ELP prediction boost
New algorithm using AS-OCT parameters shows potential for increasing accuracy
Ikko Iehisa MD
A new algorithm incorporating novel preoperative parameters obtained by anterior segment optical coherence tomography (AS-OCT) predicts postoperative effective lens position (ELP) more accurately than established methods, reported Ikko Iehisa MD at the XXXIV Congress of the ESCRS in Copenhagen, Denmark.
The new algorithm was developed and validated in a retrospective study using data from 60 eyes. The two parameters it uses to calculate postoperative ELP are angle-to-angle (ATA) depth and the sum of the preoperative anterior chamber depth plus one-half of the crystalline lens thickness (ACD + LT/2). The values for ATA depth, ACD and LT are all measured using a new Fourier domain AS-OCT device (CASIA2, Tomey).
“Fourier domain AS-OCT allows for accurate measurements of ocular anatomical parameters at one time. By improving the prediction of the postoperative ELP, this new algorithm may increase the accuracy of intraocular lens (IOL) power calculations,” said Dr Iehisa, National Hospital Organization, Tokyo Medical Center, Japan. The 60 eyes included in the study were all operated on by a single surgeon using a standard technique with insertion of an acrylic toric IOL (AcrySof® IQ Toric, Alcon) through a 2.2mm temporal clear corneal incision. Preoperative measurements also included corneal thickness (CT) determined with AS-OCT, as well as axial length (AL) and corneal curvature (K) obtained with an optical biometer (OA-2000, Tomey). AS-OCT imaging was repeated at one month after surgery to measure ATA depth, CT, ELP and IOL thickness.
Using AS-OCT, ATA depth is measured as the perpendicular distance between the posterior corneal surface and the intersection point of a line joining both angle recesses on the cross-sectional horizontal image with the corneal vertex, Dr Iehisa explained.
The 60 eyes were randomised into two groups of 30 eyes each representing a training set and a validation set. There were no significant differences between the training and validation sets of eyes for any ocular, IOL, or demographic characteristics.
The training set was used to evaluate the performance of various parameters for predicting postoperative ELP and to develop a new regression algorithm for ELP prediction based on multiple linear regression analysis. A simple linear regression analysis considering five parameters – ACD + LT/2, ATA depth, ACD, AL, and LT – showed that ACD + LT/2 had the highest correlation coefficient (R=0.913) followed by ATA depth (R=0.809).
The multiple linear regression analysis showed that the combination of ATA depth and ACD + LT/2 had the strongest correlation with postoperative ELP compared with other combinations tested (ATA depth, ACD; ATA depth, AL; ACD + LT/2, AL; and ACD, AL). The coefficient of determination (R2) was 0.856 for the combination of ATA depth with ACD + LT/2, and it ranged from 0.691 to 0.829 for the four other combinations of variables.
TESTING AND ACCURACY
The study assessed the performance of the new algorithm for predicting ELP in the validation set by determining the R2 between the measured and predicted postoperative ELP. The analyses showed that the new algorithm predicted the postoperative ELP with a higher R2 than a ray-tracing formula (OKULIX, Tomey) and two modern theoretical formulas – the SRK/T and Haigis (R2=0.830 vs 0.285, 0.379, and 0.510, respectively).
The validation set was also used to calculate the absolute ELP prediction error associated with each of the formulas. The new algorithm had a median absolute prediction error of 0.100mm, and that result was significantly smaller (P<0.001) than for the SRK/T (1.026mm), Haigis (0.336mm), and OKULIX (0.193mm).
In a previous study (Goto S, Ophthalmology. 2016;123(12):2474-2480), Dr Iehisa and colleagues reported that ATA depth was an effective parameter for predicting postoperative ACD. They had hypothesised that it would be useful because ATA depth remains unchanged after phacoemulsification and IOL implantation.
Ikko Iehisa: firstname.lastname@example.org