• Algorithm to Identify Ocular Conditions From EHR Data

    By Lynda Seminara
    Selected and Reviewed By: Neil M. Bressler, MD, and Deputy Editors

    Journal Highlights

    JAMA Ophthalmology, May 2019

    Download PDF

    For “big data” research, investigators are tasked with identifying many patients with a disease or phenotype of interest. Often this is accomplished by relying on administrative billing codes alone. Stein et al. set out to devise a method to identify the presence or absence of specific ocular conditions using data from electronic health records (EHR). They developed, tested, and validated an algorithm to determine the pres­ence/absence of exfoliation syndrome (XFS). Their approach proved superior to using billing codes alone.

    This retrospective analysis involved EHR data for 122,339 patients in the Sight Outcomes Research Collabora­tive Ophthalmology Data Repository who received eye care at participating centers from August 2012 through Au­gust 2017. The researchers developed a comprehensive algorithm that searches structured and unstructured (free text) EHR data for conditions of interest. They then tested its ability to detect the presence or absence of XFS among a sample of patients with and without XFS (n = 200) by reviewing ICD-9/ICD-10 billing codes, the patient’s problem list, and text within the ocular exam section and the unstructured (free-text) section of the EHR.

    The likelihood of XFS was estimated for each patient using logistic least ab­solute shrinkage and selection operator regression. The EHR data of all patients were run through the algorithm to gen­erate an XFS probability score for each patient, and the algorithm was validat­ed through EHR review by glaucoma specialists. The positive predictive value (PPV) and negative predictive value (NPV) of the algorithm were computed as the proportion of patients classified correctly as having or not having XFS.

    The algorithm assigned XFS probability of less than 10% to 99% of patients (n = 121,085), probability of greater than 75% to 0.4% (n = 543), probability of greater than 90% to 0.3% (n = 353), and probability of greater than 99% to 0.07% (n = 83). According to the analysis by glaucoma specialists, the algorithm’s PPV was 95% and NPV was 100%. When there was an ICD-9 or ICD-10 billing code for XFS, there also was XFS evidence elsewhere in the EHR in 86% or 96% of records, respectively. However, with clinical or free-text evidence of XFS, coexistence of ICD-9 codes was less common (~40%), and ICD-10 codes were even more scant (~20%). (Also see related commentary by Kurt K. Benke, PhD, in the same issue.)

    The original article can be found here.