Senior Scientist Optum Epidemiology Boston, United States
Background: Ocular surface disorders (OSD), particularly conjunctivitis and keratitis, have been reported among patients with atopic dermatitis (AD). Nonetheless, validation studies of claims-based algorithms for conjunctivitis and keratitis in the AD population are lacking.
Objectives: To develop and validate claims-based algorithms for conjunctivitis and keratitis among patients with AD using medical records.
Methods: Within the Optum Research Database, a source population of patients with AD was identified between March 2017 and November 2019. Candidate algorithms to identify conjunctivitis and keratitis were developed within the AD source population. The algorithms included International Classification of Diseases, 10th Revision (ICD-10) diagnosis codes indicative of conjunctivitis or keratitis, alone and in combination with medications for treatment of these conditions. Among patients who met ≥ 1 candidate algorithms for conjunctivitis or keratitis, a subset was randomly selected for medical record review. Overall, 201 medical records were sought and 162 were adjudicated by an expert ophthalmologist, including 117 from patients with potential conjunctivitis and 45 from patients with potential keratitis. The positive predictive value (PPV) of each algorithm was calculated and compared to a pre-specified threshold of ≥ 70%.
Results: For conjunctivitis, the PPV of the algorithm that required ≥ 1 conjunctivitis diagnosis and no conjunctivitis medications (PPV=80%) was similar to the PPV of the algorithms that required a conjunctivitis diagnosis and medication dispensing (PPVs ranging from 77-87%). As such, the final algorithm was simplified to ≥ 1 conjunctivitis diagnosis, for which the PPV was 81% (95% confidence interval [CI]: 73% - 87%). For keratitis, the algorithm that required a keratitis diagnosis and ≥ 1 dispensing of an ophthalmic antibiotic or topical antibiotic-steroid combination had the highest PPV (PPV=91%). The next best performing algorithm, which required ≥ 1 keratitis and ≥ 1 dispensing of an ophthalmic corticosteroid, topical immune-modulator, or topical NSAID nearly met the pre-specified PPV threshold (PPV=68%). As the performance of the other keratitis algorithms was poor, the final algorithm was a combination of these 2 algorithms and had a PPV of 80% (95% CI: 55% - 93%).
Conclusions: The first well-performing claims-based algorithms for identifying conjunctivitis and keratitis among patients with AD were developed and validated using expert ophthalmologist review of 162 medical records. These algorithms are available for use in future studies, in particular to better understand conjunctivitis and keratitis occurrence among patients receiving treatments for AD.
