Algorithms for the identification of ophthalmic diseases in medico-administrative databases: A systematic review

Medico-administrative databases (MADs) are increasingly used in comparative effectiveness research.

To conduct a systematic review of algorithms used for the identification of key ophthalmic diseases in MADs: age-related macular degeneration (AMD), diabetic retinopathy (DR)/ diabetic macular edema (DME), glaucoma, cataract and uveitis.

We searched PubMed between June 30, 2016, and August 6, 2024, using keywords related to MADs and the ophthalmic diseases of interest. Two reviewers independently selected studies and extracted data.

From the 1719 references identified, 315 were selected describing a total of 523 algorithms. Approximately half of the study objectives were related to the identification of factors associated with the onset of ophthalmic diseases, exacerbation, or hospitalization (48%, n = 151). Only 2% (n = 6) focused exclusively on the development and/or validation of algorithms. Most studies were from Taiwan (36%, n = 113), Korea (27%, n = 84) and the United States (20%, n = 64). From the 523 algorithms, a validation was mentioned for 47 (9%). After regrouping close algorithms, 433 different algorithms were identified concerning glaucoma (34%, n = 146), DR/DME (26%, n = 114), AMD (18%, n = 80), cataract (11%, n = 47) and uveitis (11%, n = 46). About half of these algorithms used diagnosis codes only (58%, n = 251), while others combined diagnosis codes and procedures (16%, n = 69), or diagnosis codes and drugs (12%, n = 50).

This systematic review showed heterogeneity between algorithms used to identify the same pathology, raising the question of which ones are more appropriate to use in a particular context. Moreover, most algorithms were not validated despite the potential impact on study results.