A LARGE POPULATION WITH CLINICALLY MEANINGFUL EVIDENCE OF ALCOHOL USE DISORDER (AUD) IS NOT CAPTURED BY DIAGNOSIS CODES ALONE: EVIDENCE FROM NORSTELLALINQ STRUCTURED AND HUMAN-IN-THE-LOOP LARGE LANGUAGE MODEL (LLM) EXTRACTED CLINICAL DATA
Author(s)
Isabella Even-Chen, BA1, ilan behm, MPH2, Rahul Das, PhD3, Allison Perry, PhD1.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA, 3Norstella, Yardley, PA, USA.
1Norstella, New York, NY, USA, 2Norstella, Englewood, CO, USA, 3Norstella, Yardley, PA, USA.
OBJECTIVES: To evaluate the extent thatstructured and unstructured clinical data identify patients with evidence of AUD beyond patients with diagnosis codes and AUD-specific treatment records.
METHODS: A cross-sectional analysis was conducted using NorstellaLinQ’s US real-world linked open claims, structured EHR, and clinical notes (June 2020-present). From 10,711,582 patients with diagnosed or undiagnosed AUD indicators in claims and EHR, 1,908,909 had EHR activity, 1,376,492 had at least one clinical note, and 403,771 had meaningful AUD-relevant note signals. These were stratified into confirmed AUD (N=318,270; ICD-10 F10.xx codes or AUD-specific pharmacotherapy), likely AUD (N=12,724; structured clinical indicators without formal diagnosis), and potential AUD (N=72,777; human-in-the-loop large language model (LLM) extracted notesignals or alcohol exposure).
RESULTS: Among 403,771 patients with clinically meaningful AUD note signals, 318,270 (79%) had confirmed AUD, while 85,501 were identified through likely or potential AUD indicators. AUD was documented in clinical notes for 86% of confirmed, 82% of likely, and 38% of potential AUD patients; among likely AUD patients, note documentation preceded formal diagnosis in 82% of cases. Across all cohorts, alcohol consumption was documented in 286,768 clinical notes, AUD-related treatment in 185,048, and care team interactions in 194,455. Formal AUD screening, however, was documented in only 51,855 patients, the smallest footprint across all note concepts, despite diagnosis documentation reaching 344,381 — indicating that clinicians document alcohol use and treatment far more consistently than they document structured screening.
CONCLUSIONS: Diagnosis codes alone may underestimate the identifiable AUD population, with clinical notes revealing an additional 85,501 patients including a pre-diagnosis cohort. Formal AUD screening is documented far less frequently than alcohol use or treatment in clinical notes, suggesting a systemic gap between clinical recognition of AUD and structured screening practice. These findings have direct implications for care pathway design, disease burden estimation, and HTA submissions requiring accurate AUD population sizing
METHODS: A cross-sectional analysis was conducted using NorstellaLinQ’s US real-world linked open claims, structured EHR, and clinical notes (June 2020-present). From 10,711,582 patients with diagnosed or undiagnosed AUD indicators in claims and EHR, 1,908,909 had EHR activity, 1,376,492 had at least one clinical note, and 403,771 had meaningful AUD-relevant note signals. These were stratified into confirmed AUD (N=318,270; ICD-10 F10.xx codes or AUD-specific pharmacotherapy), likely AUD (N=12,724; structured clinical indicators without formal diagnosis), and potential AUD (N=72,777; human-in-the-loop large language model (LLM) extracted notesignals or alcohol exposure).
RESULTS: Among 403,771 patients with clinically meaningful AUD note signals, 318,270 (79%) had confirmed AUD, while 85,501 were identified through likely or potential AUD indicators. AUD was documented in clinical notes for 86% of confirmed, 82% of likely, and 38% of potential AUD patients; among likely AUD patients, note documentation preceded formal diagnosis in 82% of cases. Across all cohorts, alcohol consumption was documented in 286,768 clinical notes, AUD-related treatment in 185,048, and care team interactions in 194,455. Formal AUD screening, however, was documented in only 51,855 patients, the smallest footprint across all note concepts, despite diagnosis documentation reaching 344,381 — indicating that clinicians document alcohol use and treatment far more consistently than they document structured screening.
CONCLUSIONS: Diagnosis codes alone may underestimate the identifiable AUD population, with clinical notes revealing an additional 85,501 patients including a pre-diagnosis cohort. Formal AUD screening is documented far less frequently than alcohol use or treatment in clinical notes, suggesting a systemic gap between clinical recognition of AUD and structured screening practice. These findings have direct implications for care pathway design, disease burden estimation, and HTA submissions requiring accurate AUD population sizing
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
RWD87
Topic
Clinical Outcomes, Epidemiology & Public Health, Real World Data & Information Systems
Topic Subcategory
Health & Insurance Records Systems
Disease
Mental Health (including addiction)