Zewen Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zewen Wu has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Michael L. Burns
Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan
3 shared publications - Aaron Retsky
Department of Anesthesiology University of Michigan Health System Ann Arbor, MI United States
2 shared publications - Alexander Hoffmann
Department of Anesthesiology University of Michigan Health System Ann Arbor, MI United States
2 shared publications - Anik Sinha
DIAL, ECE, North South University, Dhaka, Bangladesh
2 shared publications - David Chesney
Department of Electrical Engineering and Computer Science University of Michigan Ann Arbor, MI United States
2 shared publications - Nirav Shah
Department of Anesthesiology, University of Michigan, Ann Arbor, MI, United States
2 shared publications
Latest Publications
- Open Access
- 7 August 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e84454
Machine Learning to Identify Point-of-Care Ultrasound and Evaluate Standardized Documentation: Retrospective Operational Cohort Study
Background: Point-of-care ultrasound (POCUS) is integral to obstetrics and gynecology (OBGYN), offering bedside diagnostic and therapeutic advantages. Despite its widespread adoption, accurate documentation and billing remain challenging due to inconsistent workflows, variable free-text note quality, and inefficiencies within electronic health record (EHR) systems. These barriers often result in missed procedural charges and hinder operational, educational, and reimbursement efforts. Objective: This study leveraged machine learning (ML) to automatically identify POCUS procedures within clinical notes and assessed the effect of implementing standardized procedure documentation (ProcDoc) templates on billing capture accuracy and efficiency. Methods: We conducted a multipart retrospective cohort study at a large academic medical center using EHRs from January 2018 to August 2024 across 11 OBGYN clinic sites. ML models (LightGBM [light gradient boosting machine] and BioClinBERT [biomedical and clinical bidirectional encoder representations from transformers]) were trained on clinical encounter notes to classify POCUS procedures and validated against Current Procedural Terminology (CPT) manual code assignments. In February 2023, a standardized ProcDoc smart form was introduced to streamline POCUS documentation and automatically trigger CPT billing codes. Preintervention and postintervention periods were compared using ML metrics and manual billing audits. Outcomes included model accuracy, recall, precision, adoption rates, improvement in billing recapture, and usage of ProcDoc templates. Results: A total of 559,029 encounters from 109,776 unique patients were analyzed. The BioClinBERT model (accuracy 0.97; F 1 -score 0.55-0.63) demonstrated a robust ability to identify documented and missed procedures in free-text clinical notes. ProcDoc adoption reached 75.1% within 12 months, supported by comprehensive staff education. Billing recapture—the proportion of charges missed by providers but later identified—dropped from 10.0% preintervention to 2.4% postintervention, primarily arising from the shift toward auto-capturing documentation (odds ratio 0.22, 95% CI 0.17‐0.30; P <.001), with overall POCUS billing slightly increased (+0.6%). Most postintervention CPT codes (1812/2404, 75.4%) originated from ProcDoc templates, confirming improved workflow efficiency and reduced manual audit burden. Model analysis and billing metrics demonstrated that improvements were associated with workflow changes and not an increase in procedure frequency. Conclusions: ML modeling proved effective for extracting POCUS procedures from clinical documentation and serving as an evaluation tool for workflow interventions. Standardized documentation with ProcDoc significantly enhanced charge capture accuracy and reduced dependence on manual chart reviews and billing reconciliation. This approach highlights the use of ML as a retrospective auditing and evaluation tool for assessing clinical workflow interventions. Broader application of similar strategies could address documentation inefficiencies and promote sustainability across health care settings.
- 7 August 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e84454
Open AccessMachine Learning to Identify Point-of-Care Ultrasound and Evaluate Standardized Documentation: Retrospective Operational Cohort Study
Background: Point-of-care ultrasound (POCUS) is integral to obstetrics and gynecology (OBGYN), offering bedside diagnostic and therapeutic advantages. Despite its widespread adoption, accurate documentation and billing remain challenging due to inconsistent workflows, variable free-text note quality, and inefficiencies within electronic health record (EHR) systems. These barriers often result in missed procedural charges and hinder operational, educational, and reimbursement efforts. Objective: This study leveraged machine learning (ML) to automatically identify POCUS procedures within clinical notes and assessed the effect of implementing standardized procedure documentation (ProcDoc) templates on billing capture accuracy and efficiency. Methods: We conducted a multipart retrospective cohort study at a large academic medical center using EHRs from January 2018 to August 2024 across 11 OBGYN clinic sites. ML models (LightGBM [light gradient boosting machine] and BioClinBERT [biomedical and clinical bidirectional encoder representations from transformers]) were trained on clinical encounter notes to classify POCUS procedures and validated against Current Procedural Terminology (CPT) manual code assignments. In February 2023, a standardized ProcDoc smart form was introduced to streamline POCUS documentation and automatically trigger CPT billing codes. Preintervention and postintervention periods were compared using ML metrics and manual billing audits. Outcomes included model accuracy, recall, precision, adoption rates, improvement in billing recapture, and usage of ProcDoc templates. Results: A total of 559,029 encounters from 109,776 unique patients were analyzed. The BioClinBERT model (accuracy 0.97; F 1 -score 0.55-0.63) demonstrated a robust ability to identify documented and missed procedures in free-text clinical notes. ProcDoc adoption reached 75.1% within 12 months, supported by comprehensive staff education. Billing recapture—the proportion of charges missed by providers but later identified—dropped from 10.0% preintervention to 2.4% postintervention, primarily arising from the shift toward auto-capturing documentation (odds ratio 0.22, 95% CI 0.17‐0.30; P <.001), with overall POCUS billing slightly increased (+0.6%). Most postintervention CPT codes (1812/2404, 75.4%) originated from ProcDoc templates, confirming improved workflow efficiency and reduced manual audit burden. Model analysis and billing metrics demonstrated that improvements were associated with workflow changes and not an increase in procedure frequency. Conclusions: ML modeling proved effective for extracting POCUS procedures from clinical documentation and serving as an evaluation tool for workflow interventions. Standardized documentation with ProcDoc significantly enhanced charge capture accuracy and reduced dependence on manual chart reviews and billing reconciliation. This approach highlights the use of ML as a retrospective auditing and evaluation tool for assessing clinical workflow interventions. Broader application of similar strategies could address documentation inefficiencies and promote sustainability across health care settings.
- Open Access
- 7 September 2023
- Published by JMIR Publications Inc. in JMIR Biomedical Engineering
- Vol. 8 (1) , e46653
Development and Testing of a Data Capture Device for Use With Clinical Incentive Spirometers: Testing and Usability Study
Background: The incentive spirometer is a basic and common medical device from which electronic health care data cannot be directly collected. As a result, despite numerous studies investigating clinical use, there remains little consensus on optimal device use and sparse evidence supporting its intended benefits such as prevention of postoperative respiratory complications. Objective: The aim of the study is to develop and test an add-on hardware device for data capture of the incentive spirometer. Methods: An add-on device was designed, built, and tested using reflective optical sensors to identify the real-time location of the volume piston and flow bobbin of a common incentive spirometer. Investigators manually tested sensor level accuracies and triggering range calibrations using a digital flowmeter. A valid breath classification algorithm was created and tested to determine valid from invalid breath attempts. To assess real-time use, a video game was developed using the incentive spirometer and add-on device as a controller using the Apple iPad. Results: In user testing, sensor locations were captured at an accuracy of 99% (SD 1.4%) for volume and 100% accuracy for flow. Median and average volumes were within 7.5% (SD 6%) of target volume sensor levels, and maximum sensor triggering values seldom exceeded intended sensor levels, showing a good correlation to placement on 2 similar but distinct incentive spirometer designs. The breath classification algorithm displayed a 100% sensitivity and a 99% specificity on user testing, and the device operated as a video game controller in real time without noticeable interference or delay. Conclusions: An effective and reusable add-on device for the incentive spirometer was created to allow the collection of previously inaccessible incentive spirometer data and demonstrate Internet-of-Things use on a common hospital device. This design showed high sensor accuracies and the ability to use data in real-time applications, showing promise in the ability to capture currently inaccessible clinical data. Further use of this device could facilitate improved research into the incentive spirometer to improve adoption, incentivize adherence, and investigate the clinical effectiveness to help guide clinical care.
- 7 September 2023
- Published by JMIR Publications Inc. in JMIR Biomedical Engineering
- Vol. 8 (1) , e46653
Open AccessDevelopment and Testing of a Data Capture Device for Use With Clinical Incentive Spirometers: Testing and Usability Study
Background: The incentive spirometer is a basic and common medical device from which electronic health care data cannot be directly collected. As a result, despite numerous studies investigating clinical use, there remains little consensus on optimal device use and sparse evidence supporting its intended benefits such as prevention of postoperative respiratory complications. Objective: The aim of the study is to develop and test an add-on hardware device for data capture of the incentive spirometer. Methods: An add-on device was designed, built, and tested using reflective optical sensors to identify the real-time location of the volume piston and flow bobbin of a common incentive spirometer. Investigators manually tested sensor level accuracies and triggering range calibrations using a digital flowmeter. A valid breath classification algorithm was created and tested to determine valid from invalid breath attempts. To assess real-time use, a video game was developed using the incentive spirometer and add-on device as a controller using the Apple iPad. Results: In user testing, sensor locations were captured at an accuracy of 99% (SD 1.4%) for volume and 100% accuracy for flow. Median and average volumes were within 7.5% (SD 6%) of target volume sensor levels, and maximum sensor triggering values seldom exceeded intended sensor levels, showing a good correlation to placement on 2 similar but distinct incentive spirometer designs. The breath classification algorithm displayed a 100% sensitivity and a 99% specificity on user testing, and the device operated as a video game controller in real time without noticeable interference or delay. Conclusions: An effective and reusable add-on device for the incentive spirometer was created to allow the collection of previously inaccessible incentive spirometer data and demonstrate Internet-of-Things use on a common hospital device. This design showed high sensor accuracies and the ability to use data in real-time applications, showing promise in the ability to capture currently inaccessible clinical data. Further use of this device could facilitate improved research into the incentive spirometer to improve adoption, incentivize adherence, and investigate the clinical effectiveness to help guide clinical care.
- Open Access
- 20 February 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Development and Testing of a Data Capture Device for Use With Clinical Incentive Spirometers: Testing and Usability Study (Preprint)
BACKGROUND The incentive spirometer is a basic and common medical device from which electronic health care data cannot be directly collected. As a result, despite numerous studies investigating clinical use, there remains little consensus on optimal device use and sparse evidence supporting its intended benefits such as prevention of postoperative respiratory complications. OBJECTIVE The aim of the study is to develop and test an add-on hardware device for data capture of the incentive spirometer. METHODS An add-on device was designed, built, and tested using reflective optical sensors to identify the real-time location of the volume piston and flow bobbin of a common incentive spirometer. Investigators manually tested sensor level accuracies and triggering range calibrations using a digital flowmeter. A valid breath classification algorithm was created and tested to determine valid from invalid breath attempts. To assess real-time use, a video game was developed using the incentive spirometer and add-on device as a controller using the Apple iPad. RESULTS In user testing, sensor locations were captured at an accuracy of 99% (SD 1.4%) for volume and 100% accuracy for flow. Median and average volumes were within 7.5% (SD 6%) of target volume sensor levels, and maximum sensor triggering values seldom exceeded intended sensor levels, showing a good correlation to placement on 2 similar but distinct incentive spirometer designs. The breath classification algorithm displayed a 100% sensitivity and a 99% specificity on user testing, and the device operated as a video game controller in real time without noticeable interference or delay. CONCLUSIONS An effective and reusable add-on device for the incentive spirometer was created to allow the collection of previously inaccessible incentive spirometer data and demonstrate Internet-of-Things use on a common hospital device. This design showed high sensor accuracies and the ability to use data in real-time applications, showing promise in the ability to capture currently inaccessible clinical data. Further use of this device could facilitate improved research into the incentive spirometer to improve adoption, incentivize adherence, and investigate the clinical effectiveness to help guide clinical care.
- 20 February 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessDevelopment and Testing of a Data Capture Device for Use With Clinical Incentive Spirometers: Testing and Usability Study (Preprint)
BACKGROUND The incentive spirometer is a basic and common medical device from which electronic health care data cannot be directly collected. As a result, despite numerous studies investigating clinical use, there remains little consensus on optimal device use and sparse evidence supporting its intended benefits such as prevention of postoperative respiratory complications. OBJECTIVE The aim of the study is to develop and test an add-on hardware device for data capture of the incentive spirometer. METHODS An add-on device was designed, built, and tested using reflective optical sensors to identify the real-time location of the volume piston and flow bobbin of a common incentive spirometer. Investigators manually tested sensor level accuracies and triggering range calibrations using a digital flowmeter. A valid breath classification algorithm was created and tested to determine valid from invalid breath attempts. To assess real-time use, a video game was developed using the incentive spirometer and add-on device as a controller using the Apple iPad. RESULTS In user testing, sensor locations were captured at an accuracy of 99% (SD 1.4%) for volume and 100% accuracy for flow. Median and average volumes were within 7.5% (SD 6%) of target volume sensor levels, and maximum sensor triggering values seldom exceeded intended sensor levels, showing a good correlation to placement on 2 similar but distinct incentive spirometer designs. The breath classification algorithm displayed a 100% sensitivity and a 99% specificity on user testing, and the device operated as a video game controller in real time without noticeable interference or delay. CONCLUSIONS An effective and reusable add-on device for the incentive spirometer was created to allow the collection of previously inaccessible incentive spirometer data and demonstrate Internet-of-Things use on a common hospital device. This design showed high sensor accuracies and the ability to use data in real-time applications, showing promise in the ability to capture currently inaccessible clinical data. Further use of this device could facilitate improved research into the incentive spirometer to improve adoption, incentivize adherence, and investigate the clinical effectiveness to help guide clinical care.