenesis of ischemic stroke (IS), along with the reactive medical approach used in the treatment and clinical management of IS patients, resulting in clinical challenges associated with occult cancer in these patients. The lack of reliable and simple tools hinders the effectiveness of the predictive, preventive, and personalized medicine (PPPM/3PM) approach. Therefore, we conducted a multicenter study that focused on multiparametric analysis to facilitate early diagnosis of occult cancer and personalized treatment for stroke associated with cancer. Admission routine clinical examination indicators of IS patients were retrospectively collated from the electronic medical records. The training dataset comprised 136 IS patients with concurrent cancer, matched at a 1:1 ratio with a control group. The risk of occult cancer in IS patients was assessed through logistic regression and five alternative machine-learning models. Subsequently, select the model with the highest predictive efficacy to create a nomogram, which is a quantitative tool for predicting diagnosis in clinical practice. Internal validation employed a ten-fold cross-validation, while external validation involved 239 IS patients from six centers. Validation encompassed receiver operating characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and comparison with models from prior research. The ultimate prediction model was based on logistic regression and incorporated the following variables: regions of ischemic lesions, multiple vascular territories, hypertension, D-dimer, fibrinogen(FIB), and hemoglobin(Hb). The area under the ROC curve (AUC) for the nomogram was 0.871 in the training dataset and 0.834 in the external test dataset. Both calibration curves and DCA underscored the nomogram's strong performance. The nomogram enables early occult cancer diagnosis in hospitalized IS patients and helps to accurately identify the cause of IS, while the promotion of IS stratification makes personalized treatment feasible. The online nomogram based on routine clinical examination indicators of IS patients offered a cost-effective platform for secondary care in the framework of PPPM. The online version contains supplementary material available at 10.1007/s13167-024-00354-8.

Machine learning-based prediction of in-hospital mortality for critically ill patients with sepsis-associated acute kidney injury

Objectives This study aims to develop and validate a prediction model in-hospital mortality in critically ill patients with sepsis-associated acute kidney injury (SA-AKI) based on machine learning algorithms. Methods Patients who met the criteria for inclusion were identified in the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database and divided according to the validation (n = 2440) and development (n = 9756, 80%) queues. Ensemble stepwise feature selection method was used to screen for effective features. The prediction models of short-term mortality were developed by seven machine learning algorithms. Ten-fold cross-validation was used to verify the performance of the algorithm in the development queue. The area under the receiver operating characteristic curve (ROC-AUC) was used to evaluate the differentiation accuracy and performance of the prediction model in the validation queue. The best-performing model was interpreted by Shapley additive explanations (SHAP). Results A total of 12,196 patients were enrolled in this study. Eleven variables were finally chosen to develop the prediction model. The AUC of the random forest (RF) model was the highest value both in the Ten-fold cross-validation and evaluation (AUC: 0.798, 95% CI: 0.774–0.821). According to the SHAP plots, old age, low Glasgow Coma Scale (GCS) score, high AKI stage, reduced urine output, high Simplified Acute Physiology Score (SAPS II), high respiratory rate, low temperature, low absolute lymphocyte count, high creatinine level, dysnatremia, and low body mass index (BMI) increased the risk of poor prognosis. Conclusions The RF model developed in this study is a good predictor of in-hospital mortality for patients with SA-AKI in the intensive care unit (ICU), which may have potential applications in mortality prediction.

Prediction of thermoelectric-figure-of-merit based on autoencoder and light gradient boosting machine

The evaluation of thermoelectric materials relies significantly on the thermoelectric figure of merit, ZT, which serves as a crucial parameter in assessing their properties. The accurate prediction of ZT values can be accomplished by utilizing machine learning models to learn material characteristics. However, factors such as the size of the dataset, model hyperparameters, and data quality can all impact the accuracy of machine learning. In contrast to previous research where high-dimensional features were simply discarded to transform them into low-dimensional ones, deep learning models such as autoencoder can extract more effective information. Therefore, in this article, the combination of autoencoders and the Light Gradient Boosting Machine (LightGBM) is employed to learn the chemical characteristics and ZT values of various materials. The reliability of the model was confirmed by achieving an R2 score of 0.94 during tenfold cross-validation. 130 000 materials were predicted and screened, the temperature dependence of the screened materials was studied in depth, and 13 materials with high ZT values were identified. Four of the 13 most promising candidates identified are existing thermoelectric materials, while nine are ideal candidates for future experimental studies and validation. This work utilizes autoencoders for extensive prediction and screening of promising materials, providing an effective approach for handling high-dimensional material data.

Population-Based Prognostic Models for Head and Neck Cancers Using National Cancer Registry Data from Taiwan

PurposeThis study aims to raise awareness of the disparities in survival predictions among races in head and neck cancer (HNC) patients by developing and validating population-based prognostic models specifically tailored for Taiwanese and Asian populations.MethodsA total of 49,137 patients diagnosed with HNCs were included from the Taiwan Cancer Registry (TCR). Six prognostic models, divided into three categories based on surgical status, were developed to predict both overall survival (OS) and cancer-specific survival using the registered demographic and clinicopathological characteristics in the Cox proportional hazards model. The prognostic models underwent internal evaluation through a tenfold cross-validation among the TCR Taiwanese datasets and external validation across three primary racial populations using the Surveillance, Epidemiology, and End Results database. Predictive performance was assessed using discrimination analysis employing Harrell’s c-index and calibration analysis with proportion tests.ResultsThe TCR training and testing datasets demonstrated stable and favorable predictive performance, with all Harrell’s c-index values ≥ 0.7 and almost all differences in proportion between the predicted and observed mortality being < 5%. In external validation, Asians exhibited the best performance compared with white and black populations, particularly in predicting OS, with all Harrell’s c-index values > 0.7.ConclusionsSurvival predictive disparities exist among different racial groups in HNCs. We have developed population-based prognostic models for Asians that can enhance clinical practice and treatment plans.

Machine learning and molecular simulation-based protocols to identify novel potential inhibitors for reverse transcriptase against HIV infections

Acquired immunodeficiency syndrome (AIDS) is a potentially fatal condition affecting the human immune system, which is attributed to the human immunodeficiency virus (HIV). The suppression of reverse transcriptase activity is a promising and feasible strategy for the therapeutic management of AIDS. In this study, we employed machine learning algorithms, such as support vector machines (SVM), k-nearest neighbor (k-NN), random forest (RF), and Gaussian naive base (GNB), which are fast and effective tools commonly used in drug design. For model training, we initially obtained a dataset of 5,159 compounds from BindingDB. The models were assessed using tenfold cross-validation to ensure their accuracy and reliability. Among these compounds, 1,645 compounds were labeled as active, having an IC50 below 0.49 µM, while 3,514 compounds were labeled “inactive against reverse transcriptase. Random forest achieved 86% accuracy on the train and test set among the different machine learning algorithms. Random forest model was then applied to an external ZINC dataset. Subsequently, only three hits-ZINC1359750464, ZINC1435357562, and ZINC1545719422-were selected based on the Lipinski Rule, docking score, and good interaction. The stability of these molecules was further evaluated by deploying molecular dynamics simulation and MM/GBSA, which were found to be −38.6013 ± 0.1103 kcal/mol for the Zidovudine/RT complex, −59.1761 ± 2.2926 kcal/mol for the ZINC1359750464/RT complex, −47.6292 ± 2.4206 kcal/mol for the ZINC1435357562/RT complex, and −50.7334 ± 2.5713 kcal/mol for the ZINC1545719422/RT complex. Communicated by Ramaswamy H. Sarma

Enhancing heart failure treatment decisions: interpretable machine learning models for advanced therapy eligibility prediction using EHR data

Timely and accurate referral of end-stage heart failure patients for advanced therapies, including heart transplants and mechanical circulatory support, plays an important role in improving patient outcomes and saving costs. However, the decision-making process is complex, nuanced, and time-consuming, requiring cardiologists with specialized expertise and training in heart failure and transplantation. In this study, we propose two logistic tensor regression-based models to predict patients with heart failure warranting evaluation for advanced heart failure therapies using irregularly spaced sequential electronic health records at the population and individual levels. The clinical features were collected at the previous visit and the predictions were made at the very beginning of the subsequent visit. Patient-wise ten-fold cross-validation experiments were performed. Standard LTR achieved an average F1 score of 0.708, AUC of 0.903, and AUPRC of 0.836. Personalized LTR obtained an F1 score of 0.670, an AUC of 0.869 and an AUPRC of 0.839. The two models not only outperformed all other machine learning models to which they were compared but also improved the performance and robustness of the other models via weight transfer. The AUPRC scores of support vector machine, random forest, and Naive Bayes are improved by 8.87%, 7.24%, and 11.38%, respectively. The two models can evaluate the importance of clinical features associated with advanced therapy referral. The five most important medical codes, including chronic kidney disease, hypotension, pulmonary heart disease, mitral regurgitation, and atherosclerotic heart disease, were reviewed and validated with literature and by heart failure cardiologists. Our proposed models effectively utilize EHRs for potential advanced therapies necessity in heart failure patients while explaining the importance of comorbidities and other clinical events. The information learned from trained model training could offer further insight into risk factors contributing to the progression of heart failure at both the population and individual levels.

Assisted design of data science pipelines

When designing data science (DS) pipelines, end-users can get overwhelmed by the large and growing set of available data preprocessing and modeling techniques. Intelligent discovery assistants (IDAs) and automated machine learning (AutoML) solutions aim to facilitate end-users by (semi-)automating the process. However, they are expensive to compute and yield limited applicability for a wide range of real-world use cases and application domains. This is due to (a) their need to execute thousands of pipelines to get the optimal one, (b) their limited support of DS tasks, e.g., supervised classification or regression only, and a small, static set of available data preprocessing and ML algorithms; and (c) their restriction to quantifiable evaluation processes and metrics, e.g., tenfold cross-validation using the ROC AUC score for classification. To overcome these limitations, we propose a human-in-the-loop approach for the assisteddesignofdatasciencepipelines using previously executed pipelines. Based on a user query, i.e., data and a DS task, our framework outputs a ranked list of pipeline candidates from which the user can choose to execute or modify in real time. To recommend pipelines, it first identifies relevant datasets and pipelines utilizing efficient similarity search. It then ranks the candidate pipelines using multi-objective sorting and takes user interactions into account to improve suggestions over time. In our experimental evaluation, the proposed framework significantly outperforms the state-of-the-art IDA tool and achieves similar predictive performance with state-of-the-art long-running AutoML solutions while being real-time, generic to any evaluation processes and DS tasks, and extensible to new operators.

A comparative study of explainable ensemble learning and logistic regression for predicting in-hospital mortality in the emergency department

This study addresses the challenges associated with emergency department (ED) overcrowding and emphasizes the need for efficient risk stratification tools to identify high-risk patients for early intervention. While several scoring systems, often based on logistic regression (LR) models, have been proposed to indicate patient illness severity, this study aims to compare the predictive performance of ensemble learning (EL) models with LR for in-hospital mortality in the ED. A cross-sectional single-center study was conducted at the ED of Imam Reza Hospital in northeast Iran from March 2016 to March 2017. The study included adult patients with one to three levels of emergency severity index. EL models using Bagging, AdaBoost, random forests (RF), Stacking and extreme gradient boosting (XGB) algorithms, along with an LR model, were constructed. The training and validation visits from the ED were randomly divided into 80% and 20%, respectively. After training the proposed models using tenfold cross-validation, their predictive performance was evaluated. Model performance was compared using the Brier score (BS), The area under the receiver operating characteristics curve (AUROC), The area and precision–recall curve (AUCPR), Hosmer–Lemeshow (H–L) goodness-of-fit test, precision, sensitivity, accuracy, F1-score, and Matthews correlation coefficient (MCC). The study included 2025 unique patients admitted to the hospital’s ED, with a total percentage of hospital deaths at approximately 19%. In the training group and the validation group, 274 of 1476 (18.6%) and 152 of 728 (20.8%) patients died during hospitalization, respectively. According to the evaluation of the presented framework, EL models, particularly Bagging, predicted in-hospital mortality with the highest AUROC (0.839, CI (0.802–0.875)) and AUCPR = 0.64 comparable in terms of discrimination power with LR (AUROC (0.826, CI (0.787–0.864)) and AUCPR = 0.61). XGB achieved the highest precision (0.83), sensitivity (0.831), accuracy (0.842), F1-score (0.833), and the highest MCC (0.48). Additionally, the most accurate models in the unbalanced dataset belonged to RF with the lowest BS (0.128). Although all studied models overestimate mortality risk and have insufficient calibration (P > 0.05), stacking demonstrated relatively good agreement between predicted and actual mortality. EL models are not superior to LR in predicting in-hospital mortality in the ED. Both EL and LR models can be considered as screening tools to identify patients at risk of mortality.

Prediction of xerostomia in elderly based on clinical characteristics and salivary flow rate with machine learning

Xerostomia may be accompanied by changes in salivary flow rate and the incidence increases in elderly. We aimed to use machine learning algorithms, to identify significant predictors for the presence of xerostomia. This study is the first to predict xerostomia with salivary flow rate in elderly based on artificial intelligence. In a cross-sectional study, 829 patients with oral discomfort were enrolled, and six features (sex, age, unstimulated and stimulated salivary flow rates (UFR and SFR, respectively), number of systemic diseases, and medication usage) were used in four machine learning algorithms to predict the presence of xerostomia. The incidence of xerostomia increased with age. The SFR was significantly higher than the UFR, and the UFR and SFR were significantly correlated. The UFR, but not SFR, decreased with age significantly. In patients more than 60 years of age, the UFR had a significantly higher predictive accuracy for xerostomia than the SFR. Using machine learning algorithms with tenfold cross-validation, the prediction accuracy increased significantly. In particular, the prediction accuracy of the multilayer perceptron (MLP) algorithm that combined UFR and SFR data was significantly better than either UFR or SFR individually. Moreover, when sex, age, number of systemic diseases, and number of medications were added to the MLP model, the prediction accuracy increased from 56 to 68%.

Choice of refractive surgery types for myopia assisted by machine learning based on doctors’ surgical selection data

In recent years, corneal refractive surgery has been widely used in clinics as an effective means to restore vision and improve the quality of life. When choosing myopia-refractive surgery, it is necessary to comprehensively consider the differences in equipment and technology as well as the specificity of individual patients, which heavily depend on the experience of ophthalmologists. In our study, we took advantage of machine learning to learn about the experience of ophthalmologists in decision-making and assist them in the choice of corneal refractive surgery in a new case. Our study was based on the clinical data of 7,081 patients who underwent corneal refractive surgery between 2000 and 2017 at the Department of Ophthalmology, Peking Union Medical College Hospital, Chinese Academy of Medical Sciences. Due to the long data period, there were data losses and errors in this dataset. First, we cleaned the data and deleted the samples of key data loss. Then, patients were divided into three groups according to the type of surgery, after which we used SMOTE technology to eliminate imbalance between groups. Six statistical machine learning models, including NBM, RF, AdaBoost, XGBoost, BP neural network, and DBN were selected, and a ten-fold cross-validation and grid search were used to determine the optimal hyperparameters for better performance. When tested on the dataset, the multi-class RF model showed the best performance, with agreement with ophthalmologist decisions as high as 0.8775 and Macro F1 as high as 0.8019. Furthermore, the results of the feature importance analysis based on the SHAP technique were consistent with an ophthalmologist’s practical experience. Our research will assist ophthalmologists in choosing appropriate types of refractive surgery and will have beneficial clinical effects.

Predicting outcomes following lower extremity open revascularization using machine learning

Lower extremity open revascularization is a treatment option for peripheral artery disease that carries significant peri-operative risks; however, outcome prediction tools remain limited. Using machine learning (ML), we developed automated algorithms that predict 30-day outcomes following lower extremity open revascularization. The National Surgical Quality Improvement Program targeted vascular database was used to identify patients who underwent lower extremity open revascularization for chronic atherosclerotic disease between 2011 and 2021. Input features included 37 pre-operative demographic/clinical variables. The primary outcome was 30-day major adverse limb event (MALE; composite of untreated loss of patency, major reintervention, or major amputation) or death. Our data were split into training (70%) and test (30%) sets. Using tenfold cross-validation, we trained 6 ML models. Overall, 24,309 patients were included. The primary outcome of 30-day MALE or death occurred in 2349 (9.3%) patients. Our best performing prediction model was XGBoost, achieving an area under the receiver operating characteristic curve (95% CI) of 0.93 (0.92–0.94). The calibration plot showed good agreement between predicted and observed event probabilities with a Brier score of 0.08. Our ML algorithm has potential for important utility in guiding risk mitigation strategies for patients being considered for lower extremity open revascularization to improve outcomes.

User-Specific Loyalty Measure and Prediction Using Deep Neural Network From Twitter Data

Analyzing the social media data posted by users reveals insights about the business such as customers’ sentiment, limitations, improvements in services, users’ expectations, customer loyalty, and satisfaction. Customer satisfaction and loyalty are measured using net promoter score (NPS), which is based on a survey and often lacks the background of customer context. Several recent methods, net brand reputation (NBR) and net sentiment score (NSS), overcome NPS limitations and utilize social media data to gauge customer loyalty. The NBR and NSS also have the limitation that they do not provide the user-level score. In social media, existing user has an influence on a new user, which cannot be estimated in NBR and NSS. The main contribution of this research is developing novel user-and product-level metrics, which measure the influence, satisfaction, and customer loyalty. User sentiment score (USS) measures customer loyalty, overall influence factor (OIF) computes influence factor, effective influence on user (EIU) gives the effective impact on a new user, and satisfaction score (SATS) estimates satisfaction using user tweets. Social data influence score (SDIS) is proposed to measure the combined impact of user satisfaction and influence. We prepare a Twitter dataset of five popular online learning platforms: Coursera, Udemy, Udacity, Khanacademy, and edX, and analyzed our proposed metrics on them. Prediction of USS into loyal and not loyal customers is carried out using Twitter data and dense neural networks (DNNs). The performance of DNN is compared with random forest classifier (RFC), eXtreme gradient boosting (XGB), and support vector classifier (SVC). DNN gave the highest classification of 98.62% on tenfold cross validation.