Phenotype analysis of early risk factors from electronic medical records improves image-derived diagnostic classifiers for optic nerve pathology

Abstract

We examine imaging and electronic medical records (EMR) of 588 subjects over five major disease groups that affect optic nerve function. An objective evaluation of the role of imaging and EMR data in diagnosis of these conditions would improve understanding of these diseases and help in early intervention. We developed an automated image-processing pipeline that identifies the orbital structures within the human eyes from computed tomography (CT) scans, calculates structural size, and performs volume measurements. We customized the EMR-based phenome-wide association study (PheWAS) to derive diagnostic EMR phenotypes that occur at least two years prior to the onset of the conditions of interest from a separate cohort of 28,411 ophthalmology patients. We used random forest classifiers to evaluate the predictive power of image-derived markers, EMR phenotypes, and clinical visual assessments in identifying disease cohorts from a control group of 763 patients without optic nerve disease. Image-derived markers showed more predictive power than clinical visual assessments or EMR phenotypes. However, the addition of EMR phenotypes to the imaging markers improves the classification accuracy against controls: the AUC improved from 0.67 to 0.88 for glaucoma, 0.73 to 0.78 for intrinsic optic nerve disease, 0.72 to 0.76 for optic nerve edema, 0.72 to 0.77 for orbital inflammation, and 0.81 to 0.85 for thyroid eye disease. This study illustrates the importance of diagnostic context for interpretation of image-derived markers and the proposed PheWAS technique provides a flexible approach for learning salient features of patient history and incorporating these data into traditional machine learning analyses.

Applications of Artificial Intelligence to Electronic Health Record Data in Ophthalmology

“…Two studies in this review focused on the field of glaucoma and used supervised machine learning techniques to improve diagnosis and predict progression. 16 , 22 In the study by Chaganti et al., 16 a good performance was obtained (AUC of glaucoma diagnosis 88%), and results showed that the addition of an EMR phenotype could improve the classification accuracy of a random forest classifier with imaging biomarkers. 16 On the other hand, Baxter et al.…”

“… 16 , 22 In the study by Chaganti et al., 16 a good performance was obtained (AUC of glaucoma diagnosis 88%), and results showed that the addition of an EMR phenotype could improve the classification accuracy of a random forest classifier with imaging biomarkers. 16 On the other hand, Baxter et al. 22 reported a moderate performance (AUC 67%) in a study that used EHR data alone to predict risk of progression to surgical intervention in patients with open-angle glaucoma.…”

“… 20 to predict myopia onset and by Chaganti et al. 16 to improve the diagnostic accuracy of glaucoma. In addition, Baxter et al.…”

“…The work performed by Chaganti et al. 16 began to explore this problem by comparing the performance of classifiers using EMR phenotypes, visual disability scores, and imaging metrics.…”

“…14 Likewise in ophthalmology, machine learning classifiers with EHR data have been used to predict risks of cataract surgery complications, improve diagnosis of glaucoma and age-related macular degeneration (AMD), and perform risk assessment of diabetic retinopathy (DR). [15][16][17][18] Although the application of AI to EHR data related to ocular diseases has increased during the past decade, there have been no published reviews of this literature. One literature review of machine learning techniques applied in ophthalmology was published in 2017 19 ; however, the included studies mainly focused on the application of machine learning techniques to imaging data, rather than EHR data.…”