Nephrology, Hypertension & Renal Transplantation

The Artificial Intelligence Learns Optimal Treatment Strategies for Hypotension in Surgery


Dr. Tezcan Ozrazgat-Baslanti

Email
Tezcan.OzrazgatBaslanti@medicine.ufl.edu

Phone
(352) 273-6668

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Case Review

Research Project Description:
Patients and physicians make essential decisions on which diagnostic and therapeutic interventions should be performed or deferred under time constraints and uncertainty regarding patients’ diagnoses and predicted response to treatment which may lead to cognitive and judgment errors. We aim to develop a deep reinforcement learning model that would provide individualized and clinically interpretable treatment decisions that could help balance blood pressure during surgery and decrease postoperative complications, and improve patient outcomes in patients undergoing surgery. Current practices in the administration of intravenous fluids and vasopressors during surgery as a treatment of hypotension are suboptimal. Reinforcement learning have the potential to play complementary roles in delivering high-value surgical care through sound judgment and optimal decision-making. We aim to develop a deep RL model that would provide individualized and clinically interpretable treatment decisions that could help balance blood pressure during surgery and decrease postoperative complications in patients undergoing surgery.

Does this project have an international component or travel?
No

Validation of Acute Kidney Injury (AKI) Etiology Computable Phenotypes

Name:
Dr. Azra Bihorac

Email
Abihorac@ufl.edu

Phone
(352) 273-9009

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Literature Review

Research Project Description:
The project aims at validating computable phenotype algorithms to automate AKI etiologies. The role of the medical student will be to review electronic health records to adjudicate AKI etiology phenotyping as well as help prepare results and do literature review and write introduction and discussion parts of the manuscript.

Does this project have an international component or travel?
No

Validation of drug-associated acute kidney injury computable phenotypes


Dr. Azra Bihorac

Email
Abihorac@ufl.edu

Phone
(352) 273-9009

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Literature Review

Research Project Description:
The project aims at validating computable phenotype algorithms to identify drug-associated acute kidney injury (D-AKI). The role of the medical student will be to review electronic health records to adjudicate D-AKI as well as help prepare results and do literature review and write introduction and discussion parts of the manuscript.

Does this project have an international component or travel?
No

Sepsis Computable Phenotypes


Dr. Azra Bihorac

Email
abihorac@ufl.edu

Phone
(325) 273-9009

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Literature Review

Research Project Description:
The project aims at validating computable phenotype algorithms to automate sepsis diagnosis. The role of the medical student will be to review electronic health records to adjudicate sepsis onset and sepsis type phenotyping as well as help prepare results and do literature review and write introduction and discussion parts of the manuscript.

Does this project have an international component or travel?
No

Defining Postoperative Intensity of Care


Dr. Azra Bihorac

Email
abihorac@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
After major inpatient surgery, patients’ risk for critical illness and death, summarized as patient acuity, should match triage destination (i.e. intensive care unit (ICU) vs. ward) and the frequency of vital sign monitoring and laboratory testing, summarized as intensity of care. There is no validated, unifying definition for postoperative intensity of care. Our objective is to develop and validate postoperative intensity of care definitions. This project will test the hypothesis that postoperative intensity of care can be autonomously adjudicated in EHRs as optimal (low-acuity and low-intensity or high-acuity and high-intensity) or suboptimal, which may be insufficient (high-acuity and low-intensity), or excessive (low-acuity and high-intensity), and that suboptimal intensity of care will be common and associated with increased in-hospital and 1-year mortality, morbidity, and costs. Our approach will be to use retrospective preoperative, intraoperative, and postoperative EHR data from surgical inpatients at two institutions, classify high- and low-acuity with a random forest classifier that predicts postoperative critical illness and death, classify high- and low-intensity of care according to ICU vs. ward location and frequency of vital sign monitoring and laboratory test orders, and compare in-hospital and 1-year mortality, morbidity, and costs among patients receiving optimal, insufficient, and excessive intensity of care.

Does this project have an international component or travel?
No

The Artificial Intelligence Learns Optimal Treatment Strategies for Hypotension in Surgery


Dr. Azra Bihorac

Email
abihorac@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
Patients and physicians make essential decisions on which diagnostic and therapeutic interventions should be performed or deferred under time constraints and uncertainty regarding patients’ diagnoses and predicted response to treatment which may lead to cognitive and judgment errors. We aim to develop a deep reinforcement learning model that would provide individualized and clinically interpretable treatment decisions that could help balance blood pressure during surgery and decrease postoperative complications, and improve patient outcomes in patients undergoing surgery. Current practices in the administration of intravenous fluids and vasopressors during surgery as a treatment of hypotension are suboptimal. Reinforcement learning have the potential to play complementary roles in delivering high-value surgical care through sound judgment and optimal decision-making. We aim to develop a deep RL model that would provide individualized and clinically interpretable treatment decisions that could help balance blood pressure during surgery and decrease postoperative complications in patients undergoing surgery

Does this project have an international component or travel?
No

ALPHA-Artificial intelligence Augmented Acute Kidney Injury Subphenotypes


Dr. Azra Bihorac

Email
abihorac@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
Acute kidney injury (AKI) is one of the most common complications among hospitalized patients and is associated with morbidity,mortality and increased health cost. There is vast heterogeneity in cause, severity, trajectory and outcomes of AKI that are influenced by inherent disease mechanisms, the patient’s response and the care process. This heterogeneous syndrome encompasses a vast, multidimensional array of clinical, socioeconomic and biological features. Different combinations of these features may naturally cluster into previously undescribed subsets or “clinical phenotypes” that may have different clinical trajectories. These phenotypes must be identifiable soon after AKI onset to guide treatment, and efforts to determine such phenotypes have remained limited. Paradoxically the abundant real-time physiologic, laboratory, clinical notes, and other clinical data are available in the electronic health records (EHR) but their magnitude and complexity often overwhelms physicians’ ability yto comprehend, retain, and organize the information in an optimal and timely way. The overall objective of this application is to identify AKI subphenotypes by developing and validating a generalizable, reproducible and fair algorithm using multimodal clinical data and novel deep learning (DL) technologies.

Does this project have an international component or travel?
No

Clinical phenotypes for AKI using clustering of time series


Dr. Azra Bihorac

Email
Abihorac@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
The purpose of this study is to use clustering to predict pathophysiologic signatures and clinical trajectory during the early stages of acute kidney injury. We hypothesized that clustering analysis of electronic health record data would reveal distinct, predictable patient phenotypes with unique pathophysiological signatures and clinical trajectories.

Does this project have an international component or travel?
No

Integration of IntraOperative data with Preoperative data for improved Postoperative Sepsis Risk prediction


Dr. Azra Bihorac

Email
Abihorac@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
The project aims at developing prediction models to help study patient risk scores for postoperative sepsis complication given preoperative medical history and intraoperative EMR responses. The role of the medical student will be to help prepare results and do literature review and write introduction and discussion parts of the manuscript.

Does this project have an international component or travel?
No

A Patient-Focused Collaborative Hospital Repository Uniting Standards (CHoRUS) for Equitable AI


Dr. Azra Bihorac

Email
abihorac@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
The overall goal of this proposal is to develop and implement an ecosystem for artificial intelligence (AI) in Critical Care (AICC) to accomplish the Grand Challenge of Improving Recovery from Acute Illness through Equitable AI. Infrastructure is urgently needed for AICC. In 2020 alone, 95,596 available U.S. intensive care unit beds streamed over 14 petabytes, but few waveforms were recovered—despite a pandemic in dire need of AI-ready data from critically ill populations. Similarly, large NIH-funded networks lack a platform to store high-resolution data. However, high-resolution data in context is key to unlocking phenotyping, prediction, and decision making for managing sepsis, seizures, cardiac arrest, heart failure, lung injury, and intracranial pressure. Existing large, high-resolution datasets are available only at single centers without geographic, demographic, and practice-pattern diversity needed to ensure model generalization. This proposal addresses a critical grand challenge to improve recovery from acute illnesses through equitable AI, creating an AICC ecosystem for our most severe and vulnerable patients
The project deliverable – the CHoRUS dataset – will be a flagship data resource for the Critical Care AI/ML community. We will develop a tool to obtain data for various exposures contributing to a person’s social determinates of health exposome (SDoH) which may impact health more than health care.56 A systematic analysis of the SDoH literature, guided by the Healthy People 2030 Social Determinants of Health Framework will identify data sources and extract relevant variables.

Does this project have an international component or travel?
No

Federated Learning in Critical Care and Surgery


Dr. Benjamin Shickel

Email
shickelb@ufl.edu

Phone
(352) 273-5995

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation

This project is primarily:
Clinical

Research Project Description:
Federated learning (FL) is a distributed method for training AI algorithms using datasets spread across multiple decentralized devices. FL is especially suited for medical AI applications, where patient privacy concerns often prevent the sharing of datasets across institutions. In this project, the role of the medical student(s) will be to research and explore existing literature on federated learning in critical care and surgical settings, and to evaluate patient-level prediction accuracy in a simulated federated learning environment using existing medical AI models.

Does this project have an international component or travel?
No

Digital pathology assessment of renal glomeruli using AI-driven user-friendly software

Dr. Pinaki Sarder
Pinaki.Sarder@medicine.ufl.edu

Phone
(352) 294-8580

Faculty Department/Division
Nephrology, Hypertension & Renal Transplantation


This project is primarily:
Clinical

Research Project Description:
For this study students will computationally interrogate a dataset consisting of renal biopsy whole slide images (WSIs) from a range of pathologies. Quantitative analysis of glomeruli, annotated using AI, will consist of extracting a large number of objective measures summarizing histological characteristics that are informative toward disease classification and sub-categorization. This project is a great opportunity for students looking to get experience in modern, high-throughput, diagnostic pipelines and designing their own machine learning (ML) models via a user-friendly interface.

Does this project have an international component or travel?
No