Personalized prediction of intradialytic hypotension in clinical practice: Development and evaluation of a novel AI dashboard incorporating risk factors from previous and current dialysis sessions.

BACKGROUND: Intradialytic hypotension (IDH) is one of the most common and critical complications of hemodialysis. Despite many proven factors associated with IDH, accurately predicting it before it occurs for individual patients during dialysis sessions remains a challenge. PURPOSE: To establish artificial intelligence (AI) predictive models for IDH, which consider risk factors from previous and ongoing dialysis to optimize model performance. We then implement a novel digital dashboard with the best model for continuous monitoring of patients' status undergoing hemodialysis. The AI dashboard can display the real-time probability of IDH for each patient in the hemodialysis center providing an objective reference for care members for monitoring IDH and treating it in advance. METHODS: Eight machine learning (ML) algorithms, including Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), K Nearest Neighbor (KNN), Light Gradient Boosting Machine (LightGBM), Multilayer Perception (MLP), eXtreme Gradient Boosting (XGBoost), and NaiveBayes, were used to establish the predictive model of IDH to determine if the patient will acquire IDH within 60 min. In addition to real-time features, we incorporated several features sourced from previous dialysis sessions to improve the model's performance. The electronic medical records of patients who had undergone hemodialysis at Chi Mei Medical Center between September 1, 2020 and December 31, 2020 were included in this research. Impact evaluation of AI assistance was conducted by IDH rate. RESULTS: The results showed that the XGBoost model had the best performance (accuracy: 0.858, sensitivity: 0.858, specificity: 0.858, area under the curve: 0.936) and was chosen for AI dashboard implementation. The care members were delighted with the dashboard providing real-time scientific probabilities for IDH risk and historic predictive records in a graphic style. Other valuable functions were appended in the dashboard as well. Impact evaluation indicated a significant decrease in IDH rate after the application of AI assistance. CONCLUSION: This AI dashboard provides high-quality results in IDH risk prediction during hemodialysis. High-risk patients for IDH will be recognized 60 min earlier, promoting individualized preventive interventions as part of the treatment plan. Our approachis believed to promise an excellent way for IDH management.

Longitudinal Risk Analysis of Second Primary Cancer after Curative Treatment in Patients with Rectal Cancer.

Predicting and improving the response of rectal cancer to second primary cancers (SPCs) remains an active and challenging field of clinical research. Identifying predictive risk factors for SPCs will help guide more personalized treatment strategies. In this study, we propose that experience data be used as evidence to support patient-oriented decision-making. The proposed model consists of two main components: a pipeline for extraction and classification and a clinical risk assessment. The study includes 4402 patient datasets, including 395 SPC patients, collected from three cancer registry databases at three medical centers; based on literature reviews and discussion with clinical experts, 10 predictive variables were considered risk factors for SPCs. The proposed extraction and classification pipelines that classified patients according to importance were age at diagnosis, chemotherapy, smoking behavior, combined stage group, and sex, as has been proven in previous studies. The C5 method had the highest predicted AUC (84.88%). In addition, the proposed model was associated with a classification pipeline that showed an acceptable testing accuracy of 80.85%, a recall of 79.97%, a specificity of 88.12%, a precision of 85.79%, and an F1 score of 79.88%. Our results indicate that chemotherapy is the most important prognostic risk factor for SPCs in rectal cancer survivors. Furthermore, our decision tree for clinical risk assessment illuminates the possibility of assessing the effectiveness of a combination of these risk factors. This proposed model may provide an essential evaluation and longitudinal change for personalized treatment of rectal cancer survivors in the future.

Implementation of a machine learning model in acute coronary syndrome and stroke risk assessment for patients with lower urinary tract symptoms.

OBJECTIVE: The global population is aging and the burden of lower urinary tract symptoms (LUTS) is expected to increase. According to the National Health Insurance Research Database, our previous studies have showed LUTS may predispose patients to cardiovascular disease. However, it is difficult to provide a personalized risk assessment in the context of "having acute coronary syndrome (ACS) and stroke." This study aimed to develop an artificial intelligence (AI)-based prediction model for patients with LUTS. MATERIAL AND METHODS: We retrospectively reviewed the electronic medical records of 1799 patients with LUTS at Chi Mei Medical Center between January 1, 2001 and December, 31, 2018. Features with >10 cases and high correlations with outcomes were imported into six machine learning algorithms. The study outcomes included ACS and stroke. Model performances was evaluated using the area under the receiver operating characteristic curve (AUC). The model with the highest AUC was used to implement the clinical risk prediction application. RESULTS: Age, systemic blood pressure, diastolic blood pressure, creatinine, glycated hemoglobin, hypertension, diabetes mellitus and hyperlipidemia were the most relevant features that affect the outcomes. Based on the AUC, our optimal model was built using multilayer perception (AUC = 0.803) to predict ACS and stroke events within 3 years. CONCLUSION: We successfully built an AI-based prediction system that can be used as a prediction model to achieve time-saving, precise, personalized risk evaluation; it can also be used to offer warning, enhance patient adherence, early intervention and better health care outcomes.

Risk Factors and Nomogram Model for Hepatocellular Carcinoma Development in Chronic Hepatitis B Patients with Low-Level Viremia.

Background and aim: Patients with chronic hepatitis B patients (CHB) with low-level viremia (LLV) are not necessarily at low risk of developing hepatocellular carcinoma (HCC). The question of whether CHB patients with LLV require immediate antiviral agent (AVT) or long-term AVT remains controversial. The study aims to investigate the risk of HCC development and the risk factors in CHB patients with LLV and construct a nomogram model predicting the risk of HCC. Methods: We conducted a retrospective cohort study that enrolled 16,895 CHB patients from January 2008 to December 2020. The patients were divided into three groups for comparison: the LLV group, maintained virological response (MVR) group and HBV-DNA>2000 group. The cumulative incidence of progression to HCC was assessed. Cox regression analysis was performed to determine the final risk factors, and a nomogram model was constructed. The 10-fold Cross-Validation method was utilized for internal validation. Results: A total of 408 new cases of HCC occurred during the average follow-up period of 5.78 years. The 3, 5, and 10-year cumulative HCC risks in the LLV group were 3.56%, 4.96%, and 9.51%, respectively. There was a significant difference in the cumulative risk of HCC between the HBV-DNA level > 2000 IU/mL and LLV groups (p = 0.049). Independent risk factors for HCC development in LLV group included male gender, age, presence of cirrhosis, and platelets count. The Area Under the Curve (AUC) values for the 3-year and 5-year prediction from our HCC risk prediction model were 0.75 and 0.76, respectively. Conclusion: Patients with LLV and MVR are still at risk for developing HCC. The nomogram established for CHB patient with LLV, incorporating identified significant risk factors, serves as an effective tool for predicting HCC-free outcomes. This nomogram model provides valuable information for determining appropriate surveillance strategies and prescribing AVT.

Machine learning approaches for practical predicting outpatient near-future AECOPD based on nationwide electronic medical records.

In this research, we aimed to harness machine learning to predict the imminent risk of acute exacerbation in chronic obstructive pulmonary disease (AECOPD) patients. Utilizing retrospective data from electronic medical records of two Taiwanese hospitals, we identified 26 critical features. To predict 3- and 6-month AECOPD occurrences, we deployed five distinct machine learning algorithms alongside ensemble learning. The 3-month risk prediction was best realized by the XGBoost model, achieving an AUC of 0.795, whereas the XGBoost was superior for the 6-month prediction with an AUC of 0.813. We conducted an explainability analysis and found that the episode of AECOPD, mMRC score, CAT score, respiratory rate, and the use of inhaled corticosteroids were the most impactful features. Notably, our approach surpassed predictions that relied solely on CAT or mMRC scores. Accordingly, we designed an interactive prediction system that provides physicians with a practical tool to predict near-term AECOPD risk in outpatients.

The intervention of artificial intelligence to improve the weaning outcomes of patients with mechanical ventilation: Practical applications in the medical intensive care unit and the COVID-19 intensive care unit: A retrospective study.

Patients admitted to intensive care units (ICU) and receiving mechanical ventilation (MV) may experience ventilator-associated adverse events and have prolonged ICU length of stay (LOS). We conducted a survey on adult patients in the medical ICU requiring MV. Utilizing big data and artificial intelligence (AI)/machine learning, we developed a predictive model to determine the optimal timing for weaning success, defined as no reintubation within 48 hours. An interdisciplinary team integrated AI into our MV weaning protocol. The study was divided into 2 parts. The first part compared outcomes before AI (May 1 to Nov 30, 2019) and after AI (May 1 to Nov 30, 2020) implementation in the medical ICU. The second part took place during the COVID-19 pandemic, where patients were divided into control (without AI assistance) and intervention (with AI assistance) groups from Aug 1, 2022, to Apr 30, 2023, and we compared their short-term outcomes. In the first part of the study, the intervention group (with AI, n = 1107) showed a shorter mean MV time (144.3 hours vs 158.7 hours, P = .077), ICU LOS (8.3 days vs 8.8 days, P = .194), and hospital LOS (22.2 days vs 25.7 days, P = .001) compared to the pre-intervention group (without AI, n = 1298). In the second part of the study, the intervention group (with AI, n = 88) exhibited a shorter mean MV time (244.2 hours vs 426.0 hours, P = .011), ICU LOS (11.0 days vs 18.7 days, P = .001), and hospital LOS (23.5 days vs 40.4 days, P < .001) compared to the control group (without AI, n = 43). The integration of AI into the weaning protocol led to improvements in the quality and outcomes of MV patients.

Real-time artificial intelligence predicts adverse outcomes in acute pancreatitis in the emergency department: Comparison with clinical decision rule.

OBJECTIVE: Artificial intelligence (AI) prediction is increasingly used for decision making in health care, but its application for adverse outcomes in emergency department (ED) patients with acute pancreatitis (AP) is not well understood. This study aimed to clarify this aspect. METHODS: Data from 8274 ED patients with AP in three hospitals from 2009 to 2018 were analyzed. Demographic data, comorbidities, laboratory results, and adverse outcomes were included. Six algorithms were evaluated, and the one with the highest area under the curve (AUC) was implemented into the hospital information system (HIS) for real-time prediction. Predictive accuracy was compared between the AI model and Bedside Index for Severity in Acute Pancreatitis (BISAP). RESULTS: The mean ± SD age was 56.1 ± 16.7 years, with 67.7% being male. The AI model was successfully implemented in the HIS, with Light Gradient Boosting Machine (LightGBM) showing the highest AUC for sepsis (AUC 0.961) and intensive care unit (ICU) admission (AUC 0.973), and eXtreme Gradient Boosting (XGBoost) showing the highest AUC for mortality (AUC 0.975). Compared to BISAP, the AI model had superior AUC for sepsis (BISAP 0.785), ICU admission (BISAP 0.778), and mortality (BISAP 0.817). CONCLUSIONS: The first real-time AI prediction model implemented in the HIS for predicting adverse outcomes in ED patients with AP shows favorable initial results. However, further external validation is needed to ensure its reliability and accuracy.

Using artificial intelligence to predict adverse outcomes in emergency department patients with hyperglycemic crises in real time.

BACKGROUND: Hyperglycemic crises are associated with high morbidity and mortality. Previous studies have proposed methods to predict adverse outcomes of patients in hyperglycemic crises; however, artificial intelligence (AI) has never been used to predict adverse outcomes. We implemented an AI model integrated with the hospital information system (HIS) to clarify whether AI could predict adverse outcomes. METHODS: We included 2,666 patients with hyperglycemic crises from emergency departments (ED) between 2009 and 2018. The patients were randomized into a 70%/30% split for AI model training and testing. Twenty-two feature variables from the electronic medical records were collected. The performance of the multilayer perceptron (MLP), logistic regression, random forest, Light Gradient Boosting Machine (LightGBM), support vector machine (SVM), and K-nearest neighbor (KNN) algorithms was compared. We selected the best algorithm to construct an AI model to predict sepsis or septic shock, intensive care unit (ICU) admission, and all-cause mortality within 1 month. The outcomes between the non-AI and AI groups were compared after implementing the HIS and predicting the hyperglycemic crisis death (PHD) score. RESULTS: The MLP had the best performance in predicting the three adverse outcomes, compared with the random forest, logistic regression, SVM, KNN, and LightGBM models. The areas under the curves (AUCs) using the MLP model were 0.852 for sepsis or septic shock, 0.743 for ICU admission, and 0.796 for all-cause mortality. Furthermore, we integrated the AI predictive model with the HIS to assist decision making in real time. No significant differences in ICU admission or all-cause mortality were detected between the non-AI and AI groups. The AI model performed better than the PHD score for predicting all-cause mortality (AUC 0.796 vs. 0.693). CONCLUSIONS: A real-time AI predictive model is a promising method for predicting adverse outcomes in ED patients with hyperglycemic crises. Further studies recruiting more patients are warranted.

Implementing a Novel Machine Learning System for Nutrition Education in Diabetes Mellitus Nutritional Clinic: Predicting 1-Year Blood Glucose Control.

(1) Background: Persistent hyperglycemia in diabetes mellitus (DM) increases the risk of death and causes cardiovascular disease (CVD), resulting in significant social and economic costs. This study used a machine learning (ML) technique to build prediction models with the factors of lifestyle, medication compliance, and self-control in eating habits and then implemented a predictive system based on the best model to forecast whether blood glucose can be well-controlled within 1 year in diabetic patients attending a DM nutritional clinic. (2) Methods: Data were collected from outpatients aged 20 years or older with type 2 DM who received nutrition education in Chi Mei Medical Center. Multiple ML algorithms were used to build the predictive models. (3) Results: The predictive models achieved accuracies ranging from 0.611 to 0.690. The XGBoost model with the highest area under the curve (AUC) of 0.738 was regarded as the best and used for the predictive system implementation. SHAP analysis was performed to interpret the feature importance in the best model. The predictive system, evaluated by dietitians, received positive feedback as a beneficial tool for diabetes nutrition consultations. (4) Conclusions: The ML prediction model provides a promising approach for diabetes nutrition consultations to maintain good long-term blood glucose control, reduce diabetes-related complications, and enhance the quality of medical care.

Implementing AI Models for Prognostic Predictions in High-Risk Burn Patients.

BACKGROUND AND OBJECTIVES: Burn injuries range from minor medical issues to severe, life-threatening conditions. The severity and location of the burn dictate its treatment; while minor burns might be treatable at home, severe burns necessitate medical intervention, sometimes in specialized burn centers with extended follow-up care. This study aims to leverage artificial intelligence (AI)/machine learning (ML) to forecast potential adverse effects in burn patients. METHODS: This retrospective analysis considered burn patients admitted to Chi Mei Medical Center from 2010 to 2019. The study employed 14 features, comprising supplementary information like prior comorbidities and laboratory results, for building models for predicting graft surgery, a prolonged hospital stay, and overall adverse effects. Overall, 70% of the data set trained the AI models, with the remaining 30% reserved for testing. Three ML algorithms of random forest, LightGBM, and logistic regression were employed with evaluation metrics of accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). RESULTS: In this research, out of 224 patients assessed, the random forest model yielded the highest AUC for predictions related to prolonged hospital stays (>14 days) at 81.1%, followed by the XGBoost (79.9%) and LightGBM (79.5%) models. Besides, the random forest model of the need for a skin graft showed the highest AUC (78.8%), while the random forest model and XGBoost model of the occurrence of adverse complications both demonstrated the highest AUC (87.2%) as well. Based on the best models with the highest AUC values, an AI prediction system is designed and integrated into hospital information systems to assist physicians in the decision-making process. CONCLUSIONS: AI techniques showcased exceptional capabilities for predicting a prolonged hospital stay, the need for a skin graft, and the occurrence of overall adverse complications for burn patients. The insights from our study fuel optimism for the inception of a novel predictive model that can seamlessly meld with hospital information systems, enhancing clinical decisions and bolstering physician-patient dialogues.

Machine Learning Algorithm Predicts Mortality Risk in Intensive Care Unit for Patients with Traumatic Brain Injury.

BACKGROUND: Numerous mortality prediction tools are currently available to assist patients with moderate to severe traumatic brain injury (TBI). However, an algorithm that utilizes various machine learning methods and employs diverse combinations of features to identify the most suitable predicting outcomes of brain injury patients in the intensive care unit (ICU) has not yet been well-established. METHOD: Between January 2016 and December 2021, we retrospectively collected data from the electronic medical records of Chi Mei Medical Center, comprising 2260 TBI patients admitted to the ICU. A total of 42 features were incorporated into the analysis using four different machine learning models, which were then segmented into various feature combinations. The predictive performance was assessed using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve and validated using the Delong test. RESULT: The AUC for each model under different feature combinations ranged from 0.877 (logistic regression with 14 features) to 0.921 (random forest with 22 features). The Delong test indicated that the predictive performance of the machine learning models is better than that of traditional tools such as APACHE II and SOFA scores. CONCLUSION: Our machine learning training demonstrated that the predictive accuracy of the LightGBM is better than that of APACHE II and SOFA scores. These features are readily available on the first day of patient admission to the ICU. By integrating this model into the clinical platform, we can offer clinicians an immediate prognosis for the patient, thereby establishing a bridge for educating and communicating with family members.

Development of real-time individualized risk prediction models for contrast associated acute kidney injury and 30-day dialysis after contrast enhanced computed tomography.

PURPOSE: This study aimed to develop preprocedural real-time artificial intelligence (AI)-based systems for predicting individualized risks of contrast-associated acute kidney injury (CA-AKI) and dialysis requirement within 30 days following contrast-enhanced computed tomography (CECT). METHOD: This single-center, retrospective study analyzed adult patients from emergency or in-patient departments who underwent CECT; 18,895 patients were included after excluding those who were already on dialysis, had stage V chronic kidney disease, or had missing data regarding serum creatinine levels within 7 days before and after CECT. Clinical parameters, laboratory data, medication exposure, and comorbid diseases were selected as predictive features. The patients were randomly divided into model training and testing groups at a 7:3 ratio. Logistic regression (LR) and random forest (RF) were employed to create prediction models, which were evaluated using receiver operating characteristic curves. RESULTS: The incidence rates of CA-AKI and dialysis within 30 days post-CECT were 6.69% and 0.98%, respectively. For CA-AKI prediction, LR and RF exhibited similar performance, with areas under curve (AUCs) of 0.769 and 0.757, respectively. For 30-day dialysis prediction, LR (AUC, 0.863) and RF (AUC, 0.872) also exhibited similar performance. Relative to eGFR-alone, the LR and RF models produced significantly higher AUCs for CA-AKI prediction (LR vs. eGFR alone, 0.769 vs. 0.626, p < 0.001) and 30-day dialysis prediction (RF vs. eGFR alone, 0.872 vs. 0.738, p < 0.001). CONCLUSIONS: The proposed AI prediction models significantly outperformed eGFR-alone for predicting the CA-AKI and 30-day dialysis risks of emergency department and hospitalized patients who underwent CECT.

Real-time artificial intelligence system for bacteremia prediction in adult febrile emergency department patients.

BACKGROUND: Artificial intelligence (AI) holds significant potential to be a valuable tool in healthcare. However, its application for predicting bacteremia among adult febrile patients in the emergency department (ED) remains unclear. Therefore, we conducted a study to provide clarity on this issue. METHODS: Adult febrile ED patients with blood cultures at Chi Mei Medical Center were divided into derivation (January 2017 to June 2019) and validation groups (July 2019 to December 2020). The derivation group was utilized to develop AI models using twenty-one feature variables and five algorithms to predict bacteremia. The performance of these models was compared with qSOFA score. The AI model with the highest area under the receiver operating characteristics curve (AUC) was chosen to implement the AI prediction system and tested on the validation group. RESULTS: The study included 5,647 febrile patients. In the derivation group, there were 3,369 patients with a mean age of 61.4 years, and 50.7% were female, including 508 (13.8%) with bacteremia. The model with the best AUC was built using the random forest algorithm (0.761), followed by logistic regression (0.755). All five models demonstrated better AUC than the qSOFA score (0.560). The random forest model was adopted to build a real-time AI prediction system integrated into the hospital information system, and the AUC achieved 0.709 in the validation group. CONCLUSION: The AI model shows promise to predict bacteremia in adult febrile ED patients; however, further external validation in different hospitals and populations is necessary to verify its effectiveness.

Machine Learning Approach for Prediction of the Test Results of Gonadotropin-Releasing Hormone Stimulation: Model Building and Implementation.

Precocious puberty in girls is defined as the onset of pubertal changes before 8 years of age, and gonadotropin-releasing hormone (GnRH) agonist treatment is available for central precocious puberty (CPP). The gold standard for diagnosing CPP is the GnRH stimulation test. However, the GnRH stimulation test is time-consuming, costly, and requires repeated blood sampling. We aimed to develop an artificial intelligence (AI) prediction model to assist pediatric endocrinologists in decision making regarding the optimal timing to perform the GnRH stimulation test. We reviewed the medical charts of 161 girls who received the GnRH stimulation test from 1 August 2010 to 31 August 2021, and we selected 15 clinically relevant features for machine learning modeling. We chose the models with the highest area under the receiver operating characteristic curve (AUC) to integrate into our computerized physician order entry (CPOE) system. The AUC values for the CPP diagnosis prediction model (LH ≥ 5 IU/L) were 0.884 with logistic regression, 0.912 with random forest, 0.942 with LightGBM, and 0.942 with XGBoost. For the Taiwan National Health Insurance treatment coverage prediction model (LH ≥ 10 IU/L), the AUC values were 0.909, 0.941, 0.934, and 0.881, respectively. In conclusion, our AI predictive system can assist pediatric endocrinologists when they are deciding whether a girl with suspected CPP should receive a GnRH stimulation test. With proper use, this prediction model may possibly avoid unnecessary invasive blood sampling for GnRH stimulation tests.

Deep Learning Approach for Negation and Speculation Detection for Automated Important Finding Flagging and Extraction in Radiology Report: Internal Validation and Technique Comparison Study.

BACKGROUND: Negation and speculation unrelated to abnormal findings can lead to false-positive alarms for automatic radiology report highlighting or flagging by laboratory information systems. OBJECTIVE: This internal validation study evaluated the performance of natural language processing methods (NegEx, NegBio, NegBERT, and transformers). METHODS: We annotated all negative and speculative statements unrelated to abnormal findings in reports. In experiment 1, we fine-tuned several transformer models (ALBERT [A Lite Bidirectional Encoder Representations from Transformers], BERT [Bidirectional Encoder Representations from Transformers], DeBERTa [Decoding-Enhanced BERT With Disentangled Attention], DistilBERT [Distilled version of BERT], ELECTRA [Efficiently Learning an Encoder That Classifies Token Replacements Accurately], ERNIE [Enhanced Representation through Knowledge Integration], RoBERTa [Robustly Optimized BERT Pretraining Approach], SpanBERT, and XLNet) and compared their performance using precision, recall, accuracy, and F1-scores. In experiment 2, we compared the best model from experiment 1 with 3 established negation and speculation-detection algorithms (NegEx, NegBio, and NegBERT). RESULTS: Our study collected 6000 radiology reports from 3 branches of the Chi Mei Hospital, covering multiple imaging modalities and body parts. A total of 15.01% (105,755/704,512) of words and 39.45% (4529/11,480) of important diagnostic keywords occurred in negative or speculative statements unrelated to abnormal findings. In experiment 1, all models achieved an accuracy of >0.98 and F1-score of >0.90 on the test data set. ALBERT exhibited the best performance (accuracy=0.991; F1-score=0.958). In experiment 2, ALBERT outperformed the optimized NegEx, NegBio, and NegBERT methods in terms of overall performance (accuracy=0.996; F1-score=0.991), in the prediction of whether diagnostic keywords occur in speculative statements unrelated to abnormal findings, and in the improvement of the performance of keyword extraction (accuracy=0.996; F1-score=0.997). CONCLUSIONS: The ALBERT deep learning method showed the best performance. Our results represent a significant advancement in the clinical applications of computer-aided notification systems.

Using an Artificial Intelligence Approach to Predict the Adverse Effects and Prognosis of Tuberculosis.

BACKGROUND: Tuberculosis (TB) is one of the leading causes of death worldwide and a major cause of ill health. Without treatment, the mortality rate of TB is approximately 50%; with treatment, most patients with TB can be cured. However, anti-TB drug treatments may result in many adverse effects. Therefore, it is important to detect and predict these adverse effects early. Our study aimed to build models using an artificial intelligence/machine learning approach to predict acute hepatitis, acute respiratory failure, and mortality after TB treatment. MATERIALS AND METHODS: Adult patients (age ≥ 20 years) who had a TB diagnosis and received treatment from January 2004 to December 2021 were enrolled in the present study. Thirty-six feature variables were used to develop the predictive models with AI. The data were randomly stratified into a training dataset for model building (70%) and a testing dataset for model validation (30%). These algorithms included XGBoost, random forest, MLP, light GBM, logistic regression, and SVM. RESULTS: A total of 2248 TB patients in Chi Mei Medical Center were included in the study; 71.7% were males, and the other 28.3% were females. The mean age was 67.7 ± 16.4 years. The results showed that our models using the six AI algorithms all had a high area under the receiver operating characteristic curve (AUC) in predicting acute hepatitis, respiratory failure, and mortality, and the AUCs ranged from 0.920 to 0.766, 0.884 to 0.797, and 0.834 to 0.737, respectively. CONCLUSIONS: Our AI models were good predictors and can provide clinicians with a valuable tool to detect the adverse prognosis in TB patients early.

External validation of geriatric influenza death score: A multicenter study.

The Geriatric Influenza Death (GID) score was developed to help decision making in older patients with influenza in the emergency department (ED), but external validation is unavailable. Thus, we conducted a study was to fill the data gap. We recruited all older patients (≥65 years) who visited the ED of three hospitals between 2009 and 2018. Demographic data and clinical characteristics were retrospectively collected. Discrimination, goodness of fit, and performance of the GID score were evaluated. Of the 5,508 patients (121 died) with influenza, the mean age was 76.6±7.4 (standard deviation) years, and 49.3% were males. The GID score was higher in the mortality group (1.7±1.1 vs. 0.8±0.8, p <0.01). With 0 as the reference, the odds ratio for morality with score of 1, 2 and ≥3 was 3.08 (95% confidence interval [CI]: 1.66-5.71), 6.69 (95% CI: 3.52-12.71), and 23.68 (95% CI: 11.95-46.93), respectively. The area under the curve was 0.722 (95% CI: 0.677-0.766), and the Hosmer-Lemeshow goodness of fit test was 1.000. The GID score had excellent negative predictive values with different cut-offs. The GID score had good external validity, and further studies are warranted for wider application.

Implementation of a patient-centered mobile shared decision making platform and healthcare workers' evaluation: a case in a medical center.

Shared decision making is a patient-centered clinical decision-making process that allows healthcare workers to share the existing empirical medical outcomes with patients before making critical decisions. This study aims to explore a project in a medical center of developing a mobile SDM in Taiwan. Chi Mei Medical Center developed the mobile SDM platform and conducted a survey of evaluation from healthcare workers. A three-tier platform that based on cloud infrastructure with seven functionalities was developed. The survey revealed that healthcare workers with sufficient SDM knowledge have an antecedent effect on the three perceptive factors of acceptance of mobile SDM. Resistance to change and perceived ease of use show significant effect on behavioral intention. We provided a comprehensive architecture of mobile SDM and observed the implementation in a medical center. The majority of healthcare workers expressed their acceptancem; however, resistance to change still present. It is, therefore, necessary to be eliminated by continuously promoting activities that highlight the advantages of the Mobile SDM platform. In clinical practice, we validated that the mobile SDM provides patients and their families with an easy way to express their concerns to healthcare workers improving significantly their relationship with each other.

An artificial intelligence system to predict the optimal timing for mechanical ventilation weaning for intensive care unit patients: A two-stage prediction approach.

Background: For the intensivists, accurate assessment of the ideal timing for successful weaning from the mechanical ventilation (MV) in the intensive care unit (ICU) is very challenging. Purpose: Using artificial intelligence (AI) approach to build two-stage predictive models, namely, the try-weaning stage and weaning MV stage to determine the optimal timing of weaning from MV for ICU intubated patients, and implement into practice for assisting clinical decision making. Methods: AI and machine learning (ML) technologies were used to establish the predictive models in the stages. Each stage comprised 11 prediction time points with 11 prediction models. Twenty-five features were used for the first-stage models while 20 features were used for the second-stage models. The optimal models for each time point were selected for further practical implementation in a digital dashboard style. Seven machine learning algorithms including Logistic Regression (LR), Random Forest (RF), Support Vector Machines (SVM), K Nearest Neighbor (KNN), lightGBM, XGBoost, and Multilayer Perception (MLP) were used. The electronic medical records of the intubated ICU patients of Chi Mei Medical Center (CMMC) from 2016 to 2019 were included for modeling. Models with the highest area under the receiver operating characteristic curve (AUC) were regarded as optimal models and used to develop the prediction system accordingly. Results: A total of 5,873 cases were included in machine learning modeling for Stage 1 with the AUCs of optimal models ranging from 0.843 to 0.953. Further, 4,172 cases were included for Stage 2 with the AUCs of optimal models ranging from 0.889 to 0.944. A prediction system (dashboard) with the optimal models of the two stages was developed and deployed in the ICU setting. Respiratory care members expressed high recognition of the AI dashboard assisting ventilator weaning decisions. Also, the impact analysis of with- and without-AI assistance revealed that our AI models could shorten the patients' intubation time by 21 hours, besides gaining the benefit of substantial consistency between these two decision-making strategies. Conclusion: We noticed that the two-stage AI prediction models could effectively and precisely predict the optimal timing to wean intubated patients in the ICU from ventilator use. This could reduce patient discomfort, improve medical quality, and lower medical costs. This AI-assisted prediction system is beneficial for clinicians to cope with a high demand for ventilators during the COVID-19 pandemic.

Does AI explainability affect physicians' intention to use AI?

BACKGROUND: Artificial Intelligence (AI) is increasingly being developed to support clinical decisions for better health service quality, but the adoption of AI in hospitals is not as popular as expected. A possible reason is that the unclear AI explainability (XAI) affects the physicians' consideration of adopting the model. PURPOSE: To propose and validate an innovative conceptual model aimed at exploring physicians' intention to use AI with XAI as an antecedent variable of technology trust (TT) and perceived value (PV). METHODS: A questionnaire survey was conducted to collect data from physicians of three hospitals in Taiwan. Structural equation modeling (SEM) was used to validate the proposed model and test the hypotheses. RESULTS: A total of 295 valid questionnaires were collected. The research results showed that physicians expressed a high intention to use AI. The XAI was found to be of great importance and had a significant impact both on AI TT and PV. We also observed that TT in AI had a significant impact on PV. Moreover, physicians' PV and TT in AI had a significant impact on their behavioral intention to use AI (BI). However, XAI's impact on BI cannot be proved. CONCLUSIONS: The conceptual model developed in this study provides empirical evidence that could be used as guidelines to effectively explore physicians' intention to use medical AI from the antecedent of XAI. Our findings contribute crucial AI-human interaction insights in health care studies.