(P-617) Use of Natural Language Processed Clinical Notes in an Electronic Health Record (EHR) to Characterize Adverse Events of Mood Disorders

Background: Mood disorders (anxiety, depression) are comorbidities of endometriosis and are potential adverse events for some medications used to treat endometriosis-associated pain, due to their estrogen-reducing mechanism. Accurate characterization of mood disorders in these women may inform medical management options. Post-marketing surveillance of these events in administrative databases may be impacted by misclassification of disease onset and patient history, due to lack of data on early signs and symptoms.

Objectives: To characterize mood disorders identified by ICD diagnosis codes using additional information within physician notes in electronic health record (EHR) databases.

Methods: The analysis used data from Optum Market Clarity EHR (Jan 2007 to Mar 2019) and was restricted to women with endometriosis <50 years old, who had available provider notes and evidence of activity in the system over 5 years prior to the index date. The index date for the anxiety cohort (N = 46,195) and depression cohort (N = 35,264) was the date of first ICD diagnosis code for anxiety and depression, respectively. Early mentions were defined as the presence of ≥1 term of anxiety or depression without a negative sentiment from natural language processed (NLP) clinical notes within 5 years prior to the diagnosis code. A total of 38 mental health terms was selected for depression and anxiety.

Results: In the anxiety cohort, 59% of women had an early mention of anxiety in the notes, with a mean (SD) of 1.8 (2.4) years between first mention and ICD diagnosis code and 42% of women had ≥2 mentions of anxiety on separate days. In the depression cohort, 52% of women had an early mention of depression, with a mean (SD) of 1.6 (2.3) years between first mention and ICD diagnosis code and 35% had ≥2 mentions of depression on separate days. The most frequent terms in the provider notes for anxiety were anxiety (54%) and anxious(33%) and for depression were depression (48%) and depressed (15%). The length of time between first mention and ICD diagnosis code was highest in women >35 years old, African American race, and in the West region for both anxiety and depression.

Conclusions: Women with an ICD diagnosis code for mood disorders had mentions of the disease on average 1.7 years prior to the diagnosis code. Restricting post-marketing surveillance to only ICD codes for adverse events such as mental health may lead to misclassification of disease onset and/or patient history of the disease. Unstructured text from physician notes may provide additional important information to characterize the adverse event in real- world data sources.