Predictive approach for liberation from acute dialysis in ICU patients using interpretable machine learning.

Renal recovery following dialysis-requiring acute kidney injury (AKI-D) is a vital clinical outcome in critical care, yet it remains an understudied area. This retrospective cohort study, conducted in a medical center in Taiwan from 2015 to 2020, enrolled patients with AKI-D during intensive care unit stays. We aimed to develop and temporally test models for predicting dialysis liberation before hospital discharge using machine learning algorithms and explore early predictors. The dataset comprised 90 routinely collected variables within the first three days of dialysis initiation. Out of 1,381 patients who received acute dialysis, 27.3% experienced renal recovery. The cohort was divided into the training group (N = 1135) and temporal testing group (N = 251). The models demonstrated good performance, with an area under the receiver operating characteristic curve of 0.85 (95% CI, 0.81-0.88) and an area under the precision-recall curve of 0.69 (95% CI, 0.62-0.76) for the XGBoost model. Key predictors included urine volume, Charlson comorbidity index, vital sign derivatives (trend of respiratory rate and SpO2), and lactate levels. We successfully developed early prediction models for renal recovery by integrating early changes in vital signs and inputs/outputs, which have the potential to aid clinical decision-making in the ICU.

Prolonged use of neuromuscular blocking agents is associated with increased long-term mortality in mechanically ventilated medical ICU patients: a retrospective cohort study.

BACKGROUND: Neuromuscular blockade agents (NMBAs) can be used to facilitate mechanical ventilation in critically ill patients. Accumulating evidence has shown that NMBAs may be associated with intensive care unit (ICU)-acquired weakness and poor outcomes. However, the long-term impact of NMBAs on mortality is still unclear. METHODS: We conducted a retrospective analysis using the 2015-2019 critical care databases at Taichung Veterans General Hospital, a referral center in central Taiwan, as well as the Taiwan nationwide death registry profile. RESULTS: A total of 5709 ventilated patients were eligible for further analysis, with 63.8% of them were male. The mean age of enrolled subjects was 67.8 ± 15.8 years, and the one-year mortality was 48.3% (2755/5709). Compared with the survivors, the non-survivors had a higher age (70.4 ± 14.9 vs 65.4 ± 16.3, p < 0.001), Acute Physiology and Chronic Health Evaluation II score (28.0 ± 6.2 vs 24.7 ± 6.5, p < 0.001), a longer duration of ventilator use (12.6 ± 10.6 days vs 7.8 ± 8.5 days, p < 0.001), and were more likely to receive NMBAs for longer than 48 h (11.1% vs 7.8%, p < 0.001). After adjusting for age, sex, and relevant covariates, the use of NMBAs for longer than 48 h was found to be independently associated with an increased risk of mortality (adjusted HR: 1.261; 95% CI: 1.07-1.486). The analysis of effect modification revealed that this association was tended to be strong in patients with a Charlson Comorbidity Index of 3 or higher. CONCLUSIONS: Our study demonstrated that prolonged use of NMBAs was associated with an increased risk of long-term mortality in critically ill patients requiring mechanical ventilation. Further studies are needed to validate our findings.

Federated machine learning for predicting acute kidney injury in critically ill patients: a multicenter study in Taiwan.

Purpose: To address the contentious data sharing across hospitals, this study adopted a novel approach, federated learning (FL), to establish an aggregate model for acute kidney injury (AKI) prediction in critically ill patients in Taiwan. Methods: This study used data from the Critical Care Database of Taichung Veterans General Hospital (TCVGH) from 2015 to 2020 and electrical medical records of the intensive care units (ICUs) between 2018 and 2020 of four referral centers in different areas across Taiwan. AKI prediction models were trained and validated thereupon. An FL-based prediction model across hospitals was then established. Results: The study included 16,732 ICU admissions from the TCVGH and 38,424 ICU admissions from the other four hospitals. The complete model with 60 features and the parsimonious model with 21 features demonstrated comparable accuracies using extreme gradient boosting, neural network (NN), and random forest, with an area under the receiver-operating characteristic (AUROC) curve of approximately 0.90. The Shapley Additive Explanations plot demonstrated that the selected features were the key clinical components of AKI for critically ill patients. The AUROC curve of the established parsimonious model for external validation at the four hospitals ranged from 0.760 to 0.865. NN-based FL slightly improved the model performance at the four centers. Conclusion: A reliable prediction model for AKI in ICU patients was developed with a lead time of 24 h, and it performed better when the novel FL platform across hospitals was implemented. Supplementary Information: The online version contains supplementary material available at 10.1007/s13755-023-00248-5.

Adaptive Autonomous Protocol for Secured Remote Healthcare Using Fully Homomorphic Encryption (AutoPro-RHC).

The outreach of healthcare services is a challenge to remote areas with affected populations. Fortunately, remote health monitoring (RHM) has improved the hospital service quality and has proved its sustainable growth. However, the absence of security may breach the health insurance portability and accountability act (HIPAA), which has an exclusive set of rules for the privacy of medical data. Therefore, the goal of this work is to design and implement the adaptive Autonomous Protocol (AutoPro) on the patient's remote healthcare (RHC) monitoring data for the hospital using fully homomorphic encryption (FHE). The aim is to perform adaptive autonomous FHE computations on recent RHM data for providing health status reporting and maintaining the confidentiality of every patient. The autonomous protocol works independently within the group of prime hospital servers without the dependency on the third-party system. The adaptiveness of the protocol modes is based on the patient's affected level of slight, medium, and severe cases. Related applications are given as glucose monitoring for diabetes, digital blood pressure for stroke, pulse oximeter for COVID-19, electrocardiogram (ECG) for cardiac arrest, etc. The design for this work consists of an autonomous protocol, hospital servers combining multiple prime/local hospitals, and an algorithm based on fast fully homomorphic encryption over the torus (TFHE) library with a ring-variant by the Gentry, Sahai, and Waters (GSW) scheme. The concrete-ML model used within this work is trained using an open heart disease dataset from the UCI machine learning repository. Preprocessing is performed to recover the lost and incomplete data in the dataset. The concrete-ML model is evaluated both on the workstation and cloud server. Also, the FHE protocol is implemented on the AWS cloud network with performance details. The advantages entail providing confidentiality to the patient's data/report while saving the travel and waiting time for the hospital services. The patient's data will be completely confidential and can receive emergency services immediately. The FHE results show that the highest accuracy is achieved by support vector classification (SVC) of 88% and linear regression (LR) of 86% with the area under curve (AUC) of 91% and 90%, respectively. Ultimately, the FHE-based protocol presents a novel system that is successfully demonstrated on the cloud network.

Machine Learning Models to Predict the Risk of Rapidly Progressive Kidney Disease and the Need for Nephrology Referral in Adult Patients with Type 2 Diabetes.

Early detection of rapidly progressive kidney disease is key to improving the renal outcome and reducing complications in adult patients with type 2 diabetes mellitus (T2DM). We aimed to construct a 6-month machine learning (ML) predictive model for the risk of rapidly progressive kidney disease and the need for nephrology referral in adult patients with T2DM and an initial estimated glomerular filtration rate (eGFR) ≥ 60 mL/min/1.73 m2. We extracted patients and medical features from the electronic medical records (EMR), and the cohort was divided into a training/validation and testing data set to develop and validate the models on the basis of three algorithms: logistic regression (LR), random forest (RF), and extreme gradient boosting (XGBoost). We also applied an ensemble approach using soft voting classifier to classify the referral group. We used the area under the receiver operating characteristic curve (AUROC), precision, recall, and accuracy as the metrics to evaluate the performance. Shapley additive explanations (SHAP) values were used to evaluate the feature importance. The XGB model had higher accuracy and relatively higher precision in the referral group as compared with the LR and RF models, but LR and RF models had higher recall in the referral group. In general, the ensemble voting classifier had relatively higher accuracy, higher AUROC, and higher recall in the referral group as compared with the other three models. In addition, we found a more specific definition of the target improved the model performance in our study. In conclusion, we built a 6-month ML predictive model for the risk of rapidly progressive kidney disease. Early detection and then nephrology referral may facilitate appropriate management.

Explainable machine learning approach to predict extubation in critically ill ventilated patients: a retrospective study in central Taiwan.

BACKGROUND: Weaning from mechanical ventilation (MV) is an essential issue in critically ill patients, and we used an explainable machine learning (ML) approach to establish an extubation prediction model. METHODS: We enrolled patients who were admitted to intensive care units during 2015-2019 at Taichung Veterans General Hospital, a referral hospital in central Taiwan. We used five ML models, including extreme gradient boosting (XGBoost), categorical boosting (CatBoost), light gradient boosting machine (LightGBM), random forest (RF) and logistic regression (LR), to establish the extubation prediction model, and the feature window as well as prediction window was 48 h and 24 h, respectively. We further employed feature importance, Shapley additive explanations (SHAP) plot, partial dependence plot (PDP) and local interpretable model-agnostic explanations (LIME) for interpretation of the model at the domain, feature, and individual levels. RESULTS: We enrolled 5,940 patients and found the accuracy was comparable among XGBoost, LightGBM, CatBoost and RF, with the area under the receiver operating characteristic curve using XGBoost to predict extubation was 0.921. The calibration and decision curve analysis showed well applicability of models. We also used the SHAP summary plot and PDP plot to demonstrate discriminative points of six key features in predicting extubation. Moreover, we employed LIME and SHAP force plots to show predicted probabilities of extubation and the rationale of the prediction at the individual level. CONCLUSIONS: We developed an extubation prediction model with high accuracy and visualised explanations aligned with clinical workflow, and the model may serve as an autonomous screen tool for timely weaning.

Artificial intelligence-aided diagnosis model for acute respiratory distress syndrome combining clinical data and chest radiographs.

Objective: The aim of this study was to develop an artificial intelligence-based model to detect the presence of acute respiratory distress syndrome (ARDS) using clinical data and chest X-ray (CXR) data. Method: The transfer learning method was used to train a convolutional neural network (CNN) model with an external image dataset to extract the image features. Then, the last layer of the model was fine-tuned to determine the probability of ARDS. The clinical data were trained using three machine learning algorithms-eXtreme Gradient Boosting (XGB), random forest (RF), and logistic regression (LR)-to estimate the probability of ARDS. Finally, ensemble-weighted methods were proposed that combined the image model and the clinical data model to estimate the probability of ARDS. An analysis of the importance of clinical features was performed to explore the most important features in detecting ARDS. A gradient-weighted class activation mapping (Grad-CAM) model was used to explain what our CNN sees and understands when making a decision. Results: The proposed ensemble-weighted methods improved the performances of the ARDS classifiers (XGB + CNN, area under the curve [AUC] = 0.916; RF + CNN, AUC = 0.920; LR + CNN, AUC = 0.920; XGB + RF + LR + CNN, AUC = 0.925). In addition, the ML model using clinical data to present the top 15 important features to identify the risk factors of ARDS. Conclusion: This study developed combined machine learning models with clinical data and CXR images to detect ARDS. According to the results of the Shapley Additive exPlanations values and the Grad-CAM techniques, an explicable ARDS diagnosis model is suitable for a real-life scenario.

Multi-Modal Data Analysis for Pneumonia Status Prediction Using Deep Learning (MDA-PSP).

Evaluating several vital signs and chest X-ray (CXR) reports regularly to determine the recovery of the pneumonia patients at general wards is a challenge for doctors. A recent study shows the identification of pneumonia by the history of symptoms and signs including vital signs, CXR, and other clinical parameters, but they lack predicting the recovery status after starting treatment. The goal of this paper is to provide a pneumonia status prediction system for the early affected patient's discharge from the hospital within 7 days or late discharge more than 7 days. This paper aims to design a multimodal data analysis for pneumonia status prediction using deep learning classification (MDA-PSP). We have developed a system that takes an input of vital signs and CXR images of the affected patient with pneumonia from admission day 1 to day 3. The deep learning then classifies the health status improvement or deterioration for predicting the possible discharge state. Therefore, the scope is to provide a highly accurate prediction of the pneumonia recovery on the 7th day after 3-day treatment by the SHAP (SHapley Additive exPlanation), imputation, adaptive imputation-based preprocessing of the vital signs, and CXR image feature extraction using deep learning based on dense layers-batch normalization (BN) with class weights for the first 7 days' general ward patient in MDA-PSP. A total of 3972 patients with pneumonia were enrolled by de-identification with an adult age of 71 mean ± 17 sd and 64% of them were male. After analyzing the data behavior, appropriate improvement measures are taken by data preprocessing and feature vectorization algorithm. The deep learning method of Dense-BN with SHAP features has an accuracy of 0.77 for vital signs, 0.92 for CXR, and 0.75 for the combined model with class weights. The MDA-PSP hybrid method-based experiments are proven to demonstrate higher prediction accuracy of 0.75 for pneumonia patient status. Henceforth, the hybrid methods of machine and deep learning for pneumonia patient discharge are concluded to be a better approach.

Explainable machine learning to predict long-term mortality in critically ill ventilated patients: a retrospective study in central Taiwan.

BACKGROUND: Machine learning (ML) model is increasingly used to predict short-term outcome in critically ill patients, but the study for long-term outcome is sparse. We used explainable ML approach to establish 30-day, 90-day and 1-year mortality prediction model in critically ill ventilated patients. METHODS: We retrospectively included patients who were admitted to intensive care units during 2015-2018 at a tertiary hospital in central Taiwan and linked with the Taiwanese nationwide death registration data. Three ML models, including extreme gradient boosting (XGBoost), random forest (RF) and logistic regression (LR), were used to establish mortality prediction model. Furthermore, we used feature importance, Shapley Additive exPlanations (SHAP) plot, partial dependence plot (PDP), and local interpretable model-agnostic explanations (LIME) to explain the established model. RESULTS: We enrolled 6994 patients and found the accuracy was similar among the three ML models, and the area under the curve value of using XGBoost to predict 30-day, 90-day and 1-year mortality were 0.858, 0.839 and 0.816, respectively. The calibration curve and decision curve analysis further demonstrated accuracy and applicability of models. SHAP summary plot and PDP plot illustrated the discriminative point of APACHE (acute physiology and chronic health exam) II score, haemoglobin and albumin to predict 1-year mortality. The application of LIME and SHAP force plots quantified the probability of 1-year mortality and algorithm of key features at individual patient level. CONCLUSIONS: We used an explainable ML approach, mainly XGBoost, SHAP and LIME plots to establish an explainable 1-year mortality prediction ML model in critically ill ventilated patients.

A Positive Fluid Balance in the First Week Was Associated With Increased Long-Term Mortality in Critically Ill Patients: A Retrospective Cohort Study.

Introduction: Early fluid balance has been found to affect short-term mortality in critically ill patients; however, there is little knowledge regarding the association between early cumulative fluid balance (CFB) and long-term mortality. This study aims to determine the distinct association between CFB day 1-3 (CFB 1-3) and day 4-7 (CFB 4-7) and long-term mortality in critically ill patients. Patients and Methods: This study was conducted at Taichung Veterans General Hospital, a tertiary care referral center in central Taiwan, by linking the hospital critical care data warehouse 2015-2019 and death registry data of the Taiwanese National Health Research Database. The patients followed up until deceased or the end of the study on 31 December 2019. We use the log-rank test to examine the association between CFB 1-3 and CFB 4-7 with long-term mortality and multivariable Cox regression to identify independent predictors during index admission for long-term mortality in critically ill patients. Results: A total of 4,610 patients were evaluated. The mean age was 66.4 ± 16.4 years, where 63.8% were men. In patients without shock, a positive CFB 4-7, but not CFB 1-3, was associated with 1-year mortality, while a positive CFB 1-3 and CFB 4-7 had a consistent and excess hazard of 1-year mortality among critically ill patients with shock. The multivariate Cox proportional hazard regression model identified that CFB 1-3 and CFB 4-7 (with per 1-liter increment, HR: 1.047 and 1.094; 95% CI 1.037-1.058 and 1.080-1.108, respectively) were independently associated with high long-term mortality in critically ill patients after adjustment of relevant covariates, including disease severity and the presence of shock. Conclusions: We found that the fluid balance in the first week, especially on days 4-7, appears to be an early predictor for long-term mortality in critically ill patients. More studies are needed to validate our findings and elucidate underlying mechanisms.

Impact of Early Fluid Balance on Long-Term Mortality in Critically Ill Surgical Patients: A Retrospective Cohort Study in Central Taiwan.

Fluid balance is an essential issue in critical care; however, the impact of early fluid balance on the long-term mortality in critically ill surgical patients remains unknown. This study aimed to address the impact of day 1-3 and day 4-7 fluid balance on the long-term mortality in critically ill surgical patients. We enrolled patients who were admitted to surgical intensive care units (ICUs) during 2015-2019 at a tertiary hospital in central Taiwan and retrieved date-of-death from the Taiwanese nationwide death registration profile. We used a Log-rank test and a multivariable Cox proportional hazards regression model to determine the independent mortality impact of early fluid balance. A total of 6978 patients were included for analyses (mean age: 60.9 ± 15.9 years; 63.9% of them were men). In-hospital mortality, 90-day mortality, 1-year and overall mortality was 10.3%, 15.8%, 23.8% and 31.7%, respectively. In a multivariable Cox proportional hazard regression model adjusted for relevant covariates, we found that positive cumulative day 4-7 fluid balance was independently associated with long-term mortality (aHR 1.083, 95% CI 1.062-1.105), and a similar trend was found on day 1-3 fluid balance, although to a lesser extent (aHR 1.027, 95% CI 1.011-1.043). In conclusion, the fluid balance in the first week of ICU stay, particularly day 4-7 fluid balance, may affect the long-term outcome in critically ill surgical patients.

An Artificial Intelligence Approach to Bloodstream Infections Prediction.

This study aimed to develop an early prediction model for identifying patients with bloodstream infections. The data resource was taken from 2015 to 2019 at Taichung Veterans General Hospital, and a total of 1647 bloodstream infection episodes and 3552 non-bloodstream infection episodes in the intensive care unit (ICU) were included in the model development and evaluation. During the data analysis, 30 clinical variables were selected, including patients' basic characteristics, vital signs, laboratory data, and clinical information. Five machine learning algorithms were applied to examine the prediction model performance. The findings indicated that the area under the receiver operating characteristic curve (AUROC) of the prediction performance of the XGBoost model was 0.825 for the validation dataset and 0.821 for the testing dataset. The random forest model also presented higher values for the AUROC on the validation dataset and testing dataset, which were 0.855 and 0.851, respectively. The tree-based ensemble learning model enabled high detection ability for patients with bloodstream infections in the ICU. Additionally, the analysis of importance of features revealed that alkaline phosphatase (ALKP) and the period of the central venous catheter are the most important predictors for bloodstream infections. We further explored the relationship between features and the risk of bloodstream infection by using the Shapley Additive exPlanations (SHAP) visualized method. The results showed that a higher prothrombin time is more prominent in a bloodstream infection. Additionally, the impact of a lower platelet count and albumin was more prominent in a bloodstream infection. Our results provide additional clinical information for cut-off laboratory values to assist clinical decision-making in bloodstream infection diagnostics.