Comparing prediction accuracy between total keratometry and conventional keratometry in cataract surgery with refractive multifocal intraocular lens implantation

We aimed to compare refractive outcomes between total keratometry using a swept-source optical biometer and conventional keratometry in cataract surgery with refractive multifocal intraocular lens (IOL) implantation. We included patients who underwent cataract surgery with refractive multifocal IOL implantation. The IOL power was calculated using conventional formulas (Haigis, SRK/T, Holladay 2, and Barrett Universal II) as well as a new formula (Barrett TK Universal II). The refractive mean error, mean absolute error, and median absolute error were compared, as were the proportions of eyes within ± 0.25 diopters (D), ± 0.50 D, and ± 1.00 D of prediction error. In total 543 eyes of 543 patients, the absolute prediction error of total keratometry was significantly higher than that of conventional keratometry using the SRK/T (P = 0.034) and Barrett Universal II (P = 0.003). The proportion of eyes within ± 0.50 D of the prediction error using the SRK/T and Barrett Universal II was also significantly higher when using conventional keratometry than total keratometry (P = 0.010 for SRK/T and P = 0.005 for Barrett Universal II). Prediction accuracy of conventional keratometry was higher than that of total keratometry in cataract surgery with refractive multifocal IOL implantation.


Discussion
In the present study, we demonstrated that the prediction accuracy of conventional K was higher than that of TK in cataract surgery with refractive multifocal IOL implantation. When using the SRK/T and Barrett Universal II formulas, K had a significantly higher prediction accuracy than TK. Previous studies on the clinical usefulness of TK have used several kinds of monofocal or toric IOLs. In the present study, we focused on the use of refractive multifocal IOLs to investigate the accuracy of TK in IOL power prediction. In addition, one strength of the study was its large sample size.
With all formulas except the Haigis formula, MAE and MedAE were higher when using TK than when using K, and the difference was significant in the SRK/T and Barrett Universal II formulas. When comparing the proportion of eyes within ± 0.50 D of the prediction error, K showed a significantly higher proportion than TK in the SRK/T and Barrett Universal II formulas. Previous studies have reported that, when the Barrett Universal II and SRK/T formulas were applied to conventional K, the MAE and MedAE were higher than when those formulas were applied to TK 2,3 . The present study contradicts this finding. It may be that the SRK/T formula showed more accurate results when using conventional K because it was developed using that measurement. However, the  Table 3. Proportion of eyes within ± 0.25 D, ± 0.50 D, and ± 1.00 D of spherical equivalent prediction error. D diopters, K conventional keratometry, TK total keratometry. *Statistically significant difference between the K and TK.      3 . In the present study, we found values of 82.5% and 89.9%, respectively when applying K to the Haigis and Barrett Universal II formulas, and 82.1% and 86.6% when the TK was applied. Compared to the two studies mentioned, when using TK, the prediction accuracy was similar or higher, while when using K, the accuracy was much higher. Therefore, the unexpected results of the present study likely resulted from higher accuracy when using K rather than from low accuracy when using TK. The patients included in the present study underwent multifocal IOL implantation to correct presbyopia. As such, they were younger and better preoperative best corrected visual acuity (BCVA) than the patients in other IOL power calculation studies. Therefore, it may be that we could make more accurate biometric measurements and that the IOL power calculation had greater accuracy.
The Haigis formula was the only one that showed lower MAE when using TK than when using K. In a previous study, Haigis using TK exhibited higher accuracy than Haigis using K 2 . When estimating the cumulative percentage of eyes, Haigis K showed 77% within 0.50 D and 95% within 1.00 D, while Haigis TK showed 79% within 0.50 D and 96% within 1.00 D. All of these proportions were higher in the present study than in the previous study 2 .
When comparing the APE between formulas, not all comparisons were significant, but the overall comparison was significant (Fig. 3). In a recently published study on TK, when using both K and TK, MAE and MedAE was lowest when applying Barrett Universal II, followed by SRK/T, Holladay 2, and Haigis 3 . Our study found similar results, although the order between SRK/T and Holladay 2 was reversed, and the difference was not significant using either K or TK. As mentioned above, the Barrett TK Universal II formula showed significantly inferior results than the Barrett Universal II using K, but it did show superior results compared to other formulas using K, perhaps because the Barrett TK Universal II was developed based on TK. Conversely, when the previously used formulas SRK/T, Holladay 2, and Haigis used TK, the result was inferior to all formulas using K except Haigis.
A review article on IOL power calculation mentioned that the Haigis and Holladay 2 formulas have higher accuracy than the Barrett Universal II formula in short eyes, and that the Barrett Universal II, Haigis, Olsen, and SRK/T formulas have greater accuracy in long eyes 9 . In the AXL subgroup analyses of the present study, Barrett formulas resulted in the lowest MAE, regardless of the AXL-Barrett TK Universal II in short eyes, Barrett Universal II in medium and long eyes, while Haigis formulas conferred the highest MAE-Haigis K in short and medium eyes, Haigis TK in long eyes. Both Haigis and Holladay 2, which were mentioned as high accuracy in the review article, showed high MAE in short eyes, and Haigis even showed high MAE in long eyes, whereas Barrett and SRK/T showed lower MAE in long eyes than in short eyes, showing high accuracy in long eyes. Although these results were somewhat contrary to those previously published, our results included only 11.0% of all subjects for short eyes and 3.87% for long eyes, so a larger number of patients should be analysed in a future study.
The present study was limited by its retrospective nature and because the average ME was not set to zero using the A constant as a User Group of the Laser Interference Biometry (ULIB). To allow more accurate IOL power prediction, an optimally personalized lens constant that can set the average ME to 0 should be applied to eliminate bias caused by the lens factor. Excluding this limitation, the present study followed all protocols suggested for studies into IOL formula accuracy proposed by Hoffer et al. 10 . The cataract surgery in the present study was performed by one expert surgeon using only one IOL model, one eye from each patient was randomly included, and postoperative refraction was measured 2 months after surgery.
In conclusion, when applying the formulas currently in use to TK during cataract surgery with refractive multifocal IOL implantation, the refractive results were not superior to those of K. The APE was higher when using TK than when using K, so more TK data must be gathered for each formula to allow more accurate IOL power calculation using TK.

Subject. This study was conducted with the approval of the Institutional Review Board of the Asan Medical
Center and University of Ulsan College of Medicine, Seoul, South Korea (Approval number: 2020-1290). The study adhered to the tenets of the Declaration of Helsinki and followed good clinical practice guidelines. This study was conducted retrospectively and the informed consent from the patients was waived.
In this retrospective study, all patients underwent cataract surgery with phacoemulsification and implantation of refractive multifocal IOL (Lentis M plus; Oculentis GmbH., Berlin, Germany). All procedures were performed by a single experienced surgeon. In previous studies, this refractive multifocal IOL has shown stable distant and near vision, as well as fewer photopic phenomena and better intermediate vision than diffractive multifocal IOLs 11,12 . The study included patients who underwent cataract surgery between January 2019 and December 2019, who underwent a complete preoperative examination, uncomplicated cataract surgery, and a follow-up of at least 2 months. Patients who had undergone previous intraocular surgery were excluded, as were those with a BCVA of less than 0.1 logMAR at 2 months after surgery. One eye was randomly selected to participate in the study. All subjects underwent a complete ophthalmologic examination before surgery, including BCVA and automated keratometry using the RK-F2 Full Auto Ref-Keratometer (Canon, Tokyo, Japan). The AXL, anterior chamber depth, K, and TK were measured using the IOLMaster 700. At least 2 months after surgery, patients underwent ophthalmologic examination, including BCVA, automated keratometry, and manifest refraction. www.nature.com/scientificreports/ were measured twice by an experienced technician to ensure accuracy. Patients were categorised into three groups according to the preoperative axial length: short (< 22.5 mm), medium (22.5-25.5 mm), and long (> 25.5 mm).
Calculating IOL prediction error. IOL power calculations were carried out on an IOLMaster 700 unit (Barrett Suite, software version 1.80.6.60340). We applied the TK and K of all patients to the currently widely used formulas (Haigis, SRK/T, Holladay 2). In the Barrett formula, K is applied to Barrett Universal II, and TK is applied to Barrett TK Universal II, a new formula based on TK. We used the lens constants of the User Group of the Laser Interference Biometry (ULIB) website. ME was defined as the mean value obtained by subtracting the preoperative predicted spherical equivalent from the actual spherical equivalent. APE was defined as the absolute value obtained by subtracting the actual spherical equivalent from the preoperative predicted spherical equivalent. MAE was defined as the mean APE value, while MedAE was defined as the median APE. The proportion of eyes within ± 0.25 D, ± 0.50 D, and ± 1.00 D of the prediction error were measured.

Statistical analyses.
A Wilk-Shapiro test was used to assess the distribution of numerical data. The Wilcoxon signed-rank test was used to compare the APE between K and TK. The APE was compared between the formulas using the Friedman test, with Bonferroni post-hoc correction for multiple comparisons. The proportions of eyes within ± 0.25 D, ± 0.50 D, and ± 1.00 D were compared using the McNemar's test. For each formula, APE differences between conventional K and TK were obtained, the differences were analysed using the Bland-Altman plot. All data were statistically analysed using SPSS software version 25.0 (IBM, Armonk, NY, USA), and all P-values less than 0.05 were considered statistically significant.

Data availability
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request.