Evaluation of an Algorithm for Identifying Ocular Conditions in Electronic Health Record Data

Area of Science:

• Ophthalmology
• Medical Informatics
• Big Data Analytics

Background:

• Accurate patient identification is crucial for big data research in ophthalmology.
• Administrative billing codes alone are insufficient for identifying patients with specific ocular diseases or phenotypes.
• Electronic Health Records (EHR) offer a rich data source but require sophisticated methods for accurate phenotyping.

Purpose of the Study:

• To develop and validate a novel algorithm for accurately identifying the presence or absence of ocular conditions using EHR data.
• To improve the efficiency and accuracy of patient cohort identification for ophthalmology research.

Main Methods:

• Retrospective analysis of EHR data from 122,339 patients over five years.
• Development of an algorithm to search structured and unstructured EHR data for exfoliation syndrome (XFS).
• Algorithm trained using logistic least absolute shrinkage and selection operator (LASSO) regression and validated by glaucoma specialists.

Main Results:

• The algorithm achieved a positive predictive value (PPV) of 95.0% and a negative predictive value (NPV) of 100% for identifying XFS.
• Billing codes alone captured XFS evidence in only 86% (ICD-9) or 96% (ICD-10) of cases where it was documented elsewhere.
• Clinical evidence of XFS was often under-documented in billing codes (approx. 40% for ICD-9, less for ICD-10).

Conclusions:

• The developed algorithm demonstrates superior accuracy in identifying XFS compared to relying solely on billing codes.
• This algorithm can significantly enhance the utility of EHR data for studying patients with ocular diseases.
• Improved phenotyping using EHR data facilitates large-scale epidemiological and clinical research in ophthalmology.