**Background:** The COVID-19 pandemic caused widespread disruptions to healthcare, with ophthalmology experiencing particularly severe reductions in care due to close-proximity exams, an older patient population, and many elective procedures. Previous studies examined broad categories or limited diagnoses, but a comprehensive analysis across many specific ophthalmic conditions was lacking. This study aimed to characterize utilization patterns for 261 granular ophthalmic diagnoses during the first two years of the pandemic in the US, using a common analytical framework to identify differential impacts and associations with disease severity.
**Methods:** Data were extracted from the American Academy of Ophthalmology IRIS Registry, encompassing 44.62 million unique patients and 2455 practices. Monthly patient counts for each of 261 diagnosis entities (grouped from ICD-10 codes) were analyzed from January 2017 to December 2021. Counterfactual predictions of expected utilization in the absence of the pandemic were generated using predictive models (generalized linear models with seasonality and trend terms) trained on pre-pandemic data (2017–2019) via leave-out-one-year cross-validation. The primary outcome was the deviation (δ) between observed and expected patient volumes, with 95% prediction intervals and empirical p-values. Average deviations were computed for the hiatus (March–May 2020) and post-hiatus (June 2020–December 2021) periods. Recovery was defined as three or more consecutive months with no significant negative deviation. Hierarchical clustering identified groups with similar deviation patterns. Associations with disease severity were tested using univariable linear regressions (based on severity rankings from Bourges et al.) and Wilcoxon rank-sum tests comparing vision-threatening vs. non-vision-threatening subtypes of AMD, diabetic retinopathy, and glaucoma.
**Key Results:** The mean out-of-sample RMSPE for predictive models was 10.3% (SD 7.3%). Across all 261 diagnoses, the mean deviation during the hiatus nadir (April 2020) was −0.67 (67% below expectation, SD 0.14), and during the post-hiatus period was −0.13 (13% below expectation, SD 0.09). 94.3% of diagnoses remained below pre-pandemic volumes post-hiatus. Less severe diagnostic categories (e.g., refractive error: hiatus δ = −0.89, post-hiatus δ = −0.14) experienced greater reductions than more severe categories (e.g., ocular globe injuries: hiatus δ = −0.49, post-hiatus δ = −0.08). Among 36 ocular emergencies, median deviations increased with severity: for each unit increase in severity ranking, deviations increased by 5.5% (p < 0.001) during hiatus and 1.9% (p = 0.04) post-hiatus. Non-vision-threatening subtypes of AMD, DR, and glaucoma showed greater reductions than vision-threatening subtypes during April 2020 (p = 0.06, 0.03, 0.01 respectively), but this difference was not significant post-hiatus. 33 conditions had intense post-hiatus reductions (δ ≤ −0.20, p ≤ 0.05), mostly non-vision-threatening (e.g., infectious keratoconjunctivitis: δ = −0.38, 95% CI −0.41 to −0.35, p < 0.001). Only 15 conditions (6%) met or exceeded counterfactual predictions post-hiatus, including retinopathy of prematurity stage 3 (δ = 0.12, 95% CI 0.06 to 0.18, p < 0.001) and unspecified diabetic retinopathy with DME (δ = 0.46, 95% CI 0.37 to 0.55, p < 0.001). 44% of conditions showed some recovery, but 57% of those were not sustained. The most common recovery month was June 2020 (36% of recovered conditions).
**Clinical Implications:** The study reveals widespread and lasting reductions in ophthalmic care utilization across nearly all diagnoses, with more severe conditions being relatively prioritized. Persistent declines in visits for leading causes of visual impairment (AMD, diabetic retinopathy, cataract, glaucoma) raise concerns about delayed diagnosis and treatment, potentially leading to increased irreversible vision loss. The inverse relationship between severity and underutilization suggests that patients and systems prioritized urgent care, but the lack of separation post-hiatus may reflect broader behavioral or economic barriers. Conditions with above-average utilization (e.g., ROP, ocular injuries) indicate continued attention to emergencies, while intense reductions in non-vision-threatening conditions (e.g., conjunctivitis) may be less clinically harmful. The analytical framework can be adapted to other specialties to monitor healthcare disruptions and identify unmet needs.