loading
loading
NIH grant 1R03DE036394-01: Clinical Informatics of Xerostomia: Characterizing Comorbidities, Social Risk Factors, and Predicting Oral Health Outcomes Using Structured and Unstructured EHR Data
Institute: {'code': 'DE', 'abbreviation': 'NIDCR', 'name': 'National Institute of Dental and Craniofacial Research', 'admin_org_id': '12500070', 'admin_funding_url': ''}. PI: Sepideh Banava. Award: $328000.
Start: 2026-08-28T00:00:00.
Project Summary
Xerostomia (dry mouth) is a widespread and underrecognized oral health condition that affects an estimated
46% of adults in the United States and 22% globally, with the highest prevalence among older adults aged 50–
90. Etiologies include polypharmacy, autoimmune diseases such as Sjögren’s syndrome, prior head and neck
radiation or chemotherapy, and other systemic conditions. Xerostomia significantly impairs oral function and
quality of life, leading to poor oral hygiene, rampant dental caries, periodontal disease, tooth loss, fungal
infections, halitosis, problems with denture retention or fit, and difficulty with chewing, swallowing, tasting, and
speaking. Patients with xerostomia also experience 33% higher dental utilization and expenditures, highlighting
its importance as a major dental public health burden.
Despite its clinical and economic impact, a critical gap exists in our ability to identify xerostomia patients most
at risk for severe oral health complications. Current diagnostic and research approaches rely heavily on
structured data such as ICD codes, which have poor sensitivity for xerostomia. In practice, xerostomia is more
often documented in unstructured clinical notes, making it largely invisible to traditional analytic methods.
Recent advances in natural language processing (NLP) now enable the extraction of patient-reported
symptoms and social risk factors from clinical text, providing new opportunities for more accurate case
identification and comprehensive phenotyping.
The objective of this R03 project is to develop and validate a novel EHR-based phenotype of xerostomia and
use it to build a preliminary AI-driven risk prediction model of oral health complications. We will analyze a large
real-world dataset from UCSF’s De-Identified Clinical Data Warehouse, which contains more than 4.5 million
patients and 198 million clinical notes. Aim 1 will characterize xerostomia and related oral health outcomes
using AI-enabled multimodal phenotyping that integrates structured data, clinical notes, and social factors. Aim
2 will develop and internally validate a preliminary risk prediction model using logistic regression and machine
learning methods to estimate oral health complication risk over time.
The expected outcomes of this project are: (1) a rigorously validated, scalable EHR phenotype of xerostomia,
and (2) a preliminary, interpretable risk model for oral health complications. These deliverables will lay the
groundwork for future R01/R21/R34 applications to design and test an EHR-integrated Clinical Decision
Support (CDS) tool, with the ultimate goal of providing personalized prevention and treatment plans for
xerostomia and related oral health complications, and advancing the quality of life of patients with xerostomia.document fetched 2026-09-06 02:47