Accuracy of PEARL-DGS Formula for Intraocular Lens Power Calculation in Patients With Previous Myopic Laser Vision Correction.

Abstract

Purpose: To determine the accuracy of a new machine learning-based open-source IOL formula (PEARLS-DGS) in 100 patients who underwent uncomplicated cataract surgery and had a history of laser refractive surgery for myopic defects.

Methods: The setting for this retrospective study was HUMANITAS Research Hospital, Milan, Italy. Data from 100 patients with a history of photorefractive keratectomy or laser in situ keratomileusis were retrospectively analyzed to assess the accuracy of the formula. The primary outcome measures were absolute refractive prediction error, refractive prediction error, and cumulative distribution of absolute refractive prediction error within multiple thresholds. These parameters were estimated post-hoc using the Shammas, Haigis-L, Barrett True-K without history, ASCRS calculator average, EVO, Hoffer QST, and PEARL-DGS formulas. The cumulative distribution of the absolute refraction prediction error was analyzed and statistically tested.

Results: EVO 2.0 showed the lowest median absolute error (MedAE) of 0.36 diopters (D), followed by Hoffer QST (0.38 D) and PEARL-DGS (0.41 D). The cumulative distribution of the absolute refractive prediction error at ±0.50 D threshold showed the following ranking: Hoffer QST (0.65), PEARL-DGS (0.61), EVO 2.0 (0.60), Barrett-True-K (0.56), Haigis-L, ASCRS (0.52), and Shammas (0.45). A significant difference was recorded between Shammas and Hoffer QST only at this threshold (P < .05). Statistical differences could not be detected otherwise.

Conclusions: The new PEARL-DGS IOL formula demonstrated similar accuracy and comparability in median refractive prediction error to the other current formulas in eyes with a history of myopic laser vision correction. The cumulative distribution of refractive prediction error of the PEARLS-DGS performed well even compared to the Hoffer QST results.