Abstract
Purpose:
To develop and test an automated algorithm that classifies patients with age-related macular degeneration (AMD) as either wet or dry sub-types based on electronic medical records (EMR).
Methods:
We developed an automated EMR-based algorithm to identify patients as having either neovascular or non-neovascular AMD using billing codes and keywords suggested by clinicians (MAS, AAF). We tested the algorithm in a pilot study of 20 patients enrolled in the NUGene project, an EMR-linked DNA biorepository developed by Northwestern University Center for Genetic Medicine, and seen at NU Ophthalmology from 2005 to 2013. The algorithm identified 10 patients as having dry and 10 patients as having wet AMD. A trained grader (MAS) then independently reviewed the chart notes and fundus drawings and verified the diagnosis.
Results:
Of the 20 charts reviewed, 10% (2/20) of patients were under the age of 60 and 90% (18/20) were over the age of 60. Overall, 45% (9/20) of patients were correctly classified by the algorithm as either dry or wet AMD. Excluding patients under the age of 60, accuracy improved to 50% (9/18). Overall, 56% and 44% of patients were correctly classified by the algorithm in the dry and wet AMD categories, respectively. In the dry AMD category, misclassified cases included a case of pattern dystrophy, a case of diabetic retinopathy and a case of central serous chorioretinopathy. In the wet AMD category, misclassified cases included a case of proliferative diabetic retinopathy, a case of high myopia with lacquer crack, a case of idiopathic macular scar, a case of end-stage retinopathy of prematurity, and a case of birdshot chorioretinopathy.
Conclusions:
With the increasing availability of EMRs, the ability to identify and automatically classify patients with AMD using an EMR-based algorithm can be of benefit in genotype-phenotype studies and for targeted therapeutic interventions. We have found that using billing codes alone is not sufficient, and that use of critically selected keywords is required to achieve a more rigorous algorithm. We are modifying the algorithm using this preliminary data with a goal of creating an unsupervised algorithm that can accurately identify AMD subjects with more than 80% specificity. We have identified 182 AMD patients of whom 108 are already genotyped by NUGene whom we plan to study.
Keywords: 412 age-related macular degeneration •
453 choroid: neovascularization •
539 genetics