Comparable performance of 3D and 2D anterior segment optical coherence tomography in predicting intraocular pressure...

To develop predictive models using three-dimensional (3D) and conventional two-dimensional (2D) anterior-segment optical coherence tomography (AS-OCT) parameters for intraocular pressure (IOP) reduction following phacoemulsification in glaucoma and non-glaucoma cohorts.

This prospective study included patients with and without glaucoma who underwent phacoemulsification. Preoperative predictors of 1-month IOP reduction, including clinical, ocular biometry, and AS-OCT (CASIA2) parameters, were analyzed. 3D AS-OCT measurements were assessed in two approaches: (1) averaging 360-degree values and (2) estimating circumferential areas or volumes. 2D variables were obtained from horizontal cross-sectional images. Model selection with least absolute shrinkage and selection operator (LASSO) for 2D and 3D variables was performed separately. R-squared (R²) used to represent model accuracy. The performance was validated by bootstrap resampling.

A total of 130 eyes (64 glaucoma, 66 non-glaucoma) from 103 patients were included. The average IOP change was -1.20 ± 3.29 mmHg. Preoperative IOP (preIOP) was the strongest single predictor across all models. In the entire cohort, the final models showed moderate predictability (R² 38% for 3D and 36% for 2D; optimism-corrected R² 31% for both). In glaucoma eyes, models incorporating AS-OCT angle status and anterior chamber width achieved R² values of 36% (3D) and 38% (2D), with optimism-corrected R² of 27% (3D) and 28% (2D). In non-glaucoma eyes, both 3D and 2D models showed higher predictability (R² 53%, optimism-corrected R² 45%), with final predictors incorporating AS-OCT anterior chamber area and iris thickness. Both 3D and 2D models significantly outperformed the reference preIOP models across all cohorts (all p > 0.05).

AS-OCT parameters improved the predictability of IOP reduction after phacoemulsification, with notably better performance in non-glaucoma eyes. 2D and 3D models showed comparable predictive ability.