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.

Early absolute lymphocyte count was associated with one-year mortality in critically ill surgical patients: A propensity score-matching and weighting study.

BACKGROUND: Absolute lymphocyte count (ALC) is a crucial indicator of immunity in critical illness, but studies focusing on long-term outcomes in critically ill patients, particularly surgical patients, are still lacking. We sought to explore the association between week-one ALC and long-term mortality in critically ill surgical patients. METHODS: We used the 2015-2020 critical care database of Taichung Veterans General Hospital (TCVGH), a referral hospital in central Taiwan, and the primary outcome was one-year all-cause mortality. We assessed the association between ALC and long-term mortality by measuring hazard ratios (HRs) with 95% confidence intervals (CIs). Furthermore, we used propensity score-matching and -weighting analyses, consisting of propensity score matching (PSM), inverse probability of treatment weighting (IPTW), and covariate balancing propensity score (CBPS), to validate the association. RESULTS: A total of 8052 patients were enrolled, with their one-year mortality being 24.2%. Cox regression showed that low ALC was independently associated with mortality (adjHR 1.140, 95% CI 1.091-1.192). Moreover, this association tended to be stronger among younger patients, patients with fewer comorbidities and lower severity. The association between low ALC and mortality in original, PSM, IPTW, and CBPS populations were 1.497 (95% CI 1.320-1.697), 1.391 (95% CI 1.169-1.654), 1.512 (95% CI 1.310-1.744), and 1.511 (95% CI 1.310-1.744), respectively. Additionally, the association appears to be consistent, using distinct cutoff levels to define the low ALC. CONCLUSIONS: We identified that early low ALC was associated with increased one-year mortality in critically ill surgical patients, and prospective studies are warranted to confirm the finding.

Deep-Learning-Based Automated Anomaly Detection of EEGs in Intensive Care Units.

An intensive care unit (ICU) is a special ward in the hospital for patients who require intensive care. It is equipped with many instruments monitoring patients' vital signs and supported by the medical staff. However, continuous monitoring demands a massive workload of medical care. To ease the burden, we aim to develop an automatic detection model to monitor when brain anomalies occur. In this study, we focus on electroencephalography (EEG), which monitors the brain electroactivity of patients continuously. It is mainly for the diagnosis of brain malfunction. We propose the gated-recurrent-unit-based (GRU-based) model for detecting brain anomalies; it predicts whether the spike or sharp wave happens within a short time window. Based on the banana montage setting, the proposed model exploits characteristics of multiple channels simultaneously to detect anomalies. It is trained, validated, and tested on separated EEG data and achieves more than 90% testing performance on sensitivity, specificity, and balanced accuracy. The proposed anomaly detection model detects the existence of a spike or sharp wave precisely; it will notify the ICU medical staff, who can provide immediate follow-up treatment. Consequently, it can reduce the medical workload in the ICU significantly.

An automated ICU agitation monitoring system for video streaming using deep learning classification.

OBJECTIVE: To address the challenge of assessing sedation status in critically ill patients in the intensive care unit (ICU), we aimed to develop a non-contact automatic classifier of agitation using artificial intelligence and deep learning. METHODS: We collected the video recordings of ICU patients and cut them into 30-second (30-s) and 2-second (2-s) segments. All of the segments were annotated with the status of agitation as "Attention" and "Non-attention". After transforming the video segments into movement quantification, we constructed the models of agitation classifiers with Threshold, Random Forest, and LSTM and evaluated their performances. RESULTS: The video recording segmentation yielded 427 30-s and 6405 2-s segments from 61 patients for model construction. The LSTM model achieved remarkable accuracy (ACC 0.92, AUC 0.91), outperforming other methods. CONCLUSION: Our study proposes an advanced monitoring system combining LSTM and image processing to ensure mild patient sedation in ICU care. LSTM proves to be the optimal choice for accurate monitoring. Future efforts should prioritize expanding data collection and enhancing system integration for practical application.

Lower body mass index is an independent predictor of mortality in older patients with acute respiratory distress syndrome.

Background: Acute respiratory distress syndrome (ARDS) is associated with high mortality. The impacts of body mass index (BMI) on the morality of older patients with ARDS remain unclear. Methods: This is a single-center cohort study which was conducted at Taichung Veterans General Hospital, Taiwan. Adult patients admitted to the ICU needing mechanical ventilation with ARDS were included for analysis. We compared the data of older patients (age ≥65 years) with those of younger patients (Age <65 years). The factors associated with in-hospital mortality of older patients were investigated. Results: This study included a total of 728 (mean age: 66 years; men: 63%) patients, and 425 (58.4%) of them aged ≥65 years. Older patients exhibited lower body mass index (BMI) (23.8 vs 25.2), higher Acute Physiology and Chronic Health Evaluation (APACHE) II scores (28.9 vs 26.3), higher Charlson Comorbidity Index (CCI) (4.0 vs 3.4), and lower Sequential Organ Failure Assessment (SOFA) scores (10.0 vs 11.1) than younger patients. Furthermore, older patients had mortality rates similar to younger patients (40.5% vs 42.9%, P = 0.542), but had longer length of stay in the ICU (17.6 vs 15.6 days, P = 0.047). For older patients, BMI <18.5 (odds ratio [OR], 2.78; 95% confidence interval [CI], 1.45-5.34), high SOFA score (OR, 1.20; 95% CI, 1.12-1.28), and moderate (OR, 1.95; 95% CI 1.20-3.14) or severe ARDS (OR, 2.30; 95% CI 1.26-4.22) were independent risk factors for mortality. Conclusions: In this cohort, critical ill older patients with ARDS had lower BMI, more comorbidities, and higher APACHE II scores than younger patients. Mortality rate was similar between older and younger patients. Low BMI, high SOFA score, and moderate or severe ARDS were independently associated with mortality in older patients with ARDS.

Factors associated with mental illness in patients with rheumatoid arthritis initiating b/ts DMARDs: A population-based study.

AIM: Mental health is an essential issue in patients with rheumatoid arthritis (RA) but remains unclear among those receiving biological and targeted synthetic disease-modifying antirheumatic drugs (b/tsDMARDs). We aim to assess the incidence and factors associated with mental illness among patients with RA who underwent b/tsDMARD therapy. METHOD: We used Taiwan's National Health Insurance Research Database for the period 2001-2020 to identify patients with RA receiving b/tsDMARDs. The primary outcome was newly developed mental illness, including anxiety and mood disorders. We performed a Cox regression analysis to determine factors associated with mental illness and presented as hazard ratios (HR) with 95% confidence interval (CI). RESULTS: We enrolled 10 852 patients, with 7854 patients receiving tumor necrosis factors inhibitors (TNFi), 1693 patients receiving non-TNFi bDMARDs, and 1305 patients treated with tsDMARD. We found that 13.62% of enrolled patients developed mental illness, with an incidence rate of 4054 per 100 000 person-year. Those receiving tocilizumab (aHR 0.64, 95% CI: 0.51-0.82), abatacept (aHR 0.69, 95% CI: 0.55-0.86), or tsDMARDs (aHR 0.58, 95% CI: 0.47-0.73) had a lower risk of mental illness compared with those receiving TNFi. We also found that old age, low income, diabetes mellitus, use of cyclosporine, and use of steroids were associated with incident mental illness. CONCLUSION: This population-based study investigated the incidence and factors associated with mental illness among patients with RA receiving b/tsDMARDs. Our findings highlight the need for vigilance with respect to the possibility of mental illness in patients with RA.

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.

Medical malpractice in hospitals-how healthcare staff feel.

Introduction: Literature is limited on quantified acute stress reaction, the impact of event scale on medical staff when facing medical malpractice (MMP), and how to individually care for staff. Methods: We analyzed data in the Taichung Veterans General Hospital from October 2015 to December 2017, using the Stanford Acute Stress Reaction Questionnaire (SASRQ), the Impact of Event Scale-Revised (IES-R), and the medical malpractice stress syndrome (MMSS). Results and Discussion: Of all 98 participants, most (78.8%) were women. Most MMPs (74.5%) did not involve injury to patients, and most staff (85.7%) indicated receiving help from the hospital. The internal-consistency evaluations of the three questionnaires showed good validity and reliability. The highest score of IES-R was the construct of intrusion (30.1); the most severe construct of SASRQ was "Marked symptoms of anxiety or increased arousal," and the most were having mental and mild physical symptoms for MMES. A higher total IES-R was associated with younger age (<40 y/o), and more severe injury on patients (mortality). Those who indicated receiving very much help from the hospital were those having significantly lower SASRQ sores. Our study highlighted that hospital authorities should regularly follow up on staff's response to MMP. With timely interventions, vicious cycles of bad feelings can be avoided, especially in young, non-doctor, and non-administrative staff.

Three-year hospital-wide pain management system implementation at a tertiary medical center: Pain prevalence analysis.

We developed a pain management system over a 3-year period. In this project, "Towards a pain-free hospital", we combined evidence-based medicine and medical expertise to develop a series of policies. The intervention mainly included the development of standard procedures for inpatient pain management, the implementation of hospital-wide pain medicine education and training, the establishment of a dashboard system to track pain status, and regular audits and feedback. This study aimed to gain an understanding of the changes in the prevalence of pain in inpatients under the care of the pain management system. The subjects of the survey are inpatients over 20 years old, and who had been hospitalized in the general ward for at least 3 days. The patients would be excluded if they were unable to respond to the questions. We randomly selected eligible patients in the general ward. Our trained interviewers visited inpatients to complete the questionnaires designed by our pain care specialists. A total of 3,094 inpatients completed the survey from 2018 to 2020. During the three-year period, the prevalence of pain was 69.5% (2018) (reference), 63.3% (2019) (OR:0.768, p<0.01), and 60.1% (2020) (OR:0.662, p <0.001). The prevalence rates of pain in patients undergoing surgery during the 3-year period were 81.4% (2018), 74.3% (2019), and 68.8% (2020), respectively. As for care-related causes of pain, injection, change in position/chest percussion, and rehabilitation showed a decreasing trend over the 3-year period of study. Our pain management system provided immediate professional pain management, and achieved a good result in the management of acute moderate to severe pain, especially perioperative pain. Studies on pain prevalence and Pain-Free Hospitals are scarce in Asia. With the aid of the policies based on evidence-based medicine and the dashboard information system, from 2018 to 2020, the prevalence of pain has decreased year by year.

Anaemia in the first week may be associated with long-term mortality among critically ill patients: propensity score-based analyses.

BACKGROUND: Anaemia is highly prevalent in critically ill patients; however, the long-term effect on mortality remains unclear. METHODS: We retrospectively included patients admitted to the medical intensive care units (ICUs) during 2015-2020 at the Taichung Veterans General Hospital. The primary outcome of interest was one-year mortality, and hazard ratios (HRs) with 95% confidence intervals (CIs) were determined to assess the association. We used propensity score matching (PSM) and propensity score matching methods, including inverse probability of treatment weighting (IPTW) as well as covariate balancing propensity score (CBPS), in the present study. RESULTS: A total of 7,089 patients were eligible for analyses, and 45.0% (3,189/7,089) of them had anaemia, defined by mean levels of haemoglobin being less than 10 g/dL. The standardised difference of covariates in this study were lower than 0.20 after matching and weighting. The application of CBPS further reduced the imbalance among covariates. We demonstrated a similar association, and adjusted HRs in original, PSM, IPTW and CBPS populations were 1.345 (95% CI 1.227-1.474), 1.265 (95% CI 1.145-1.397), 1.276 (95% CI 1.142-1.427) and 1.260 (95% CI 1.125-1.411), respectively. CONCLUSIONS: We used propensity score-based analyses to identify that anaemia within the first week was associated with increased one-year mortality in critically ill patients.

Differences in Pain Assessments Between Inpatients and Nurses Leads to Considerable Misestimated Pain.

BACKGROUND: Pain assessments are an important aspect of health care quality because the high prevalence of pain in inpatients may contribute to complications. Several studies revealed a gap in the pain intensity evaluated by nurses (PEN) and patients (PEP). The aim of the present study was to analyze the correlation and agreement between pain assessments conducted by nurses and patients, and to determine patients at high risk of misestimated pain. OBJECTIVES: To compare the difference of pain intensity between the questionnaires conducted by additional assessors and electronic records by nursing staff. STUDY DESIGN: A retrospective study. SETTING: A medical center in Taichung, Taiwan. METHODS: We approached 1,034 patients admitted from January 1, 2018 to December 31, 2018 in our hospital. We compared the assessments of pain intensity using questionnaires conducted by additional assessors with those entered into electronic records by nursing staff. Continuous data were reported as the mean (± standard deviation). The analysis of agreement and correlation were performed by kappa statistics or weighted kappa statistics, and correlation (Spearman rank correlation method). RESULTS: Among the 1,034 patients, 307 patients were excluded. Thus, the final analysis included 686 patients. Patients' median pain intensity was 5 in PEP and 1 in PEN. The patients' pain intensity was underestimated (PEN < PEP) in 539 patients (78.6%), matched (PEN = PEP) in 126 patients (18.3%), and overestimated (PEN > PEP) in 21 patients (3.1%). The surgical interventions (chi squared = 7.996, and P = 0.018) and pain in the past 24 hours (chi squared = 17.776, and P < 0.001) led to a significant difference. LIMITATIONS: The limitation of the study was the single-center and retrospective design. CONCLUSIONS: The gap in pain assessments between inpatients and nurses is an important issue in daily practice. The underestimations of pain were more common than overestimations (78.6% vs 3.1%). Surgical interventions and persistent pain lasting over 24 hours were high risk factors for underestimation, but patients' gender, receiving anesthesia, type of anesthesia, and patient-controlled analgesia did not contribute significantly to differences in pain estimation.

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.

Perception Disparity of Telemedicine Use between Outpatients and Medical Staff during the COVID-19 Pandemic.

We assessed the characteristics and perception of telephone appointments among outpatients and medical staff during the COVID-19 pandemic in Taiwan. Our survey was performed by giving self-administered questionnaires to the enrollees. Basic socioeconomic status data were collected. We used a valid and reliable telehealth usability questionnaire (TUQ) to assess the telemedicine experience among outpatients and medical staff. Only outpatients with chronic illness and who had regular visits before the pandemic were enrolled. We delivered the questionnaire survey to participants who used telephone appointments from 20 May 2021 to 31 July 2021 in Taichung Veterans General Hospital. A total of 471 outpatients and 203 medical staff completed the survey. Most of the respondents were aged 30-69, college-educated, women, and married. Outpatients have higher scores in all dimensions of TUQ than medical staff, especially in the dimensions of ease of use and effectiveness. Age, gender, education, and marriage have no significant associations in the medical staff group. In the outpatient group, gender is the only significant factor in the six dimensions of TUQ. We found a significant disparity in the perception gap of telemedicine among outpatient and medical staff. Outpatients are satisfied with telephone appointments during the COVID-19 pandemic, but medical staff are concerned about the ease of use and effectiveness.

Week-One Anaemia was Associated with Increased One-Year Mortality in Critically Ill Surgical Patients.

Background: Anaemia has a deleterious effect on surgical patients, but the long-term impact of anaemia in critically ill surgical patients remains unclear. Methods: We enrolled consecutive patients who were admitted to surgical intensive care units (ICUs) at a tertiary referral centre in central Taiwan between 2015 and 2020. We used both Cox proportional hazards analysis and propensity score-based analyses, including propensity score matching (PSM), inverse probability of treatment weighting (IPTW), and covariate balancing propensity score (CBPS) to determine hazard ratios (HRs) and 95% confidence intervals (CIs) for one-year mortality. Results: A total of 7,623 critically ill surgical patients were enrolled, and 29.9% (2,280/7,623) of them had week-one anaemia (haemoglobin <10 g/dL). We found that anaemia was independently associated with an increased risk of one-year mortality after adjustment for relevant covariates (aHR, 1.170; 95% CI, 1.045-1.310). We further identified a consistent strength of association between anaemia and one-year mortality in propensity score-based analyses, with the adjusted HRs in the PSM, IPTW, and CBPS were 1.164 (95% CI 1.025-1.322), 1.179 (95% CI 1.030-1.348), and 1.181 (1.034-1.349), respectively. Conclusions: We identified the impact on one-year mortality of anaemia in critically ill surgical patients, and more studies are needed to validate our findings.

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.

Deep Learning-Based Pain Classifier Based on the Facial Expression in Critically Ill Patients.

Objective: Pain assessment based on facial expressions is an essential issue in critically ill patients, but an automated assessment tool is still lacking. We conducted this prospective study to establish the deep learning-based pain classifier based on facial expressions. Methods: We enrolled critically ill patients during 2020-2021 at a tertiary hospital in central Taiwan and recorded video clips with labeled pain scores based on facial expressions, such as relaxed (0), tense (1), and grimacing (2). We established both image- and video-based pain classifiers through using convolutional neural network (CNN) models, such as Resnet34, VGG16, and InceptionV1 and bidirectional long short-term memory networks (BiLSTM). The performance of classifiers in the test dataset was determined by accuracy, sensitivity, and F1-score. Results: A total of 63 participants with 746 video clips were eligible for analysis. The accuracy of using Resnet34 in the polychromous image-based classifier for pain scores 0, 1, 2 was merely 0.5589, and the accuracy of dichotomous pain classifiers between 0 vs. 1/2 and 0 vs. 2 were 0.7668 and 0.8593, respectively. Similar accuracy of image-based pain classifier was found using VGG16 and InceptionV1. The accuracy of the video-based pain classifier to classify 0 vs. 1/2 and 0 vs. 2 was approximately 0.81 and 0.88, respectively. We further tested the performance of established classifiers without reference, mimicking clinical scenarios with a new patient, and found the performance remained high. Conclusions: The present study demonstrates the practical application of deep learning-based automated pain assessment in critically ill patients, and more studies are warranted to validate our findings.

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.