Accurately predicting the risk of unfavorable outcomes after endovascular coil therapy in patients with aneurysmal subarachnoid hemorrhage: an interpretable machine learning model.

BACKGROUND: Despite endovascular coiling as a valid modality in treatment of aneurysmal subarachnoid hemorrhage (aSAH), there is a risk of poor prognosis. However, the clinical utility of previously proposed early prediction tools remains limited. We aimed to develop a clinically generalizable machine learning (ML) models for accurately predicting unfavorable outcomes in aSAH patients after endovascular coiling. METHODS: Functional outcomes at 6 months after endovascular coiling were assessed via the modified Rankin Scale (mRS) and unfavorable outcomes were defined as mRS 3-6. Five ML algorithms (logistic regression, random forest, support vector machine, deep neural network, and extreme gradient boosting) were used for model development. The area under precision-recall curve (AUPRC) and receiver operating characteristic curve (AUROC) was used as main indices of model evaluation. SHapley Additive exPlanations (SHAP) method was applied to interpret the best-performing ML model. RESULTS: A total of 371 patients were eventually included into this study, and 85.4% of them had favorable outcomes. Among the five models, the DNN model had a better performance with AUPRC of 0.645 (AUROC of 0.905). Postoperative GCS score, size of aneurysm, and age were the top three powerful predictors. The further analysis of five random cases presented the good interpretability of the DNN model. CONCLUSION: Interpretable clinical prediction models based on different ML algorithms have been successfully constructed and validated, which would serve as reliable tools in optimizing the treatment decision-making of aSAH. Our DNN model had better performance to predict the unfavorable outcomes at 6 months in aSAH patients compared with Yan's nomogram model.

An interpretable artificial neural network model for predicting hypoxemia via an online tool in adult (18-64) patients during esophagogastroduodenoscopy.

Background: The hypoxemia risk in adult (18-64) patients treated with esophagogastroduodenoscopy (EGD) under sedation often poses a dilemma for anesthesiologists. We aimed to establish an artificial neural network (ANN) model to solve this problem, and introduce the Shapley additive explanations (SHAP) algorithm to further improve the interpretability. Methods: The relevant data of patients underwent routine anesthesia-assisted EGD were collected. Elastic network was used to filter the optimal features. Airway-ANN and Basic-ANN models were established based on all collected indicators and remaining variables excluding airway assessment indicators, respectively. The performance of Basic-ANN, Airway-ANN and STOP-BANG was evaluated by the area under the precision-recall curve (AUPRC) on temporal validation set. The SHAP was used for revealing the predictive behavior of our best model. Results: 999 patients were eventually included. The AUPRC value of Airway-ANN model was significantly higher than Basic-ANN model in the temporal validation set (0.532 vs 0.429, P < 0.05). And the performance of both two ANN models was significantly better than that of STOP-BANG score (both P < 0.05). The Airway-ANN model was deployed to the cloud (http://njfh-yxb.com.cn:2022/airway_ann). Conclusion: Our online interpretable Airway-ANN model achieved satisfying ability in identifying the hypoxemia risk in adult (18-64) patients undergoing EGD.

Dynamic prediction of hypoxemia risk at different time points based on preoperative and intraoperative features: machine learning applications in outpatients undergoing esophagogastroduodenoscopy.

BACKGROUND: Hypoxemia often occurs in outpatients undergoing anesthesia-assisted esophagogastroduodenoscopy (EGD). However, there is a scarcity in tools to predict the hypoxemia risk. We aimed to solve this problem by developing and validating machine learning (ML) models based on preoperative and intraoperative features. METHODS: All data were retrospectively collected from June 2021 to February 2022. The most appropriate predictive features were selected by the least absolute shrinkage and selection operator, which were incorporated and modelled by 4 ML algorithms. The area under the precision-recall curve (AUPRC) was used as the main evaluation metric to select the best models, and the selected models were compared with the STOP-BANG score. Their predictive performance was visually interpreted by SHapley Additive exPlanations. The primary endpoint of this study was hypoxemia during the procedure, defined as at least one reading of pulse oximetry < 90% without probes misplacement from the anesthesia induction beginning to the end of EGD, while the secondary endpoint was hypoxemia during induction, from the induction beginning to the start of endoscopic intubation. RESULTS: Of 1160 patients in the derivation cohort, 112 patients (9.6%) developed intraoperative hypoxemia, of which 102 (8.8%) occurred during the induction period. In temporal and external validation, no matter whether based on preoperative variables or still based on preoperative plus intraoperative variables, our models showed excellent predictive performance for the two endpoints, significantly better than STOP-BANG score. In the model interpretation section, preoperative variables (airway assessment indicators, pulse oximeter oxygen saturation and BMI) and intraoperative variables (the induced propofol dose) made the highest contribution to the predictions.To our knowledge, our ML models were the first to predict hypoxemia risk, which achieved excellent overall predictive ability integrating various clinical indicators. These models have the potential to become an effective tool for adjusting sedation strategies flexibly and reducing the workload of anesthesiologists.KEY MESSAGESThis study is the first model employing ML methods based on preoperative and preoperative plus intraoperative variables for predicting the risk of hypoxemia during induction and the whole EGD procedure respectively.Our four models achieved satisfactory predictive performance and outperformed STOP-BANG score in terms of AUPRC in the temporal and external validation cohorts respectively.We found that the relevant variables of airway assessment should be fully taken into account when analyzing the risk factor of hypoxemia, and the effect of patients' age on their hypoxemia risk should be considered in conjunction with the propofol dose.

Development and assessment of machine learning models for predicting recurrence risk after endovascular treatment in patients with intracranial aneurysms.

Intracranial aneurysms (IAs) remain a major public health concern and endovascular treatment (EVT) has become a major tool for managing IAs. However, the recurrence rate of IAs after EVT is relatively high, which may lead to the risk for aneurysm re-rupture and re-bleed. Thus, we aimed to develop and assess prediction models based on machine learning (ML) algorithms to predict recurrence risk among patients with IAs after EVT in 6 months. Patient population included patients with IAs after EVT between January 2016 and August 2019 in Hunan Provincial People's Hospital, and an adaptive synthetic (ADASYN) sampling approach was applied for the entire imbalanced dataset. We developed five ML models and assessed the models. In addition, we used SHapley Additive exPlanations (SHAP) and local interpretable model-agnostic explanation (LIME) algorithms to determine the importance of the selected features and interpret the ML models. A total of 425 IAs were enrolled into this study, and 66 (15.5%) of which recurred in 6 months. Among the five ML models, gradient boosting decision tree (GBDT) model performed best. The area under curve (AUC) of the GBDT model on the testing set was 0.842 (sensitivity: 81.2%; specificity: 70.4%). Our study firstly demonstrated that ML-based models can serve as a reliable tool for predicting recurrence risk in patients with IAs after EVT in 6 months, and the GBDT model showed the optimal prediction performance.

Prediction-Driven Decision Support for Patients With Mild Stroke: A Model Based on Machine Learning Algorithms.

Background and Purpose: Treatment for mild stroke remains an open question. We aim to develop a decision support tool based on machine learning (ML) algorithms, called DAMS (Disability After Mild Stroke), to identify mild stroke patients who would be at high risk of post-stroke disability (PSD) if they only received medical therapy and, more importantly, to aid neurologists in making individual clinical decisions in emergency contexts. Methods: Ischemic stroke patients were prospectively recorded in the National Advanced Stroke Center of Nanjing First Hospital (China) between July 2016 and September 2020. The exclusion criteria were patients who received thrombolytic therapy, age <18 years, lack of 3-month modified Rankin Scale (mRS), disabled before the index stroke, with an admission National Institute of Health stroke scale (NIHSS) > 5. The primary outcome was PSD, corresponding to 3-month mRS ≥ 2. We developed five ML models and assessed the area under curve (AUC) of receiver operating characteristic, calibration curve, and decision curve analysis. The optimal ML model was selected to be DAMS. In addition, SHapley Additive exPlanations (SHAP) approach was introduced to rank the feature importance. Finally, rapid-DAMS (R-DAMS) was constructed for a more urgent situation based on DAMS. Results: A total of 1,905 mild stroke patients were enrolled in this study, and patients with PSD accounted for 23.4% (447). There was no difference in AUCs between the five models (ranged from 0.691 to 0.823). Although there was similar discriminative performance between ML models, the support vector machine model exhibited higher net benefit and better calibration (Brier score, 0.159, calibration slope, 0.935, calibration intercept, 0.035). Therefore, this model was selected for DAMS. In addition, SHAP approach showed that the most crucial feature was NIHSS on admission. Finally, R-DAMS was constructed and there was similar discriminative performance between R-DAMS and DAMS, but the former performed worse on calibration. Conclusions: DAMS and R-DAMS, as prediction-driven decision support tools, were designed to aid clinical decision-making for mild stroke patients in emergency contexts. In addition, even within a narrow range of baseline scores, NIHSS on admission is the strongest feature that contributed to the prediction.