Wenli Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Wenli Zhang has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Qing Li
Department of Industrial & Manufacturing Systems Engineering, Iowa State University, Ames, IA, United States
11 shared publications - Yiqun Jiang
Affiliation Department of Industrial and Manufacturing Systems Engineering, Iowa State University, Ames, Iowa, United States of America
9 shared publications - Jiaheng Xie
Department of Accounting & MIS, Lerner College of Business and Economics, University of Delaware, Newark, Delaware 19716
4 shared publications - Shaodong Wang
Department of Industrial and Manufacturing Systems Engineering, Iowa State University
4 shared publications - Yu-Li Huang
Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, USA
3 shared publications - Ben Niu
Shenzhen University
2 shared publications
Latest Publications
- Open Access
- 31 July 2026
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 14, e88854
A Dynamic Graph–Based Multiobjective Optimization Method for Physician Recommendation: Development and Evaluation Study
Background: Online health care consultation provides patients with broad access to physicians, but also presents the challenge of selecting a suitable doctor in the absence of triage guidance. Patients have multifaceted needs, prioritizing not only recommendation accuracy but also physician service quality, and diversity of physician expertise. Furthermore, the sparsity of patient interaction data intensifies the difficulty of providing balanced and effective recommendations. Objective: This study aims to develop a multiobjective physician recommendation method that simultaneously optimizes recommendation accuracy, service quality, and diversity of physician expertise while addressing the data sparsity challenge inherent in online health care platforms. Methods: We propose dynamic graph–based bacteria colony optimization for multiobjective physician recommendation (DyGMO-PR), a dynamic graph–based multiobjective optimization method. It integrates bacterial colony optimization with an evolving physician relationship graph. Our approach features a novel graph–based encoding scheme, a chemotaxis-inspired graph-walking strategy for stable search, and a dynamic graph evolution mechanism that learns implicit physician relationships to enhance recommendation quality under sparse data conditions. Results: Evaluation on a real-world dataset comprising 10,493 consultation records from 1256 patients and 1377 physicians demonstrates the effectiveness of DyGMO-PR. Our method consistently outperforms 6 state-of-the-art multiobjective recommendation algorithms. Using a recommendation list length of 6 as an example, DyGMO-PR achieves an accuracy of 0.876, a service quality of 0.873, and a diversity of 0.720, surpassing the best-performing baseline by 7.4%, 3.1%, and 8.1%, respectively. DyGMO-PR also attains the highest hypervolume value (0.696 at K =6) and the highest recall (0.380 at K =6). Case studies and graph analysis further demonstrate the method’s effectiveness in generating interpretable and balanced recommendation lists. Conclusions: DyGMO-PR offers an effective solution for multiobjective physician recommendation. It provides a flexible and practical foundation for building more responsive and reliable recommender systems in online health care.
- 31 July 2026
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 14, e88854
Open AccessA Dynamic Graph–Based Multiobjective Optimization Method for Physician Recommendation: Development and Evaluation Study
Background: Online health care consultation provides patients with broad access to physicians, but also presents the challenge of selecting a suitable doctor in the absence of triage guidance. Patients have multifaceted needs, prioritizing not only recommendation accuracy but also physician service quality, and diversity of physician expertise. Furthermore, the sparsity of patient interaction data intensifies the difficulty of providing balanced and effective recommendations. Objective: This study aims to develop a multiobjective physician recommendation method that simultaneously optimizes recommendation accuracy, service quality, and diversity of physician expertise while addressing the data sparsity challenge inherent in online health care platforms. Methods: We propose dynamic graph–based bacteria colony optimization for multiobjective physician recommendation (DyGMO-PR), a dynamic graph–based multiobjective optimization method. It integrates bacterial colony optimization with an evolving physician relationship graph. Our approach features a novel graph–based encoding scheme, a chemotaxis-inspired graph-walking strategy for stable search, and a dynamic graph evolution mechanism that learns implicit physician relationships to enhance recommendation quality under sparse data conditions. Results: Evaluation on a real-world dataset comprising 10,493 consultation records from 1256 patients and 1377 physicians demonstrates the effectiveness of DyGMO-PR. Our method consistently outperforms 6 state-of-the-art multiobjective recommendation algorithms. Using a recommendation list length of 6 as an example, DyGMO-PR achieves an accuracy of 0.876, a service quality of 0.873, and a diversity of 0.720, surpassing the best-performing baseline by 7.4%, 3.1%, and 8.1%, respectively. DyGMO-PR also attains the highest hypervolume value (0.696 at K =6) and the highest recall (0.380 at K =6). Case studies and graph analysis further demonstrate the method’s effectiveness in generating interpretable and balanced recommendation lists. Conclusions: DyGMO-PR offers an effective solution for multiobjective physician recommendation. It provides a flexible and practical foundation for building more responsive and reliable recommender systems in online health care.
- Open Access
- 16 April 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Detecting Mental Health Signals in Social Media Text: A Large Language Model–Based Approach with Knowledge-Informed Prompt Engineering (Preprint)
BACKGROUND Mental disorders are often under detected, and social media text has emerged as a complementary source of behavioral signals for digital mental health research. Existing approaches for detecting mental health conditions from user-generated content commonly rely on disorder-specific supervised machine learning models or large language model–based discrete prompting strategies, which may require substantial labeled data and may not systematically incorporate domain knowledge. OBJECTIVE This study aimed to develop and evaluate a large language model–based method for detecting mental health signals from user-generated social media text by combining personalized continuous prompting with structured medical knowledge. METHODS We proposed a continuous multi-prompt framework built on FLAN-T5 that integrates 2 components: (1) prefix tuning to capture user-level personalized context and (2) a rule-based prompt layer that injects ontology-based medical knowledge related to symptoms, life events, and treatments. The method is evaluated on eRisk datasets for depression and pathological gambling using subject-level prediction from longitudinal social media posts. The depression dataset included 1707 subjects (214 positive, 1493 negative), and the pathological gambling dataset included 4427 subjects (245 positive, 4182 negative). Data are split into training, validation, and test sets (60%/20%/20%), and results are averaged over 10 runs. We report F1-score, precision, and recall and compare the proposed method with traditional machine learning, deep learning, discrete prompting, and large language model fine-tuning baselines. We also examined few-shot learning, early identification, performance on a rare-disorder dataset (self-harm; 340 subjects, including 41 positive cases), and alignment with psychiatrist ratings. RESULTS The proposed method achieved the highest F1-scores among the evaluated approaches on both primary tasks. For depression detection, it achieved an F1-score of 0.913, precision of 0.888, and recall of 0.939. For pathological gambling detection, it achieved an F1-score of 0.985, precision of 0.985, and recall of 0.990. On the rare-disorder self-harm dataset, the method achieved an F1-score of 0.917. In post hoc evaluation, model probabilities showed strong alignment with psychiatrist ratings (Pearson r=0.8656), while expert interrater agreement is high (Pearson r=0.9636). CONCLUSIONS A large language model approach that combines personalized continuous prompting with structured medical knowledge can improve detection of mental health–related signals in social media text. These findings support the potential of knowledge-informed prompt engineering for scalable, data-efficient digital mental health research using user-generated content. This approach should be interpreted as a complementary screening and monitoring tool for research settings rather than as a clinical diagnostic system.
- 16 April 2026
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessDetecting Mental Health Signals in Social Media Text: A Large Language Model–Based Approach with Knowledge-Informed Prompt Engineering (Preprint)
BACKGROUND Mental disorders are often under detected, and social media text has emerged as a complementary source of behavioral signals for digital mental health research. Existing approaches for detecting mental health conditions from user-generated content commonly rely on disorder-specific supervised machine learning models or large language model–based discrete prompting strategies, which may require substantial labeled data and may not systematically incorporate domain knowledge. OBJECTIVE This study aimed to develop and evaluate a large language model–based method for detecting mental health signals from user-generated social media text by combining personalized continuous prompting with structured medical knowledge. METHODS We proposed a continuous multi-prompt framework built on FLAN-T5 that integrates 2 components: (1) prefix tuning to capture user-level personalized context and (2) a rule-based prompt layer that injects ontology-based medical knowledge related to symptoms, life events, and treatments. The method is evaluated on eRisk datasets for depression and pathological gambling using subject-level prediction from longitudinal social media posts. The depression dataset included 1707 subjects (214 positive, 1493 negative), and the pathological gambling dataset included 4427 subjects (245 positive, 4182 negative). Data are split into training, validation, and test sets (60%/20%/20%), and results are averaged over 10 runs. We report F1-score, precision, and recall and compare the proposed method with traditional machine learning, deep learning, discrete prompting, and large language model fine-tuning baselines. We also examined few-shot learning, early identification, performance on a rare-disorder dataset (self-harm; 340 subjects, including 41 positive cases), and alignment with psychiatrist ratings. RESULTS The proposed method achieved the highest F1-scores among the evaluated approaches on both primary tasks. For depression detection, it achieved an F1-score of 0.913, precision of 0.888, and recall of 0.939. For pathological gambling detection, it achieved an F1-score of 0.985, precision of 0.985, and recall of 0.990. On the rare-disorder self-harm dataset, the method achieved an F1-score of 0.917. In post hoc evaluation, model probabilities showed strong alignment with psychiatrist ratings (Pearson r=0.8656), while expert interrater agreement is high (Pearson r=0.9636). CONCLUSIONS A large language model approach that combines personalized continuous prompting with structured medical knowledge can improve detection of mental health–related signals in social media text. These findings support the potential of knowledge-informed prompt engineering for scalable, data-efficient digital mental health research using user-generated content. This approach should be interpreted as a complementary screening and monitoring tool for research settings rather than as a clinical diagnostic system.
- Open Access
- 7 January 2026
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (1) , e0338922
Unbiased inference for echocardiogram urgency prediction using double machine learning
The increased utilization of echocardiography in clinical practice has witnessed a substantial rise, underscoring its pivotal role as a diagnostic tool for various cardiovascular conditions. However, due to the relative scarcity of echocardiography tests, challenges persist in efficiently prioritizing patients for echocardiographic assessments. In this study, we develop a model to assess the urgency of appointments by considering both clinical and administrative variables extracted from Electronic Health Record data. We use double machine learning techniques to analyze these variables and improve our predictions of patient urgency. Traditional methods for estimating variable effects have limitations, particularly in our research context, where clinical and administrative variables may influence one another while also directly impacting the outcome (i.e., the urgency of appointments). In this work, we address this issue by developing an urgency stratification model using double machine learning, which disentangles the complex relationships between variables. Our evaluations demonstrate that the proposed model not only outperforms traditional machine learning methods in predicting appointment urgency but also provides robust estimations of variable effects. Specifically, our results underscore the critical roles of administrative variables and cancer-related comorbidity variables in patient prioritization and appointment urgency prediction. By leveraging double machine learning techniques, our method can enhance the efficiency and effectiveness of echocardiography utilization in clinical practice. It provides clinicians with actionable insights for patient prioritization, facilitating the timely identification of urgent cases and the optimal allocation of resources. Our work contributes to the advancement of healthcare practices by leveraging sophisticated analytics to improve patient care delivery and streamline clinical workflows in echocardiography laboratories. A similar research design can also be extended to other advanced yet limited laboratory tests to help prioritize medical resources.
- 7 January 2026
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (1) , e0338922
Open AccessUnbiased inference for echocardiogram urgency prediction using double machine learning
The increased utilization of echocardiography in clinical practice has witnessed a substantial rise, underscoring its pivotal role as a diagnostic tool for various cardiovascular conditions. However, due to the relative scarcity of echocardiography tests, challenges persist in efficiently prioritizing patients for echocardiographic assessments. In this study, we develop a model to assess the urgency of appointments by considering both clinical and administrative variables extracted from Electronic Health Record data. We use double machine learning techniques to analyze these variables and improve our predictions of patient urgency. Traditional methods for estimating variable effects have limitations, particularly in our research context, where clinical and administrative variables may influence one another while also directly impacting the outcome (i.e., the urgency of appointments). In this work, we address this issue by developing an urgency stratification model using double machine learning, which disentangles the complex relationships between variables. Our evaluations demonstrate that the proposed model not only outperforms traditional machine learning methods in predicting appointment urgency but also provides robust estimations of variable effects. Specifically, our results underscore the critical roles of administrative variables and cancer-related comorbidity variables in patient prioritization and appointment urgency prediction. By leveraging double machine learning techniques, our method can enhance the efficiency and effectiveness of echocardiography utilization in clinical practice. It provides clinicians with actionable insights for patient prioritization, facilitating the timely identification of urgent cases and the optimal allocation of resources. Our work contributes to the advancement of healthcare practices by leveraging sophisticated analytics to improve patient care delivery and streamline clinical workflows in echocardiography laboratories. A similar research design can also be extended to other advanced yet limited laboratory tests to help prioritize medical resources.
- Open Access
- 14 November 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Management Information Systems
- Vol. 16 (4) , 1-29
ICU Outcome Predictions Using Real-Time Signals with Wavelet-Transform-based Deep Learning Method
Intensive care units (ICUs) serve patients with life-threatening conditions. However, the availability of ICU resources remains scarce worldwide. Therefore, it is critical to conduct ICU outcome predictions at an early stage and promote efficient use of ICU resources. However, the current ICU outcome prediction methods have limitations such as unsatisfactory accuracy and dependence on resource-demanding laboratory tests or expert domain knowledge. In this research, we design a wavelet transform based additive convolutional neural network (WT-A-CNN) that requires only patients’ vital sign series and information at ICU admission for real-time ICU outcome predictions. The model is evaluated using a large real-world ICU database and outperforms state-of-the-art baselines on both ICU mortality and length-of-stay prediction tasks. Furthermore, the additive structure of the model can be used for model interpretation and analyzing which input signal is more useful for ICU outcome prediction for different patient cohorts. Our work provides an efficient tool for ICU outcome predictions, allowing healthcare providers to act promptly on patients at risk and reduce the negative impacts on patient outcomes.
- 14 November 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Management Information Systems
- Vol. 16 (4) , 1-29
Open AccessICU Outcome Predictions Using Real-Time Signals with Wavelet-Transform-based Deep Learning Method
Intensive care units (ICUs) serve patients with life-threatening conditions. However, the availability of ICU resources remains scarce worldwide. Therefore, it is critical to conduct ICU outcome predictions at an early stage and promote efficient use of ICU resources. However, the current ICU outcome prediction methods have limitations such as unsatisfactory accuracy and dependence on resource-demanding laboratory tests or expert domain knowledge. In this research, we design a wavelet transform based additive convolutional neural network (WT-A-CNN) that requires only patients’ vital sign series and information at ICU admission for real-time ICU outcome predictions. The model is evaluated using a large real-world ICU database and outperforms state-of-the-art baselines on both ICU mortality and length-of-stay prediction tasks. Furthermore, the additive structure of the model can be used for model interpretation and analyzing which input signal is more useful for ICU outcome prediction for different patient cohorts. Our work provides an efficient tool for ICU outcome predictions, allowing healthcare providers to act promptly on patients at risk and reduce the negative impacts on patient outcomes.
- Open Access
- 26 August 2025
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 4 (1) , e72671
Intensive Care Unit Patient Outcome Prediction Using ν-Support Vector Classification and Stochastic Signal Processing–Based Feature Extraction Techniques: Algorithm Development and Validation Study
Background: Intensive care units (ICUs) treat patients with life-threatening illnesses. Worldwide, intensive care demand is massive. Predicting patient outcomes in ICUs holds significant importance for health care operation management. Nevertheless, it remains a challenging problem that researchers and health care practitioners have yet to overcome. While the newly emerging health digital trace data offer new possibilities, such data contain complex time series and patterns. Although researchers have devised severity score systems, traditional machine learning models with feature engineering, and deep learning models that use raw clinical data to predict ICU outcomes, existing methods have limitations. Objective: This study aimed to develop a novel feature extraction and machine learning framework to repurpose and extract features with strong predictive power from patients’ health digital traces for ICU outcome prediction. Methods: Guided by signal processing techniques and medical domain knowledge, the proposed framework introduces a novel, signal processing–based feature engineering method to extract highly predictive features from ICU digital trace data. We rigorously evaluated this method on a real-world ICU dataset, demonstrating significant improvements over both traditional and deep learning baseline methods. The method was then evaluated using a real-world database to assess prediction accuracy and feature representativeness. Results: The prediction results obtained by the proposed framework significantly outperformed state-of-the-art benchmarks. This demonstrated the framework’s effectiveness in capturing key patterns from complex health digital traces for improving ICU outcome prediction. Conclusions: Our study contributes to health care operation management by leveraging digital traces from health care information systems to address challenges with significant implications for health care.
- 26 August 2025
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 4 (1) , e72671
Open AccessIntensive Care Unit Patient Outcome Prediction Using ν-Support Vector Classification and Stochastic Signal Processing–Based Feature Extraction Techniques: Algorithm Development and Validation Study
Background: Intensive care units (ICUs) treat patients with life-threatening illnesses. Worldwide, intensive care demand is massive. Predicting patient outcomes in ICUs holds significant importance for health care operation management. Nevertheless, it remains a challenging problem that researchers and health care practitioners have yet to overcome. While the newly emerging health digital trace data offer new possibilities, such data contain complex time series and patterns. Although researchers have devised severity score systems, traditional machine learning models with feature engineering, and deep learning models that use raw clinical data to predict ICU outcomes, existing methods have limitations. Objective: This study aimed to develop a novel feature extraction and machine learning framework to repurpose and extract features with strong predictive power from patients’ health digital traces for ICU outcome prediction. Methods: Guided by signal processing techniques and medical domain knowledge, the proposed framework introduces a novel, signal processing–based feature engineering method to extract highly predictive features from ICU digital trace data. We rigorously evaluated this method on a real-world ICU dataset, demonstrating significant improvements over both traditional and deep learning baseline methods. The method was then evaluated using a real-world database to assess prediction accuracy and feature representativeness. Results: The prediction results obtained by the proposed framework significantly outperformed state-of-the-art benchmarks. This demonstrated the framework’s effectiveness in capturing key patterns from complex health digital traces for improving ICU outcome prediction. Conclusions: Our study contributes to health care operation management by leveraging digital traces from health care information systems to address challenges with significant implications for health care.
- Open Access
- 20 August 2025
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 4, e71247
A Real-Time Signal-Based Wavelet Long Short-Term Memory Method for Length-of-Stay Prediction for the Intensive Care Unit: Development and Evaluation Study
Background: Efficient allocation of health care resources is essential for long-term hospital operation. Effective intensive care unit (ICU) management is essential for alleviating the financial strain on health care systems. Accurate prediction of length-of-stay in ICUs is vital for optimizing capacity planning and resource allocation, with the challenge of achieving early, real-time predictions. Objective: This study aimed to develop a predictive model, namely wavelet long short-term memory model (WT-LSTM), for ICU length-of-stay using only real-time vital sign data. The model is designed for urgent care settings where demographic and historical patient data or laboratory results may be unavailable; the model leverages real-time inputs to deliver early and accurate ICU length-of-stay predictions. Methods: The proposed model integrates discrete wavelet transformation and long short-term memory (LSTM) neural networks to filter noise from patients’ vital sign series and improve length-of-stay prediction accuracy. Model performance was evaluated using the electronic ICU database, focusing on 10 common ICU admission diagnoses in the database. Results: The results demonstrate that WT-LSTM consistently outperforms baseline models, including linear regression, LSTM, and bidirectional long short-term memory, in predicting ICU length-of-stay using vital sign data, achieving significant improvements in mean square error. Specifically, the wavelet transformation component of the model enhances the overall performance of WT-LSTM. Removing this component results in an average decrease of 3.3% in mean square error; such a phenomenon is particularly pronounced in specific patient cohorts. The model’s adaptability is highlighted through real-time predictions using only 3-hour, 6-hour, 12-hour, and 24-hour input data. Using only 3 hours of input data, the WT-LSTM model delivers competitive results across the 10 most common ICU admission diagnoses, often outperforming Acute Physiology and Chronic Health Evaluation IV, the leading ICU outcome prediction system currently implemented in clinical practice. WT-LSTM effectively captures patterns from vital signs recorded during the initial hours of a patient’s ICU stay, making it a promising tool for early prediction and resource optimization in the ICU. Conclusions: Our proposed WT-LSTM model, based on real-time vital sign data, offers a promising solution for ICU length-of-stay prediction. Its high accuracy and early prediction capabilities hold significant potential for enhancing clinical practice, optimizing resource allocation, and supporting critical clinical and administrative decisions in ICU management.
- 20 August 2025
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 4, e71247
Open AccessA Real-Time Signal-Based Wavelet Long Short-Term Memory Method for Length-of-Stay Prediction for the Intensive Care Unit: Development and Evaluation Study
Background: Efficient allocation of health care resources is essential for long-term hospital operation. Effective intensive care unit (ICU) management is essential for alleviating the financial strain on health care systems. Accurate prediction of length-of-stay in ICUs is vital for optimizing capacity planning and resource allocation, with the challenge of achieving early, real-time predictions. Objective: This study aimed to develop a predictive model, namely wavelet long short-term memory model (WT-LSTM), for ICU length-of-stay using only real-time vital sign data. The model is designed for urgent care settings where demographic and historical patient data or laboratory results may be unavailable; the model leverages real-time inputs to deliver early and accurate ICU length-of-stay predictions. Methods: The proposed model integrates discrete wavelet transformation and long short-term memory (LSTM) neural networks to filter noise from patients’ vital sign series and improve length-of-stay prediction accuracy. Model performance was evaluated using the electronic ICU database, focusing on 10 common ICU admission diagnoses in the database. Results: The results demonstrate that WT-LSTM consistently outperforms baseline models, including linear regression, LSTM, and bidirectional long short-term memory, in predicting ICU length-of-stay using vital sign data, achieving significant improvements in mean square error. Specifically, the wavelet transformation component of the model enhances the overall performance of WT-LSTM. Removing this component results in an average decrease of 3.3% in mean square error; such a phenomenon is particularly pronounced in specific patient cohorts. The model’s adaptability is highlighted through real-time predictions using only 3-hour, 6-hour, 12-hour, and 24-hour input data. Using only 3 hours of input data, the WT-LSTM model delivers competitive results across the 10 most common ICU admission diagnoses, often outperforming Acute Physiology and Chronic Health Evaluation IV, the leading ICU outcome prediction system currently implemented in clinical practice. WT-LSTM effectively captures patterns from vital signs recorded during the initial hours of a patient’s ICU stay, making it a promising tool for early prediction and resource optimization in the ICU. Conclusions: Our proposed WT-LSTM model, based on real-time vital sign data, offers a promising solution for ICU length-of-stay prediction. Its high accuracy and early prediction capabilities hold significant potential for enhancing clinical practice, optimizing resource allocation, and supporting critical clinical and administrative decisions in ICU management.
- Open Access
- 14 February 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Intensive Care Unit Patient Outcome Prediction Using ν-Support Vector Classification and Stochastic Signal Processing–Based Feature Extraction Techniques: Algorithm Development and Validation Study (Preprint)
BACKGROUND Intensive care units (ICUs) treat patients with life-threatening illnesses. Worldwide, intensive care demand is massive. Predicting patient outcomes in ICUs holds significant importance for health care operation management. Nevertheless, it remains a challenging problem that researchers and health care practitioners have yet to overcome. While the newly emerging health digital trace data offer new possibilities, such data contain complex time series and patterns. Although researchers have devised severity score systems, traditional machine learning models with feature engineering, and deep learning models that use raw clinical data to predict ICU outcomes, existing methods have limitations. OBJECTIVE This study aimed to develop a novel feature extraction and machine learning framework to repurpose and extract features with strong predictive power from patients’ health digital traces for ICU outcome prediction. METHODS Guided by signal processing techniques and medical domain knowledge, the proposed framework introduces a novel, signal processing–based feature engineering method to extract highly predictive features from ICU digital trace data. We rigorously evaluated this method on a real-world ICU dataset, demonstrating significant improvements over both traditional and deep learning baseline methods. The method was then evaluated using a real-world database to assess prediction accuracy and feature representativeness. RESULTS The prediction results obtained by the proposed framework significantly outperformed state-of-the-art benchmarks. This demonstrated the framework’s effectiveness in capturing key patterns from complex health digital traces for improving ICU outcome prediction. CONCLUSIONS Our study contributes to health care operation management by leveraging digital traces from health care information systems to address challenges with significant implications for health care. CLINICALTRIAL
- 14 February 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessIntensive Care Unit Patient Outcome Prediction Using ν-Support Vector Classification and Stochastic Signal Processing–Based Feature Extraction Techniques: Algorithm Development and Validation Study (Preprint)
BACKGROUND Intensive care units (ICUs) treat patients with life-threatening illnesses. Worldwide, intensive care demand is massive. Predicting patient outcomes in ICUs holds significant importance for health care operation management. Nevertheless, it remains a challenging problem that researchers and health care practitioners have yet to overcome. While the newly emerging health digital trace data offer new possibilities, such data contain complex time series and patterns. Although researchers have devised severity score systems, traditional machine learning models with feature engineering, and deep learning models that use raw clinical data to predict ICU outcomes, existing methods have limitations. OBJECTIVE This study aimed to develop a novel feature extraction and machine learning framework to repurpose and extract features with strong predictive power from patients’ health digital traces for ICU outcome prediction. METHODS Guided by signal processing techniques and medical domain knowledge, the proposed framework introduces a novel, signal processing–based feature engineering method to extract highly predictive features from ICU digital trace data. We rigorously evaluated this method on a real-world ICU dataset, demonstrating significant improvements over both traditional and deep learning baseline methods. The method was then evaluated using a real-world database to assess prediction accuracy and feature representativeness. RESULTS The prediction results obtained by the proposed framework significantly outperformed state-of-the-art benchmarks. This demonstrated the framework’s effectiveness in capturing key patterns from complex health digital traces for improving ICU outcome prediction. CONCLUSIONS Our study contributes to health care operation management by leveraging digital traces from health care information systems to address challenges with significant implications for health care. CLINICALTRIAL
- Open Access
- 29 January 2025
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 4 (1) , e64188
Urgency Prediction for Medical Laboratory Tests Through Optimal Sparse Decision Tree: Case Study With Echocardiograms
Background: In the contemporary realm of health care, laboratory tests stand as cornerstone components, driving the advancement of precision medicine. These tests offer intricate insights into a variety of medical conditions, thereby facilitating diagnosis, prognosis, and treatments. However, the accessibility of certain tests is hindered by factors such as high costs, a shortage of specialized personnel, or geographic disparities, posing obstacles to achieving equitable health care. For example, an echocardiogram is a type of laboratory test that is extremely important and not easily accessible. The increasing demand for echocardiograms underscores the imperative for more efficient scheduling protocols. Despite this pressing need, limited research has been conducted in this area. Objective: The study aims to develop an interpretable machine learning model for determining the urgency of patients requiring echocardiograms, thereby aiding in the prioritization of scheduling procedures. Furthermore, this study aims to glean insights into the pivotal attributes influencing the prioritization of echocardiogram appointments, leveraging the high interpretability of the machine learning model. Methods: Empirical and predictive analyses have been conducted to assess the urgency of patients based on a large real-world echocardiogram appointment dataset (ie, 34,293 appointments) sourced from electronic health records encompassing administrative information, referral diagnosis, and underlying patient conditions. We used a state-of-the-art interpretable machine learning algorithm, the optimal sparse decision tree (OSDT), renowned for its high accuracy and interpretability, to investigate the attributes pertinent to echocardiogram appointments. Results: The method demonstrated satisfactory performance (F1-score=36.18% with an improvement of 1.7% and F2-score=28.18% with an improvement of 0.79% by the best-performing baseline model) in comparison to the best-performing baseline model. Moreover, due to its high interpretability, the results provide valuable medical insights regarding the identification of urgent patients for tests through the extraction of decision rules from the OSDT model. Conclusions: The method demonstrated state-of-the-art predictive performance, affirming its effectiveness. Furthermore, we validate the decision rules derived from the OSDT model by comparing them with established medical knowledge. These interpretable results (eg, attribute importance and decision rules from the OSDT model) underscore the potential of our approach in prioritizing patient urgency for echocardiogram appointments and can be extended to prioritize other laboratory test appointments using electronic health record data.
- 29 January 2025
- Published by JMIR Publications Inc. in JMIR AI
- Vol. 4 (1) , e64188
Open AccessUrgency Prediction for Medical Laboratory Tests Through Optimal Sparse Decision Tree: Case Study With Echocardiograms
Background: In the contemporary realm of health care, laboratory tests stand as cornerstone components, driving the advancement of precision medicine. These tests offer intricate insights into a variety of medical conditions, thereby facilitating diagnosis, prognosis, and treatments. However, the accessibility of certain tests is hindered by factors such as high costs, a shortage of specialized personnel, or geographic disparities, posing obstacles to achieving equitable health care. For example, an echocardiogram is a type of laboratory test that is extremely important and not easily accessible. The increasing demand for echocardiograms underscores the imperative for more efficient scheduling protocols. Despite this pressing need, limited research has been conducted in this area. Objective: The study aims to develop an interpretable machine learning model for determining the urgency of patients requiring echocardiograms, thereby aiding in the prioritization of scheduling procedures. Furthermore, this study aims to glean insights into the pivotal attributes influencing the prioritization of echocardiogram appointments, leveraging the high interpretability of the machine learning model. Methods: Empirical and predictive analyses have been conducted to assess the urgency of patients based on a large real-world echocardiogram appointment dataset (ie, 34,293 appointments) sourced from electronic health records encompassing administrative information, referral diagnosis, and underlying patient conditions. We used a state-of-the-art interpretable machine learning algorithm, the optimal sparse decision tree (OSDT), renowned for its high accuracy and interpretability, to investigate the attributes pertinent to echocardiogram appointments. Results: The method demonstrated satisfactory performance (F1-score=36.18% with an improvement of 1.7% and F2-score=28.18% with an improvement of 0.79% by the best-performing baseline model) in comparison to the best-performing baseline model. Moreover, due to its high interpretability, the results provide valuable medical insights regarding the identification of urgent patients for tests through the extraction of decision rules from the OSDT model. Conclusions: The method demonstrated state-of-the-art predictive performance, affirming its effectiveness. Furthermore, we validate the decision rules derived from the OSDT model by comparing them with established medical knowledge. These interpretable results (eg, attribute importance and decision rules from the OSDT model) underscore the potential of our approach in prioritizing patient urgency for echocardiogram appointments and can be extended to prioritize other laboratory test appointments using electronic health record data.
- Open Access
- 13 January 2025
- Published by JMIR Publications Inc. in JMIR Preprints
WT-LSTM for Intensive Care Unit Length-of-Stay Prediction with Real-Time Signal (Preprint)
BACKGROUND Efficient allocation of healthcare resources is essential for sustainable hospital operation. Effective intensive care unit (ICU) management is essential for alleviating the financial strain on healthcare systems. Accurate prediction of length-of-stay in ICUs is vital for optimizing capacity planning and resource allocation, with the challenge of achieving early, real-time predictions. OBJECTIVE This study aims to develop a predictive model, namely WT-LSTM, for ICU length-of-stay using only real-time vital sign data. The model is designed for urgent care settings where demographic and historical patient data or lab results may be unavailable; the model leverages real-time inputs to deliver early and accurate ICU length-of-stay predictions. METHODS The proposed model integrates discrete wavelet transformation and Long Short-Term Memory (LSTM) neural networks to filter noise from patients’ vital sign series and improve length-of-stay prediction accuracy. Model performance was evaluated using the eICU database, focusing on ten common ICU admission diagnoses in the database. RESULTS The results demonstrate that WT-LSTM consistently outperforms baseline models, including linear regression, LSTM, and BiLSTM, in predicting ICU length-of-stay using vital sign data, achieving significant improvements in Mean Squared Error (MSE). Specifically, the wavelet transformation component of the model enhances the overall performance of WT-LSTM. Removing this component results in an average decrease of 3.3% in MSE; such a phenomenon is particularly pronounced in specific patient cohorts. The model's adaptability is highlighted through real-time predictions using only 3-hour, 6-hour, 12-hour, and 24-hour input data. Using only three hours of input data, the WT-LSTM model delivers competitive results across the ten most common ICU admission diagnoses, often outperforming APACHE IV, the leading ICU outcome prediction system currently implemented in clinical practice. WT-LSTM effectively captures patterns from vital signs recorded during the initial hours of a patient’s ICU stay, making it a promising tool for early prediction and resource optimization in the ICU. CONCLUSIONS Our proposed WT-LSTM model, based on real-time vital sign data, offers a promising solution for ICU length-of-stay prediction. Its high accuracy and early prediction capabilities hold significant potential for enhancing clinical practice, optimizing resource allocation, and supporting critical clinical and administrative decisions in ICU management.
- 13 January 2025
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessWT-LSTM for Intensive Care Unit Length-of-Stay Prediction with Real-Time Signal (Preprint)
BACKGROUND Efficient allocation of healthcare resources is essential for sustainable hospital operation. Effective intensive care unit (ICU) management is essential for alleviating the financial strain on healthcare systems. Accurate prediction of length-of-stay in ICUs is vital for optimizing capacity planning and resource allocation, with the challenge of achieving early, real-time predictions. OBJECTIVE This study aims to develop a predictive model, namely WT-LSTM, for ICU length-of-stay using only real-time vital sign data. The model is designed for urgent care settings where demographic and historical patient data or lab results may be unavailable; the model leverages real-time inputs to deliver early and accurate ICU length-of-stay predictions. METHODS The proposed model integrates discrete wavelet transformation and Long Short-Term Memory (LSTM) neural networks to filter noise from patients’ vital sign series and improve length-of-stay prediction accuracy. Model performance was evaluated using the eICU database, focusing on ten common ICU admission diagnoses in the database. RESULTS The results demonstrate that WT-LSTM consistently outperforms baseline models, including linear regression, LSTM, and BiLSTM, in predicting ICU length-of-stay using vital sign data, achieving significant improvements in Mean Squared Error (MSE). Specifically, the wavelet transformation component of the model enhances the overall performance of WT-LSTM. Removing this component results in an average decrease of 3.3% in MSE; such a phenomenon is particularly pronounced in specific patient cohorts. The model's adaptability is highlighted through real-time predictions using only 3-hour, 6-hour, 12-hour, and 24-hour input data. Using only three hours of input data, the WT-LSTM model delivers competitive results across the ten most common ICU admission diagnoses, often outperforming APACHE IV, the leading ICU outcome prediction system currently implemented in clinical practice. WT-LSTM effectively captures patterns from vital signs recorded during the initial hours of a patient’s ICU stay, making it a promising tool for early prediction and resource optimization in the ICU. CONCLUSIONS Our proposed WT-LSTM model, based on real-time vital sign data, offers a promising solution for ICU length-of-stay prediction. Its high accuracy and early prediction capabilities hold significant potential for enhancing clinical practice, optimizing resource allocation, and supporting critical clinical and administrative decisions in ICU management.
- Open Access
- 1 January 2025
- Published by HICSS Conference Office
Few-Shot Learning for Chronic Disease Management: Leveraging Large Language Models and Multi-Prompt Engineering with Medical Knowledge Injection
The recent breakthrough in Artificial Intelligence (AI) has resulted in a profound impact on various domains including healthcare. Among them, this study harnesses state-of-the-art AI technology for chronic disease management, specifically in detecting various mental disorders through user-generated textual content. We propose a novel framework that leverages advanced AI techniques, including large language models and multi-prompt engineering. On the depression detection task, our method (F1 = 0.975~0.978) significantly outperforms traditional supervised learning paradigms, including feature engineering (F1 = 0.760) and architecture engineering (F1 = 0.756). Our method can be generalized to other mental disorder detection tasks, including anorexia, pathological gambling, and self-harm (F1 = 0.919~0.978). In addition to the technical contributions, our proposed framework has the potential to improve the well-being of patients, control costs, and establish a more efficient and accessible healthcare system.
- 1 January 2025
- Published by HICSS Conference Office
Open AccessFew-Shot Learning for Chronic Disease Management: Leveraging Large Language Models and Multi-Prompt Engineering with Medical Knowledge Injection
The recent breakthrough in Artificial Intelligence (AI) has resulted in a profound impact on various domains including healthcare. Among them, this study harnesses state-of-the-art AI technology for chronic disease management, specifically in detecting various mental disorders through user-generated textual content. We propose a novel framework that leverages advanced AI techniques, including large language models and multi-prompt engineering. On the depression detection task, our method (F1 = 0.975~0.978) significantly outperforms traditional supervised learning paradigms, including feature engineering (F1 = 0.760) and architecture engineering (F1 = 0.756). Our method can be generalized to other mental disorder detection tasks, including anorexia, pathological gambling, and self-harm (F1 = 0.919~0.978). In addition to the technical contributions, our proposed framework has the potential to improve the well-being of patients, control costs, and establish a more efficient and accessible healthcare system.