AI/QI Incubator

The AI/QI Incubator is a multidisciplinary team of AI engineers, analysts, and project managers working to advance the science and practice of patient safety using AI tools.

32 Approved projects submitted for RAPiDS Cycle 2

24 UF College of Medicine departments submitted projects for RAPiDS Cycle 2

7 Full-time support staff involved with AI modeling

140 + Stakeholders involved in various QI-related projects

RapIds

rapids

Rapid AI Prototyping and Development for patient Safety

Rapid AI Prototyping and Development for Patient Safety (RAPiDS) is an internal mechanism for developing, testing, and advancing innovative AI quality improvement efforts by bringing physicians and the AI/QI Incubator together to translate AI models into clinical improvements.

MRI scan with healthcare workers.

Cycle 1

RAPiDS Cycle 1 focused on developing a Patient Safety Graph (PSG) as a test of rapidly prototyping AI-enabled infrastructure. The PSG is one of the first AI projects at UF Health that focuses on clinical care pathways. On average, PSG model construction time was reduced from months to weeks. Using graph analysis methods, we created a multigraph that maps how patients, patient characteristics, caregivers, medications, laboratory values, surgeries, intra-hospital transfers, and more are interconnected. Understanding how these connections evolve will enable us to improve AI predictions of patient outcomes.


RAPiDS Cycle 1 Patient Safety Graph.
Example Patient Safety Graph of health care providers involved in the perioperative care of cardiothoracic surgical patients. Colors denote providers, and the size of nodes depicts the degree of centrality of each provider in the network of care.

Accomplishments

  • The establishment of the Rapid AI Prototyping and Development for Patient Safety (RAPiDS) Cycle 1 grant program enabled effective collaboration among physicians, AI engineers, analysts, and project managers, enhancing our AI-driven quality improvement efforts.
  • RAPiDS Cycle 1 grants developed and tested the AI Labs for Patient Safety (ALPS), providing a secure data and computing infrastructure for quality improvement. Notably, developing the Patient Safety Graph (PSG) advanced AI predictions by highlighting the interconnectedness of patients, caregivers, and care delivery processes.
  • Initiating RAPiDS Cycle 2 grants marked our ongoing commitment to progress. By fostering collaboration between College of Medicine physicians and the AI/QI Incubator, we are working to leverage AI tools for clinical quality improvement in critical health areas across Florida.

Cycle 2

RAPiDS Cycle 2 builds on the RAPiDS Cycle 1 and ALPS infrastructure to connect College of Medicine physicians with the AI/QI Incubator to solve persistent challenges in high-quality clinical care.   RAPiDS Cycle 2 is leveraging AI tools to advance the science and practice of clinical quality improvement for three top health priorities in Florida: maternal care; surgery, trauma, and acute injury; and Alzheimer’s disease and related dementias and care of older adults. RAPiDS-supported AI resources include geospatial data and social determinants of health, Patient Safety Graph and care pathway optimization using Graph Neural Networks, and Natural Language Processing (NLP) of clinical text documents.

Goal

RAPiDS Cycle 2 aims to develop transdisciplinary collaborations between teams of clinical experts and the AI/QI Incubator to apply new AI infrastructure to improve the quality of three top health priorities in Florida. 

Awarded Projects

RAPiDS Cycle 2 was administered through a competitive, peer-reviewed process to support proposals that have a high likelihood of directly improving the quality of patient care, leveraging AI capabilities available through the AI/QI Incubator, and being well-positioned to be leveraged into extramural funding. RAPiDS Cycle 2 awards were announced on July 7, 2023. Listed below are the awarded projects currently in progress: 


Adetola Louis-Jacques, M.D. Department of Obstetrics and Gynecology

Improving Maternal Health by Addressing Social Needs

This project will improve maternal health and perinatal outcomes by understanding maternal social determinants of health (SDoH) through patient and geospatial assessments, predicting adverse perinatal outcomes using AI models, and shifting measurable postpartum interventions into prenatal phases of care (e.g., clinic access, social work referrals) for at-risk moms. 

The primary AI tool being used is the geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity).

Nicole M. Iovine, M.D., Ph. D. Department of Medicine

Using AI to Improve Key Patient Safety Indicators

This project will improve PSI indicators (e.g., sepsis) by predicting sepsis-related postoperative outcomes, mapping sepsis-related postoperative outcomes to antecedent, latent, high-risk clinical care pathways, and developing internal, candidate process measures linked to sepsis-related PSI outcome measures. 

The primary AI tool being used is the patient safety graph & care pathway optimization using graph neural networks.

HELEN HU, M.D. DEPARTMENT OF PEDIATRICS, DIVISION OF NEONATOLOGY

MILK+: Maximizing Initiatives for Lactation Knowledge: An AI-Powered Solution for Breastfeeding Success

This project will help clinicians identify high-risk women for lactation failure and provide targeted interventions to promote lactation success, increasing the number of postpartum women providing any human milk at discharge.

The primary AI tools being used are geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity) and natural language processing (NLP) of clinical text documents.

KEITH HOWELL, M.D. DEPARTMENT OF ANESTHESIOLOGY

Patient Specific Maximum Surgical Blood Ordering Schedule (to guide lab ordering and blood product availability for elective surgical cases)

This project will develop personalized bleeding and transfusion risk scores to update policies surrounding perioperative transfusion of blood products, decreasing the volume of preoperative labs, crossmatched blood products, and unwarranted transfusions. 

The primary AI tool being used is the geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity).

Carol A. Mathews, M.D. DEPARTMENT OF Psychiatry

Optimization of Treatment Recommendations for Treatment-Resistant Depression in Older Adults

This project will improve access to ECT, transcranial magnetic stimulation, and intranasal ketamine for older adults with treatment-resistant depression.

The primary AI tools being used are geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity), the patient safety graph & care pathway optimization using graph neural networks and natural language processing (NLP) of clinical text documents.

Jeremy Balch, M.D. Department of Surgery

Computer Vision for Optimizing Surgical Instrument Trays

This project will use AI and computer vision to understand surgical instrument utilization and decrease waste and associated costs of infrequently used instruments and trays.

The primary AI tool being used is computer vision.

John Michael DiBianco, M.D. DEPARTMENT OF Urology

Using AI To Identify Patients With Nephrolithiasis at Risk for Unplanned Healthcare Encounters

This project will use predictive and prescriptive modeling strategies to 1) understand which ambulatory surgery nephrolithiasis patients may be at risk for unplanned postoperative healthcare encounters (baseline rate of 15%), 2) which patients may benefit from elective postoperative admission.

The primary AI tools being used are geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity), the patient safety graph & care pathway optimization using graph neural networks and natural language processing (NLP) of clinical text documents.

Naveen Baskaran, M.D., M.S.H.I. DEPARTMENT OF medicine

Individualized Health Education: AI-Enabled Recommender System for Personalized Education Targeting Social Determinants of Health (Sdoh) for Heart Failure Patients Over 55

This project will combine patient and clinical process factors in models to predict and decrease the risk of readmission following hospitalization for heart failure intervention. Interventions include clinical process modification and patient-specific, education-based interventions linked to specific cause(s) of HF readmission.

The primary AI tool being used is the geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity).

Amira Quevedo, M.D. Department of Obstetrics and Gynecology

Disparities in Women’s Healthcare Project- A Path to Quality and Equitable Care a Data-Informed Approach With Artificial Intelligence Integration

This project will decrease gaps in equitable access to appropriate diagnosis and effective therapeutic interventions for women with endometriosis and adenomyosis through the development of SDoH-aware patient care pathways.

The primary AI tools being used are geospatial data & social determinants of health (including social and economic conditions impacting health, such as food insecurity), the patient safety graph & care pathway optimization using graph neural networks and natural language processing (NLP) of clinical text documents.

Eric Jeng, N.D., M.B.A., F.A.C.S., F.A.C.C. Department of Surgery

Risk Modeling in Aortic Aneurysms

This project examines QI opportunities related to surgical management of aortic disease, including reducing post-op sepsis and post-op respiratory failure.

The primary AI tools being used are the patient safety graph & care pathway optimization using graph neural networks and natural language processing (NLP) of clinical text documents.

ALPS

Alps

AI Labs for Patient Safety

Protected health information requires special security. While research use-cases of health care data can generally use de-identified materials, QI and clinical operational use-cases may occasionally require the use of patient and clinician identifiers. These use-cases also carry a wider array of stakeholders beyond publishers and funding agencies and can include department chairs, division chiefs and college leadership. We created the AI Labs for Patient Safety, or ALPS, to address this need.

Computer wires highlighted with purple, pink and orange light.

What is ALPS?

ALPS links PHI to GPU-accelerated, high-performance computing infrastructure within a secure enclave. This permits AI analyses to be performed on identifiable data. For QI and hospital operations, this allows us to extend research-level AI models to operational datasets.

ALPS is principally managed by Ray Opoku, MSc., in consultation with UF Research Computing and Ron Ison, M.S., P.M.P., in the Department of Anesthesiology.

How is ALPS innovative?

ALPS was a key advance in allowing us to develop temporal graph neural networks (GNNs) for use with PHI.  Because temporal GNNs use date/timestamps of hospital events and require GPU acceleration for creation, and because attempts to anonymize through various obfuscation methods were deemed impermissible by the IRB, temporal GNN models would not be feasible outside of an ALPS-like enclave.

How is ALPS Being Used?

A sub-project within ALPS seeks to develop a core resource for examining social disparities using geospatial analysis. Using this core, we can attach social determinants of health to all modeling efforts, and back-translate findings to maps of disparities and social vulnerabilities. This is an operationally-enabled resource developed by Rulman Pebe, MSc., in consultation with Jiang Bian, Ph.D..