Refractive Predictability Using the IOLMaster 700 and Artificial Intelligence-Based IOL Power Formulas Compared to Standard Formulas

PURPOSE: To investigate the accuracy of intraocutar tens (IOL) power calculation formulas using swept-source optical coherence tomography (SS-OCT). METHODS: Eyes with biometry measurement by IOLMaster 700 (Cart Zeiss Meditec AG). uncomplicated phacoemulsification, and IOL implantation were enrolled in this retrospective study. Newly released artificial intelligence-based formulas including Hill-Radial Basis Function (RBF) 2.0, Kane, and PEARL-DGS were compared with Gaussian optics-based standard formulas. The refraction predicted by each formula was compared with the actual refractive outcome in spherical equivalent. RESULTS: A total of 410 eyes of 410 patients were included in this study. Using optimized constants for SS-OCT biometry led to a significant decrease in median absolute error (MedAE) for Barrett, Haigis, and Hoffer Q formulas compared with using User Group for Laser Interference Biometry constants (P < .05). Overall, Olsen (0.283 diopters [D]) and Kane (0.286 D) formulas had significantly tower MedAEs than RBF 2.0 (0.314 D), Haigis (0.322 D), SRK/T (0.371 D), Holladay 1 (0.376 D), and Hoffer 0 (0.379 D) formulas under constant optimization (P < .05). The first four formulas with the lowest standard deviations of prediction error were Kane (0.451 DI, Olsen (0.456 D), EVO 2.0 (0.460 D), and Barrett (0.470 D). Olsen (47.1%), Barrett (45.9%1, Kane 145.4%1, and EVO 2.0 (45.1%) formulas had greater proportions of eyes within +/- 0.25 D of the predicted refraction than Hoffer Q (35.9%), SRK/T (35.9%), and Holladay 1 (33.4%) formulas (P < .05). CONCLUSIONS: Constant optimization for SS-OCT biometry further improves the performance of formulas. The most accurate prediction of postoperative refraction can be achieved with Barrett, EVO 2.0, Kane, and Olsen formulas.