Evaluation of Polygenic Risk Score Prediction and Validation of Diagnostic Codes of Primary Open-Angle Glaucoma in Electronic Health Record (EHR)-linked Biobanks

Abstract

Purpose : A polygenic risk score (PRS) may improve risk stratification for primary open-angle glaucoma (POAG). In constructing a PRS, case definitions are often based on diagnostic codes which may introduce misclassification. To assess the utility of ICD-code-based PRS construction and future usage, we validated ICD codes of POAG status and tested PRS performance among ICD-code-based and manually validated cases/controls in two electronic health record (EHR)-linked biobanks.

Methods : We constructed the PRS using the Lassosum penalized regression framework from 10,238,000 directly genotyped and imputed variants for 449,186 UK Biobank samples (14,171 cases, 394,292 controls). The calculated weights were used to compute a POAG PRS in 4,824 Mount Sinai BioMe Biobank and 34,547 Mass General Brigham (MGB) Biobank participants. We reviewed records of ICD-identified POAG subjects in each biobank and confirmed cases based on chart notes, optical coherence tomograms, and visual fields (VF). In a 5% random sample, controls were verified based on intraocular pressure and cup-disc ratio. We compared the PRS area under curve (AUC) performance of manually reviewed vs. ICD-identified cases/controls.

Results : 485/610 and 545/889 patients from BioMe and MGB had ≥1 POAG ICD code and VF for chart review. Among cases manually reviewed in BioMe and MGB, the most common ancestral groups represented were Hispanic (54.7%) and European (80.5%), respectively. For BioMe and MGB, respectively: positive predictive value was 60% vs. 52%; negative predictive value was 95% vs. 96%; sensitivity 94% vs. 96%; and specificity 62% vs. 53%. When adjusted for age, sex, and genotyping array, AUC for association with POAG using manual review vs. ICD codes in BioMe (Fig 1) was, respectively: 0.77 vs. 0.74 for African, 0.82 vs. 0.76 for Hispanic, and 0.87 vs. 0.83 for European; for MGB: 0.82 vs. 0.82 for African, 0.88 vs. 0.90 for Hispanic, 0.74 vs. 0.77 for European. In BioMe, the manual AUC performed better than ICD-based coding in Hispanics (p=0.02); for MGB, the AUC between ICD-identified vs. manual review was similar (p≥0.16; Fig 1).

Conclusions : A PRS designed to identify POAG performed similarly using either manual clinical standard or ICD coding across most ancestries in two EHR-linked biobanks.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Adjusted AUC for association of POAG and PRS for manual review vs. ICD-based coding