Abstract
Purpose :
Electronic health record (EHR) data offers valuable insights into the epidemiology, demographics, and disease course of normal tension glaucoma (NTG). In our prior study, 14.6% of glaucoma patients were identified as having NTG using diagnosis codes. Computable phenotypes (CP) can improve NTG patient identification beyond the limitations of diagnosis codes alone. This study assesses the effectiveness of three phenotype algorithms in detecting NTG.
Methods :
Using EHR data of glaucoma patients at the Oregon Health & Science University, a random sample of 100 patients underwent chart review. Three methods for NTG identification were compared: 1) ICD-10 diagnosis code, 2) the presence of “low tension glaucoma,” “normal tension glaucoma,” “LTG,” or “NTG” in chart notes, and 3) maximum intraocular pressure (IOP) ≤21 mmHg (Tmax) in chart notes or recorded in EHR with no glaucoma medications at the first visit. Data analysis was performed using R and manual review to derive meaningful conclusions.
Results :
NTG diagnosis codes, NTG-related terms in chart notes, and Tmax ≤21 mmHg with no glaucoma medications at the first visit identified 14, 18, and 34 NTG cases, respectively. Six NTG cases were identified using all three methods, 10 cases using two methods, and 28 cases using one method. Figure 1 depicts the overlap of the methods in identifying NTG cases. Of patients with NTG diagnosis codes, 35.7% had at least one eye with IOP exceeding 21 mmHg. Tmax was not documented in 40% of notes. 8.4% of maximum IOP recorded in chart notes used alternative phrases other than “Tmax.”
Conclusions :
A phenotype algorithm using objective data, such as Tmax and medications, identified more NTG cases than diagnosis codes or NTG-related terms in chart notes alone. Due to discrepancies between diagnosis codes and IOP values, usage of all three methods to identify NTG cases has low yield. Our claim regarding the limitations of diagnosis codes is further supported by the observation that some patients with NTG diagnosis codes did not meet criteria for NTG diagnosis. Additionally, Tmax is inconsistently documented. Incorporating “Tmax” as a structured field may contribute to more comprehensive IOP data within EHR. This study lays the groundwork for future automated CP algorithms that extract objective data to accurately and efficiently identify NTG patients.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.