Program for Clinical AI
Advancing AI tools in healthcare
We further the science, adoption, and effective use of artificial intelligence in healthcare
Table of Contents
About P4AI
Advancing Artificial Intelligence for Improving Healthcare Delivery
Shaping the future of healthcare by advancing artificial intelligence tools for real-world impact.
The Program for Clinical AI (P4AI) focuses on developing, validating, and implementing artificial intelligence (AI) tools to enhance healthcare delivery. Like other areas of our lives, the role of AI in healthcare delivery will only increase over time. A key objective of P4AI is to investigate AI-enabled tools in real-world settings including monitoring AI model performance for drift, equity, and suitability for addressing specific healthcare challenges.
The P4AI team brings together experts from across the University of Minnesota including the Medical School, School of Public Health, Institute for Health Informatics, and the College of Science and Engineering. We foster both national and international collaborations in natural language processing (NLP) and computer vision including the Federated Computer Vision in Healthcare Collaborative across multiple U.S. academic health systems.
P4AI People
P4AI Leadership
Scientific co-Director: Christopher Tignanelli, MD, MS
Scientific co-Director: Gyorgy Simon, PhD
P4AI Full Members
Structured Healthcare Data: Gyorgy Simon, PhD
Computer Vision: Ju Sun, PhD
Natural Language Processing: Rui Zhang, PhD
Robotics and Surgical Automation: Elliot Arsoniadis, MD, PhD
Project Manager: Marley Crews-Hill
P4AI Research Projects
Rib Fracture Model
Investigator
Matt Bahr
- M Health Fairview
The Problem
Trauma is the fourth leading cause of death for Americans (Gaillard et al, 1990), costing the nation $671 billion each year (Florence et al, 2015). Rib fractures are sustained in nearly 15% of trauma patients, associated with significant morbidity and mortality (Sharma et al, 2008). Regrettably, 55% of older patients who die from thoracic trauma have no injury worse than rib fractures (Kent et al, 2008). Adherence with evidence-based rib fracture practice significantly reduces morbidity and mortality. However, despite best practice treatment algorithms for rib fracture patients and other diseases, adherence is poor.
Our solution
In 2020, we developed and deployed a comprehensive clinical decision support system to help providers and nurses at the point of care to improve the adherence with best practice for patients with rib fracture. Deployed across multiple U.S. Midwest trauma centers, this system includes in-hospital decision support and post-discharge remote patient monitoring. A prospective evaluation showed that using the system improved patient outcomes, including better survival rates, shorter hospital stays, and fewer unplanned ICU transfers.
Unfortunately, adoption of the system has been inconsistent, peaking at 76% and now stabilizing at around 50%. Given this, there is a need for artificial intelligence-based approaches which can identify patients with rib fractures on admission by “interpreting” patient x-ray images and medical records. Such a system could improve utilization of decision support systems to approach 100%, facilitating maximum use of evidence-based practice and reducing healthcare disparities. The objective of this study was to develop and rigorously validate an AI-based diagnostic system for rib fractures through external and real-time prospective validation. Future research aims to compare AI-driven activation of decision support systems with provider-driven activation, in a randomized trial.
Status: in progress
Early Detection Sepsis Model
Investigator
Mike Usher, MD, PhD
- M Health Fairview
The Problem
Sepsis is the body’s uncontrolled response to an infection, leading to organ dysfunction (Singer et al, 2016). It can progress quickly and has a high mortality rate. Early identification and prompt intervention are crucial to saving lives and reducing morbidity.
Our Solution
A cognitive computing model can provide clinical decision support, helping clinicians identify patients who might need to be screened for sepsis. This model can assist clinicians in earlier identification of sepsis, helping them intervene sooner. It can be used directly by on‐site clinicians or as part of a remote patient monitoring program.
Version 1 of the Early Detection Sepsis model helps clinicians identify patients developing sepsis before their condition worsens. It is a Chronicles‐based logistic regression model with approximately 80 variables.
Status: in progress
Publications
Practical Guide to the Use of AI-Enabled Analytics in Research
Kothari AN, Kaji AH, Melton GB
Trustworthiness of a machine learning early warning model in medical and surgical inpatients
Caraballo PJ, Meehan AM, Fischer KM, Rahman P, Simon GJ, Melton GB, Salehinejad H, Borah BJ
A Machine Learning Readmission Risk Prediction Model for Classical and Malignant Hematologic Disease
Bailey A, Wang W, Shannon IV C, Huling J, Tignanelli CJ
A Machine Learning Readmission Risk Prediction Model for Cardiac Disease
Bailey A, Wang W, Shannon IV C, Huling J, Tignanelli CJ
Meeting the Artificial Intelligence Needs of U.S. Health Systems
Lyons PG, Dorr DA, Melton GB, Singh K, Payne PRO
Incorporating Patient Values in Large Language Model Recommendations for Surrogate and Proxy Decisions
Nolan VJ, Balch JA, Baskaran NP, Shickel B, Efron PA, Upchurch GR Jr, Bihorac A, Tignanelli CJ, Moseley RE, Loftus TJ
Consensus modeling: Safer transfer learning for small health systems
Tourani R, Murphree DH, Sheka A, Melton GB, Kor DJ, Simon GJ
Prospective validation of a hospital triage predictive model to decrease undertriage: an EAST multicenter study
Biesboer EA, Pokrzywa CJ, Karam BS, Chen B, Szabo A, Teng BQ, Bernard MD, Bernard A, Chowdhury S, Hayudini AE, Radomski MA, Doris S, Yorkgitis BK, Mull J, Weston BW, Hemmila MR, Tignanelli CJ, de Moya MA, Morris RS
A Symptom-Based Natural Language Processing Surveillance Pipeline for Post-COVID-19 Patients
Silverman GM, Rajamani G, Ingraham NE, Glover JK, Sahoo HS, Usher M, Zhang R, Ikramuddin F, Melnik TE, Melton GB, Tignanelli CJ
Re-Aiming Equity Evaluation in Clinical Decision Support: A Scoping Review of Equity Assessments in Surgical Decision Support Systems
Ingraham N, Jones E, King S, Dries J, Phillips M, Loftus T, Evans H, Melton G, Tignanelli CJ
Artificial Intelligence-enabled Decision Support in Surgery: State-of-the-art and Future Directions
Loftus T, Altieri M, Balch J, Abbott K, Choi J, Marwaha J, Hashimoto D, Brat G, Raftopoulos Y, Evans H, Jackson G, Walsh D, Tignanelli CJ
Validation of a Proprietary Deterioration Index Model and Performance in Hospitalized Adults
Byrd T, Southwell B, Ravishankar A, Tran T, Kc A, Phelan T, Melton-Meaux G, Usher M, Scheppmann D, Switzer S, Simon G, Tignanelli CJ
A novel, evidence-based, comprehensive clinical decision support system improves outcomes for patients with traumatic rib fractures
Jones E, Ninkovic I, Bahr M, Dodge S, Doering M, Martin D, Ottosen J, Allen T, Melton G, Tignanelli CJ
Ability of artificial intelligence to identify self-reported race in chest x-ray using pixel intensity counts
Burns JL, Zaiman Z, Vanschaik J, Luo G, Peng L, Price B, Mathias G, Mittal V, Sagane A, Tignanelli C, Chakraborty S, Gichoya JW, Purkayastha S
P4AI News
Federated Learning Collaborative
We are part of a national community of health data science and AI researchers who are facilitating research collaboration and the development of AI and machine learning healthcare algorithms.