Prediction models for postoperative delirium in elderly patients with machine-learning algorithms and SHapley Additive exPlanations

Translational Psychiatry, Год журнала: 2024, Номер 14(1)

Опубликована: Янв. 25, 2024

Abstract Postoperative delirium (POD) is a common and severe complication in elderly patients with hip fractures. Identifying high-risk POD can help improve the outcome of We conducted retrospective study on (≥65 years age) who underwent orthopedic surgery fracture between January 2014 August 2019. Conventional logistic regression five machine-learning algorithms were used to construct prediction models POD. A nomogram for was built method. The area under receiver operating characteristic curve (AUC-ROC), accuracy, sensitivity, precision calculated evaluate different models. Feature importance individuals interpreted using Shapley Additive Explanations (SHAP). About 797 enrolled study, incidence at 9.28% (74/797). age, renal insufficiency, chronic obstructive pulmonary disease (COPD), use antipsychotics, lactate dehydrogenase (LDH), C-reactive protein are build an AUC 0.71. AUCs 0.81 (Random Forest), 0.80 (GBM), 0.68 (AdaBoost), 0.77 (XGBoost), 0.70 (SVM). sensitivities six range from 68.8% (logistic SVM) 91.9% Forest). precisions 18.3% regression) 67.8% Six fractures constructed algorithms. application could provide convenient risk stratification benefit patients.

Язык: Английский

---

Development of a Medication-Related Osteonecrosis of the Jaw Prediction Model Using the FDA Adverse Event Reporting System Database and Machine Learning

Pharmaceuticals, Год журнала: 2025, Номер 18(3), С. 423 - 423

Опубликована: Март 17, 2025

Background: Medication-related osteonecrosis of the jaw (MRONJ) is a rare but serious adverse event. Herein, we conducted quantitative structure–activity relationship analysis using U.S. Food and Drug Administration Adverse Reaction Database System (FAERS) machine learning to construct drug prediction model for MRONJ induction based solely on chemical structure information. Methods: A total 4815 drugs from FAERS were evaluated, including 70 139 MRONJ-positive MRONJ-negative drugs, respectively, identified reporting odds ratios, Fisher’s exact tests, ≥100 event reports. Then, calculated 326 descriptors each compared three supervised algorithms (random forest, gradient boosting, artificial neural networks). We also number (5, 6, 7, 8, 9, 10, 20, 30 descriptors). Results: indicated that an network algorithm eight achieved highest validation receiver operating characteristic curve value 0.778. Notably, polar surface area (ASA_P) was among top-ranking descriptors, such as bisphosphonates anticancer showed high values. Our final demonstrated balanced accuracy 0.693 specificity 0.852. Conclusions: In this study, our MRONJ-inducing with properties potential causes MRONJ. This study demonstrates promising approach predicting risk, which could enhance safety assessment streamline screening in clinical preclinical settings.

Язык: Английский

---

Comparison of logistic regression and machine learning methods for predicting postoperative delirium in elderly patients: A retrospective study

CNS Neuroscience & Therapeutics, Год журнала: 2022, Номер 29(1), С. 158 - 167

Опубликована: Окт. 11, 2022

Abstract Aims To compare the performance of logistic regression and machine learning methods in predicting postoperative delirium (POD) elderly patients. Method This was a retrospective study perioperative medical data from patients undergoing non‐cardiac non‐neurology surgery over 65 years old January 2014 to August 2019. Forty‐six variables were used predict POD. A traditional five models (Random Forest, GBM, AdaBoost, XGBoost, stacking ensemble model) compared by area under receiver operating characteristic curve (AUC‐ROC), sensitivity, specificity, precision. Results In total, 29,756 enrolled, incidence POD 3.22% after variable screening. AUCs 0.783 (0.765–0.8) for method, 0.78 random forest, 0.76 0.74 0.73 0.77 model. The respective sensitivities 6 aforementioned 74.2%, 72.2%, 76.8%, 63.6%, 71.6%, 67.4%. specificities 70.7%, 99.8%, 96.5%, 98.8%, 96.1%. precision values 7.8%, 52.3%, 55.6%, 57%, 54.5%, 56.4%. Conclusions optimal application model could provide quick convenient risk identification help improve management surgical because its better fewer variables, easier interpretability than

Язык: Английский

---

Prediction of Complications and Prognostication in Perioperative Medicine: A Systematic Review and PROBAST Assessment of Machine Learning Tools

Anesthesiology, Год журнала: 2023, Номер 140(1), С. 85 - 101

Опубликована: Ноя. 9, 2023

Background The utilization of artificial intelligence and machine learning as diagnostic predictive tools in perioperative medicine holds great promise. Indeed, many studies have been performed recent years to explore the potential. purpose this systematic review is assess current state medicine, its utility prediction complications prognostication, limitations related bias validation. Methods A multidisciplinary team clinicians engineers conducted a using Preferred Reporting Items for Systematic Review Meta-Analysis (PRISMA) protocol. Multiple databases were searched, including Scopus, Cumulative Index Nursing Allied Health Literature (CINAHL), Cochrane Library, PubMed, Medline, Embase, Web Science. focused on study design, type model used, validation techniques applied, reported performance prognostication. This further classified outcomes applications an ad hoc classification system. Prediction Risk Of Bias Assessment Tool (PROBAST) was used risk applicability studies. Results total 103 identified. models literature primarily based single-center validations (75%), with only 13% being externally validated across multiple centers. Most mortality demonstrated limited ability discriminate classify effectively. PROBAST assessment indicated high errors predicted or applications. Conclusions findings indicate that development field still early stages. indicates application at stage. While suggest potential utility, several key challenges must be first overcome before their introduction into clinical practice. Editor’s Perspective What We Already Know about Topic Article Tells Us That Is New

Язык: Английский

---

Predicting risk of preterm birth in singleton pregnancies using machine learning algorithms

Frontiers in Big Data, Год журнала: 2024, Номер 7

Опубликована: Фев. 29, 2024

We aimed to develop, train, and validate machine learning models for predicting preterm birth (<37 weeks' gestation) in singleton pregnancies at different gestational intervals. Models were developed based on complete data from 22,603 a prospective population-based cohort study that was conducted 51 midwifery clinics hospitals Wenzhou City of China between 2014 2016. applied Catboost, Random Forest, Stacked Model, Deep Neural Networks (DNN), Support Vector Machine (SVM) algorithms, as well logistic regression, conduct feature selection predictive modeling. Feature implemented permutation-based importance lists derived the including all features, using balanced training set. To develop prediction models, top 10%, 25%, 50% most important features selected. Prediction with set 5-fold cross-validation internal validation. Model performance assessed area under receiver operating curve (AUC) values. The CatBoost-based model after 26 gestation performed best an AUC value 0.70 (0.67, 0.73), accuracy 0.81, sensitivity 0.47, specificity 0.83. Number antenatal care visits before 24 gestation, aspartate aminotransferase level registration, symphysis fundal height, maternal weight, abdominal circumference, blood pressure emerged strong predictors completed weeks. application pregnancy surveillance is promising approach predict we identified several modifiable predictors.

Язык: Английский

---

Interpretable machine learning model for early prediction of delirium in elderly patients following intensive care unit admission: a derivation and validation study

Frontiers in Medicine, Год журнала: 2024, Номер 11

Опубликована: Май 17, 2024

Background and objective Delirium is the most common neuropsychological complication among older adults admitted to intensive care unit (ICU) often associated with a poor prognosis. This study aimed construct validate an interpretable machine learning (ML) for early delirium prediction in ICU patients. Methods was retrospective observational cohort patient data were extracted from Medical Information Mart Intensive Care-IV database. Feature variables delirium, including predisposing factors, disease-related iatrogenic environmental selected using least absolute shrinkage selection operator regression, models built logistic decision trees, support vector machines, extreme gradient boosting (XGBoost), k-nearest neighbors naive Bayes methods. Multiple metrics used evaluation of performance models, area under receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, recall, F1 score, calibration plot, analysis. SHapley Additive exPlanations (SHAP) improve interpretability final model. Results Nine thousand seven hundred forty-eight aged 65 years or included Twenty-six features ML models. Among compared, XGBoost model demonstrated best highest AUC (0.836), accuracy (0.765), sensitivity (0.713), recall score (0.725) training set. It also exhibited excellent discrimination 0.810, good calibration, had net benefit validation cohort. The SHAP summary analysis showed that Glasgow Coma Scale, mechanical ventilation, sedation top three risk outcome prediction. dependency plot force interpreted at both factor level individual level, respectively. Conclusion reliable tool predicting critical elderly By combining SHAP, it can provide clear explanations personalized more intuitive understanding effect key establishment such would facilitate assessment prompt intervention delirium.

Язык: Английский

---

A Supervised Explainable Machine Learning Model for Perioperative Neurocognitive Disorder in Liver-Transplantation Patients and External Validation on the Medical Information Mart for Intensive Care IV Database: Retrospective Study

Journal of Medical Internet Research, Год журнала: 2025, Номер 27, С. e55046 - e55046

Опубликована: Янв. 15, 2025

Background Patients undergoing liver transplantation (LT) are at risk of perioperative neurocognitive dysfunction (PND), which significantly affects the patients’ prognosis. Objective This study used machine learning (ML) algorithms with an aim to extract critical predictors and develop ML model predict PND among LT recipients. Methods In this retrospective study, data from 958 patients who underwent between January 2015 2020 were extracted Third Affiliated Hospital Sun Yat-sen University. Six post-LT PND, performance was evaluated using area under receiver operating curve (AUC), accuracy, sensitivity, specificity, F1-scores. The best-performing additionally validated a temporal external dataset including 309 cases February August 2022, independent Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ) database 325 patients. Results development cohort, 201 out 751 (33.5%) diagnosed PND. logistic regression achieved highest AUC (0.799) in internal validation set, comparable (0.826) MIMIC-Ⅳ sets (0.72). top 3 features contributing diagnosis preoperative overt hepatic encephalopathy, platelet level, postoperative sequential organ failure assessment score, as revealed by Shapley additive explanations method. Conclusions A real-time model-based online predictor developed, providing highly interoperable tool use across medical institutions support early stratification decision making

Язык: Английский

---

Machine learning or traditional statistical methods for predictive modelling in perioperative medicine: A narrative review

Journal of Clinical Anesthesia, Год журнала: 2025, Номер 102, С. 111782 - 111782

Опубликована: Фев. 19, 2025

Язык: Английский

---

Comparison of machine learning and nomogram to predict 30-day in-hospital mortality in patients with acute myocardial infarction combined with cardiogenic shock: a retrospective study based on the eICU-CRD and MIMIC-IV databases

BMC Cardiovascular Disorders, Год журнала: 2025, Номер 25(1)

Опубликована: Март 19, 2025

To evaluate the predictive utility of machine learning and nomogram in predicting in-hospital mortality patients with acute myocardial infarction complicated by cardiogenic shock (AMI-CS), to visualize model results order analyze impact these predictors on patients' prognosis. A retrospective analysis was conducted 332 adult who were diagnosed AMI-CS admitted ICU for first time within eICU Collaborative Research Database (eICU-CRD). AdaBoost, XGBoost, LightGBM, Random Forest logistic regression developed utilizing random forest recursive elimination (RF-RFE) least absolute shrinkage selection operator (LASSO) algorithms feature selection. Compared models, demonstrated superior accuracy AMI-CS, an AUC value 0.869 (95% CI: 0.803, 0.883) F1 score 0.897 internal test set nomogram, 0.770 0.702, 0.801) 0.832 external validation set. Nomogram enhance interpretability transparency leading more reliable prognostic predictions patients. This facilitates clinicians making precise decisions, thereby enhancing patient

Язык: Английский

---

Enhanced machine learning predictive modeling for delirium in elderly ICU patients with COPD and respiratory failure: A retrospective study based on MIMIC-IV

PLoS ONE, Год журнала: 2025, Номер 20(3), С. e0319297 - e0319297

Опубликована: Март 20, 2025

Background and objective Elderly patients with Chronic obstructive pulmonary disease (COPD) respiratory failure admitted to the intensive care unit (ICU) have a poor prognosis, occurrence of delirium further worsens outcomes increases hospitalization costs. This study aimed develop predictive model for in this patient population identify associated risk factors Methods Data machine learning were obtained from MIMIC-IV database. Feature variable screening was conducted using Lasso regression best subset method. Four models—K-nearest neighbor, random forest, logistic regression, extreme gradient boosting (XGBoost)—were trained optimized predict risk. The stability is evaluated ten-fold cross validation effectiveness on set accuracy, F1 score, precision recall. SHapley Additive exPlanations (SHAP) method used explain importance each model. Results A total 1,155 between 2008 2019 included study, incidence 12.9% (149/1,155). Among four ML models evaluated, XGBoost demonstrated discriminative ability. In set, it achieved an AUC 0.932, indicating superior performance high precision, recall, scores 0.891, 0.839, 0.795, 0.810, respectively. Key features identified through SHAP analysis Glasgow Coma Scale (GCS) verbal length hospital stay, mean SpO₂ first day ICU admission, Modification Diet Renal Disease (MDRD) equation diastolic blood pressure, GCS motor gender, duration noninvasive ventilation. These findings provide valuable insights individualized management. Conclusions developed prediction effectively predicts elderly COPD ICU. can assist clinical decision-making, potentially improving reducing healthcare

Язык: Английский

---