Refractive Outcomes of Cataract Surgery Using Zeiss Artificial Intelligence Calculator versus Barrett Universal II and Kane

Purpose: To compare the performance of the Zeiss AI IOL Calculator (ZAIC) to that of Barrett Universal II (BUII) and Kane, using standard (K) and total keratometry (TK) as inputs. Patients and Methods: This was a multicenter retrospective case series which included surgery-naïve eyes undergoing cataract surgery with preoperative optical biometry on the IOLMaster 700, monofocal IOL (Bausch & Lomb enVista MX60E) implantation, and 20/40 acuity or better measured at least 3 weeks postoperatively. After taking the difference between actual and predicted postoperative spherical equivalent (SE) for each eye, optimized outcome measures included the mean absolute error (MAE), standard deviation (SD) of the mean prediction error, root mean squared absolute error (RMSAE), median prediction error (MedAE), and proportion of eyes within 0.5 D of predicted. Statistical differences were identified using Wilcox-Holladay-Wang-Koch tests with Holm correction. Results: A total of 246 eyes of 246 patients were analyzed. The MAE of ZAIC was significantly lower than that of BU2 TK [0.30 vs 0.32 D (p = 0.04)]. ZAIC had a lower SD and RMSAE than that of BU2 K and BU2 TK [0.38 vs 0.41 (p = 0.001) and 0.42 D (p = 0.002); 0.38 vs 0.41 (p = 0.008) and 0.42 (p = 0.008)]. ZAIC also had a larger proportion of eyes within 0.5 D of predicted compared to BU2 K [82% vs 76% (p = 0.02)], as well as a larger proportion of eyes within 0.75 D of predicted compared to BU2 TK [95% vs 91% (p = 0.02)]. There were no significant differences in MedAE or the proportion of eyes within 0.25 or 1 D of predicted. Conclusion: Compared to BU2, ZAIC produces a lower MAE, SD, and RMSAE, and a higher proportion of eyes within 0.5 and 0.75 D of predicted but performs similarly to Kane across all outcome measures. This remains true regardless of whether K or TK is used as an input. This study is limited by its retrospective design and use of a single MX60E IOL model. Plain Language Summary: This multi-institutional retrospective case series compares the prediction accuracy and precision of the Zeiss AI IOL Calculator (ZAIC) to the Barrett Universal II (BU2) and Kane intraocular lens power prediction formulas for cataract surgery. Results suggest ZAIC outperforms BU2 and performs at least as well as Kane. This holds true regardless of whether standard or total keratometry is used as formula inputs. This corroborates ZAIC as an accurate and precise tool for the preoperative selection of intraocular lens power, performing at least as well as other modern intraocular lens power calculation formulas. Keywords: cataract extraction, intraocular lens, artificial intelligence