Purchase this article with an account.
Sudha K Iyengar, Christopher W. Halladay, Tamer Hadi, Matthew D. Anger, Xuan-Mai Nguyen, Robert P. Igo, Paul B. Greenberg, Dana Crawford, Jack M Sullivan, Scott Damrauer, Wen-Chih Wu, Neal Peachey, Eric Konicki; Optimizing case-control classification for age-related macular degeneration in the VA electronic health record using a multi-algorithm data cube approach. Invest. Ophthalmol. Vis. Sci. 2018;59(9):1428.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The VA Million Veteran Program (MVP) consists of >600K participants, with >480K genotyped with a custom Affymetrix Axiom Biobank array. MVP provides access to array data in tandem with electronic health records (EHR). Age-related macular degeneration (AMD) is a complex oligogenic disorder leading to vision loss and blindness. As access to fundus photography, optical coherence tomography, or other gold-standard tools for AMD diagnosis is not yet facilitated MVP-wide, we utilized a multi-algorithm data cube approach together with individual validation at 3 sites to identify cases with AMD and controls without AMD.
To classify participants as cases or controls, we developed an initial algorithm utilizing structured (ICD-9-CM; ICD-10-CM; CPT) codes. We developed test sets of cases and controls across each of three VA medical centers (Cleveland, Providence, Buffalo) and refined the algorithm via detailed chart and imaging reviews by retinal specialists. We identified putative cases and controls based on age (cases ≥ 50 yrs; controls ≥ 65 yrs) and the presence of CPT codes for comprehensive eye exams (92004 and 92014) within the last 2 years. Cases had AMD codes on at least 2 clinic visits. Controls had no AMD codes. The original algorithm was then expanded to 8 algorithms (data cube) with varying definitions for cases and controls.
We reviewed charts from a total of 288 cases and 267 controls of European-American (EA) and African-American (AA) descent. The case-control algorithm performed at >90% with respect to accuracy, sensitivity, specificity and for both positive and negative predictive value. We generated a multi-algorithm phenotypic data cube and utilized p-values and odds ratios at two known loci, CFH and ARMS2/HTRA1, and q-q plots of genome-wide association, to enrich the contrast between cases and controls. A stricter case definition and expansion of controls to those without an eye visit provided 17,818 cases and 142,522 controls after quality control in the EA, and identified many AMD genes discovered recently.
AMD cases and controls can be identified with high accuracy from the VA EHR using diagnostic codes. This finding supports our ongoing efforts to use the MVP to understand the genetic basis of AMD.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
This PDF is available to Subscribers Only