Background: Early prediction of hospital mortality is crucial for ICU patients with sepsis. This study aimed to develop a novel blending machine learning (ML) model for hospital mortality prediction in ICU patients with sepsis. Methods: Two ICU databases were employed: eICU Collaborative Research Database (eICU-CRD) and Medical Information Mart for Intensive Care III (MIMIC-III). All adult patients who fulfilled Sepsis-3 criteria were identified. Samples from eICU-CRD constituted training set and samples from MIMIC-III constituted test set. Stepwise logistic regression model was used for predictor selection. Blending ML model which integrated nine sorts of basic ML models was developed for hospital mortality prediction in ICU patients with sepsis. Model performance was evaluated by various measures related to discrimination or calibration. Results: Twelve thousand five hundred fifty-eight patients from eICU-CRD were included as the training set, and 12,095 patients from MIMIC-III were included as the test set. Both the training set and the test set showed a hospital mortality of 17.9%. Maximum and minimum lactate, maximum and minimum albumin, minimum PaO2/FiO2 and age were important predictors identified by both random forest and extreme gradient boosting algorithm. Blending ML models based on corresponding set of predictors presented better discrimination than SAPS II (AUROC, 0.806 vs. 0.771; AUPRC 0.515 vs. 0.429) and SOFA (AUROC, 0.742 vs. 0.706; AUPRC 0.428 vs. 0.381) on the test set. In addition, calibration curves showed that blending ML models had better calibration than SAPS II. Conclusions: The blending ML model is capable of integrating different sorts of basic ML models efficiently, and outperforms conventional severity scores in predicting hospital mortality among septic patients in ICU.

Background: Clinical decision of extubation is a challenge in the treatment of patient with invasive mechanical ventilation (IMV), since existing extubation protocols are not capable of precisely predicting extubation failure (EF). This study aims to develop and validate interpretable recurrent neural network (RNN) models for dynamically predicting EF risk. Methods: A retrospective cohort study was conducted on IMV patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Time series with a 4-h resolution were built for all included patients. Two types of RNN models, the long short-term memory (LSTM) and the gated recurrent unit (GRU), were developed. A stepwise logistic regression model was used to select key features for developing light-version RNN models. The RNN models were compared to other five non-temporal machine learning models. The Shapley additive explanations (SHAP) value was applied to explain the influence of the features on model prediction. Results: Of 8,599 included patients, 2,609 had EF (30.3%). The area under receiver operating characteristic curve (AUROC) of LSTM and GRU showed no statistical difference on the test set (0.828 vs. 0.829). The light-version RNN models based on the 26 features selected out of a total of 89 features showed comparable performance as their corresponding full-version models. Among the non-temporal models, only the random forest (RF) (AUROC: 0.820) and the extreme gradient boosting (XGB) model (AUROC: 0.823) were comparable to the RNN models, but their calibration was deviated. Conclusions: The RNN models have excellent predictive performance for predicting EF risk and have potential to become real-time assistant decision-making systems for extubation.

Background: Although the machine learning model developed on electronic health records has become a promising method for early predicting hospital mortality, few studies focus on the approaches for handling missing data in electronic health records and evaluate model robustness to data missingness. This study proposes an attention architecture that shows excellent predictive performance and is robust to data missingness. Methods: Two public intensive care unit databases were used for model training and external validation, respectively. Three neural networks (masked attention model, attention model with imputation, attention model with missing indicator) based on the attention architecture were developed, using masked attention mechanism, multiple imputation, and missing indicator to handle missing data, respectively. Model interpretability was analyzed by attention allocations. Extreme gradient boosting, logistic regression with multiple imputation and missing indicator (logistic regression with imputation, logistic regression with missing indicator) were used as baseline models. Model discrimination and calibration were evaluated by area under the receiver operating characteristic curve, area under precision-recall curve, and calibration curve. In addition, model robustness to data missingness in both model training and validation was evaluated by three analyses. Results: In total, 65,623 and 150,753 intensive care unit stays were respectively included in the training set and the test set, with mortality of 10.1% and 8.5%, and overall missing rate of 10.3% and 19.7%. attention model with missing indicator had the highest area under the receiver operating characteristic curve (0.869; 95% CI: 0.865 to 0.873) in external validation; attention model with imputation had the highest area under precision-recall curve (0.497; 95% CI: 0.480–0.513). Masked attention model and attention model with imputation showed better calibration than other models. The three neural networks showed different patterns of attention allocation. In terms of robustness to data missingness, masked attention model and attention model with missing indicator are more robust to missing data in model training; while attention model with imputation is more robust to missing data in model validation. Conclusions: The attention architecture has the potential to become an excellent model architecture for clinical prediction task with data missingness.

Functional deterioration (FD) of various organ systems is the major cause of death in ICU patients, but few studies propose effective multi-task (MT) model to predict FD of multiple organs simultaneously. This study propose a MT deep learning model named inter-organ correlation based multi-task model (IOC-MT), to dynamically predict FD in six organ systems. Three public ICU databases were used for model training and validation. The IOC-MT was designed based on the routine MT deep learning framework, but it used a Graph Attention Networks (GAT) module to capture inter-organ correlation and an adaptive adjustment mechanism (AAM) to adjust prediction. We compared the IOC-MT to five single-task (ST) baseline models, including three deep models (LSTM-ST, GRU-ST, Transformer-ST) and two machine learning models (GRU-ST, RF-ST), and performed ablation study to assess the contribution of important components in IOC-MT. Model discrimination was evaluated by AUROC and AUPRC, and model calibration was assessed by the calibration curve. The attention weight and adjustment coefficient were analyzed at both overall and individual level to show the AAM of IOC-MT. The IOC-MT had comparable discrimination and calibration to LSTM-ST, GRU-ST and Transformer-ST for most organs under different gap windows in the internal and external validation, and obviously outperformed GRU-ST, RF-ST. The ablation study showed that the GAT, AAM and missing indicator could improve the overall performance of the model. Furthermore, the inter-organ correlation and prediction adjustment of IOC-MT were intuitive and comprehensible, and also had biological plausibility. The IOC-MT is a promising MT model for dynamically predicting FD in six organ systems. It can capture inter-organ correlation and adjust the prediction for one organ based on aggregated information from the other organs.