Wednesday, February 5, 2025
2/5/2025
ABSTRACT Diverticulitis is among the most common gastrointestinal conditions leading to emergency department visits and hospitalizations. Uncomplicated diverticulitis is usually mild and self-limiting, but about 10% of patients will develop complications including abscess, fistula, obstruction, bleeding, or perforation, and over 20-30% will experience at least one recurrence or ongoing gastrointestinal complaints. Although up to 50% of patients with complicated diverticulitis require emergent surgery, there remains considerable uncertainty about whether most other diverticulitis patients should be medically managed or offered elective segmental colectomy. Recent guidelines recommend a personalized approach, with consideration of segmental resection according to risk for recurrence or future complications. However, little is known about factors predisposing patients to unfavorable outcomes. Among patients with their first episode of diverticulitis, there is a significant unmet need to identify those at higher risk of recurrence or future complications to guide decision-making about clinical management. Thus, I propose to apply artificial intelligence (AI) techniques to clinical data including natural language processing (NLP) for narrative notes (Aim 1) and deep learning for imaging (Aim 2) in a large-scale electronic health record (EHR) database to predict recurrence and long-term complications of diverticulitis. Building upon findings from the first two aims, in Aim 3, I will validate the prediction model for diverticulitis recurrence in an external EHR dataset. My overarching goal is to develop a clinical decision support tool incorporating NLP of clinical documentation, computed tomography (CT)-based deep learning, lifestyle, and clinical factors to predict outcomes in diverticulitis. Findings from this project will enable the utilization of comprehensive information from EHR for diverticulitis research, advance understanding of novel predictors of long-term outcomes of diverticulitis, and contribute to the development of a clinical prediction tool to facilitate risk stratification and guide disease management. Through this work, I will receive formal training in using NLP for unstructured clinical notes; developing deep learning algorithms for imaging; and modeling for clinical risk prediction. I will be mentored/advised by an interdisciplinary team that includes Dr. Andrew Chan, a gastroenterologist and epidemiologist with expertise in diverticulitis; Dr. Katherine Andriole, an expert in radiology imaging informatics, deep learning, and machine learning; Dr. Ashwin Ananthakrishnan, a gastroenterologist with expertise in utilizing NLP in EHR; Dr. Lisa Strate, a gastroenterologist with expertise in diverticular disease; Dr. Avinash Kambadakone-Ramesh, a radiologist with expertise in abdominal imaging and advanced CT applications; and Dr. Alisa Manning, a biostatistician with expertise in clinical research using EHR data. The outstanding training opportunities with key leaders in their respective fields will provide me with new skills in the application of AI methods including NLP, machine learning, and deep learning to clinical data and risk prediction modeling, positioning me for a successful, independent career in advancing precision medicine in digestive diseases.
HHS’ Tracking Accountability in Government Grants System (TAGGS) website provides access to detailed descriptions of grants, loans, aggregated direct payments and other types of financial assistance awarded by the HHS Operating Divisions (OPDIVs) and the Office of the Secretary Staff Divisions (STAFFDIVs).