AI for Health
AI for Health Institute

[Team] [Updates] [Contributions] [Projects] [Talks] [Media] [Papers]


Faculty: Chenyang Lu

Collaborators in Medicine

Doctoral Students: Charles Alba, Hangyue Li, Hanyang Liu, Claire Najjuuko, Ruiqi Wang, Ben Warner, Ziqi Xu, Bing Xue, Jingwen Zhang, Daoyi Zhu

MS and Undergraduate Students: Peiqi Gao, Mingzhen Li, Jason Liu, Patrick Lynch, Seung Hyeon Ryu, Olivia Rui, Jacob Shen, Zichen Wang, Chenyu Yang

Past Collaborators and Doctoral Students Thomas Bailey, Roger Chamberlain, Yixin Chen, Octav Chipara, Ruixuan Dai, Greg Hackmann, Chet Hammill, George Despotis, York Jiao, Dingwen Li, Patrick Lyons, Michael Montana, Gruia-Catalin Roman, Susan Stark


Predictive Models for Clinical Outcomes: We have developed machine learning models for predicting clinical and mental health outcomes based on complex and multimodal data from electronic health records (EHR). Specifically, we developed and evaluated machine learning models to 1) predict postoperative complications based on preoperative and intraoperative data, 2) generate early warning alerts of clinical deterioration at general and oncology hospital units, 3) predict clinician burnout based on EHR activity logs.

Patient Monitoring with Wearable Devices:
We have developed and piloted machine learning models and informatics infrastructure to monitor patients using wearable devices. Our clinical studies demonstrated the feasibility of continuously monitoring outpatients using smart wristbands. We observed high levels of patient compliance in wearing wristbands regularly, satisfactory yield in data collection, and high predictive performance of machine learning models using a combination of wearable data and clinical patient characteristics. We have applied our informatics infrastructure and machine learning pipeline to clinical studies of patients undergoing pancreatic surgery, congestive heart failure patients, lung-transplant patients, patients undergoing lumbar fusion, patients under depression therapy, community-dwelling older adults, and youth with diabetes.

Wireless Clinical Monitoring Systems: We have developed large-scale wireless sensor networks for real-time monitoring of patient conditions. We built and deployed a clinical monitoring system integrating portable pulse oximeters, a wireless sensor network spanning multiple hospital floors, and informatics infrastructure. In a 14-month clinical trial our system was deployed in six hospital wards of Barnes-Jewish Hospital. Our experiences demonstrated the feasibility of reliable vital sign monitoring using a wireless sensor network integrated with the hospital infrastructure. We have also piloted an automated contact tracing system for hospital staff and clinicians in hospitals using Bluetooth devices.

Sample Projects

Predict Surgical Outcomes for Perioperative Care: To improve perioperative care, we developed machine learning models to predict patient risks of postoperative complications. In a cohort study of 111,888 operations at a large academic medical center, our machine learning models exhibited high performance for predicting the risk of postoperative complications related to pneumonia, acute kidney injury, deep vein thrombosis, pulmonary embolism, and delirium. These findings suggest that machine learning models using preoperative and intraoperative data can predict postoperative complications and generate reliable and clinically meaningful interpretations for supporting clinical decisions along the perioperative care continuum. Predict Post-Operative Complications with Wearables: Postoperative complications and hospital readmission are of great concern to surgical patients and health care providers. Wearable devices such as Fitbit wristbands enable long-term and non-intrusive monitoring of patients outside clinical environments. We built machine learning pipeline to make reliable predictions of postoperative complications and readmissions based on noisy and incomplete wearable data. We evaluated the feature engineering approach and machine learning models in a prospective clinical study involving 61 patients undergoing pancreatic surgery. Our machine learning models significantly outperformed the standard surgery risk model used in clinical practice and state-of-the-art machine learning approaches. Detect Mental Disorders Using Wearables: : Depression and anxiety are among the most prevalent mental disorders. Over 50% of patients are not recognized or adequately treated. We studied the problem of detecting depression and anxiety disorders with commercial wearable activity trackers based on a public dataset including 8,996 participants and 1,247 diagnosed with mental disorders. We developed an end-to-end deep learning model to directly learn from daily wearable features and detect mental disorders. Our results demonstrates the feasibility and promise of using wearables to detect mental disorders in a large and diverse community. Predict Depression Treatment Response Using Wearables: A randomized controlled trial (RCT) is used to study the safety and efficacy of new treatments. We developed machine learning models in conjunction with an RCT for personalized predictions of a depression treatment intervention, where patients were longitudinally monitored with wearable devices. We evaluated our approach in an RCT involving 106 patients with depression. Our study represented a promising step in utilizing data collected in RCTs to develop predictive models for precision medicine. Predict Physician Burnout Using Activity Logs of Electronic Health Records: Burnout is a significant public health concern affecting nearly half of the healthcare workforce. We developed the first end-to-end deep learning framework for predicting physician burnout based on electronic health record (EHR) activity logs, digital traces of physician work activities that are available in any EHR system. The experiment on over 15 million clinician activity logs collected from the EHR at a large academic medical center demonstrates the advantages of our proposed framework in predictive performance of physician burnout over state-of-the-art approaches. Early Warning System for Oncology Hospital Wards: About 9% of cancer patients experience complications while hospitalized that lead to a deterioration in their conditions, a transfer to the intensive care unit or even death. We developed a novel machine learning approach for integrating heterogenous clinical data. We applied the proposed approaches to a dataset extracted from the EHR of 20,700 hospitalizations of adult oncology patients in a research hospital, which demonstrated the advantages of the predictive model realistic clinical settings.