(P-286) Identification of a Diagnostically Complex Condition, Polycystic Ovarian Syndrome, in a Population-based Cohort Using Electronic Health Record Data

Background: Polycystic ovarian syndrome (PCOS) is a common reproductive endocrine disorder among women of child-bearing age. It is a complex condition that can be challenging to diagnose. ICD diagnosis codes (ICD-9: 256.4, ICD-10: E28.21) can be used to identify cases. Rotterdam criteria for diagnosing PCOS include at least 2 of 3 elements: oligo and amenorrhea, clinical and/or biochemical hyperandrogenism, and evidence of polycystic ovarian morphology (PCOM) on imaging. It is unknown whether women who are given a PCOS diagnosis code also fulfill the Rotterdam diagnostic criteria and vice versa.

Objectives: Compare proportion of cases identified by PCOS ICD 9/10 codes vs those identified by Rotterdam criteria using structured and unstructured electronic health record (EHR) data.

Methods: A cohort of women ages 12-60 years in 2006-2019 enrolled at Kaiser Permanente Washington was identified from EHR databases. PCOS ICD-9/10 diagnosis codes were collected as well as Rotterdam criteria data: 1) ICD-9/10 diagnosis codes for irregular bleeding; 2) indications of hyperandrogenism including ICD-9/10 codes for hirsutism, baldness and acne; elevated testosterone from laboratory data; medications from dispensing data; and searches of clinical notes via Natural Language Processing (NLP); and 3) PCOM on imaging identified by NLP. Rotterdam diagnostic algorithms will be created and incorporated 2 of the 3 elements, considering temporal relationships of the data and age. Proportions of PCOS cases identified by the 2 methods will be compared by chi-square tests to assess concordance overall and by ICD-9 vs ICD-10 era. As part of a large study to estimate PCOS incidence and prevalence over time, the analyses will incorporate chart review to validate incident case identification for the two PCOS cohorts.

Results: Among 371,495 women in 2006-2019, 5,981 (1.6%) received a PCOS ICD diagnosis. Distribution of PCOS diagnosis was between 5.4-9.3% over the 14 years. Half of these cases were diagnosed at ages 31-50 years, 40% at 18-30, 8% at 12-17 and 3% at 51-60. 27% women had a diagnosis of irregular bleeding and 0.06% had PCOM. Algorithms to define the Rotterdam cohort and concordance of the two methods to identify PCOS cases will be presented. We will also report confirmation rates of PCOS diagnosis from chart review.

Conclusions: To accurately identify patients with a diagnostically complex condition such as PCOS, it is feasible and necessary to utilize all the rich data available in the EHR, combining structured and unstructured elements to develop the most comprehensive diagnostic algorithm.