Comparing the Use of Administrative Claims with the Electronic Health Record Data for Identifying Diabetic Retinopathy Examinations.

Abstract

Purpose: To characterize differences in identification of diabetic retinopathy examinations using administrative claims and electronic health record data.

Methods: Adult patients ≥18 years with diabetes seen in the ophthalmology department at Johns Hopkins Hospital were included. Two methodologies were used to identify diabetic retinopathy examinations across the hospital system. First, a pre-specified set of Current Procedural Terminology (CPT) codes from administrative claims data were used. Second, natural language processing (NLP) was used to parse ophthalmology provider notes for mention of diabetic retinopathy screening or follow-up. The percentage of visits meeting each set of criteria was determined. Cohen's kappa of agreement between the two methodologies was calculated.

Results: A total of 59,538 patients and 1,926,828 office visits, of which 485,228 (25%) were in the ophthalmology department, were included. Most patients (86%) had at least one diabetic retinopathy examination identified using administrative codes, and 84% using the NLP-based methodology. Of all ophthalmology visits, administrative codes identified more diabetic retinopathy examinations compared to the NLP-based methodology (60%, versus 48%). Cohen's kappa for agreement was 0.57 (standard error 0.001, p < 0.001).

Conclusion: This study found only moderate agreement between the two methodologies for identifying diabetic retinopathy examinations. Given the imprecision of administrative codes, this suggests that prior studies reporting eye care utilization using only administrative claims may be over-estimating receipt of diabetic retinopathy examinations.