Novel Bayes Statistics Approach For Post-Operative Intraocular Lens (Iol) Tilt Prediction

Published 2023 - 41st Congress of the ESCRS

Reference: PP12.05 | DOI: 10.82333/6mqk-wp34

Authors: Klemens Waser* ¹ , Andreas Honeder ² , Haidar Khalil ¹ , Leon Pomberger ¹ , Peter Laubichler ¹ , Matthias Bolz ¹ , Nino Hirnschall ¹

¹Ophthalmology ,Kepleruniversitätsklinikum Linz,Linz,Austria;Ophthalmology ,Johannes Kepler Universität Linz,Linz,Austria, ²Ophthalmology ,Kepleruniversitätsklinikum Linz,Linz,Austria;Ophthalmology,Johannes Kepler Universität Linz,Linz,Austria

Intraocular lens (IOL) tilt reduces visual performance, if it exceeds the amount of physiological tilt. Aim of this study was to use preoperative biometry data to predict the post-operative amount and orientation of tilt.

Department of Ophthalmology and Optometry, Johannes Kepler University, Linz, Austria

In total, 200 eyes scheduled for cataract surgery were included in this prospective study. Preoperatively, anterior segment imaging was performed using swept source optical coherence tomography measurements (IOL Master 700; Carl Zeiss Meditec AG, Germany and Casia 2; Tomey, Japan). These measurements were repeated 8 weeks after surgery together with autorefraction and subjective refraction. Random forest plot and Bayesian statistics were used to develop a tilt prediction model.

The difference of preoperative (crystalline) lens tilt and postoperative IOL tilt in amount, orientation and tilt vector was – 0.13° (± 0.97), 2.14 °(± 12.02) and 1.2° (±0.66) respectively. A high predictive power (variable importance for projection) for post-operative tilt prediction was found for pre-operative tilt (VIP=2.2), lens thickness (VIP=1.1), axial eye length (VIP=0.9) and decentration (VIP=0.9). The out of bag of a machine learning algorithm (random forest plot) was 0.916.

Postoperative IOL tilt impairs postoperative refractive outcome, especially in toric IOLs. Excellent prediction of post- operative IOL tilt amount and orientation using a machine learning algorithm (random forest). Pre- operative tilt of the crystalline lens, crystalline lens thickness, axial eye length and pre- operative decentration had been the variables with the highest predictive value.