Predicting Postoperative Anterior Chamber Depth, Intraocular Lens Tilt, and Decentration Using an Internally Validated Machine Learning Model.

INTRODUCTION: Precise intraocular lens (IOL) positioning is critical for optimal visual outcomes in cataract surgery, particularly with advanced IOLs. Misalignment can lead to refractive errors, astigmatism, and higher-order aberrations. This study aimed to predict postoperative anterior chamber depth (ACD), IOL tilt, and decentration using machine learning. METHODS: A prospective, single-center study was conducted at the Kepler University Clinic, Linz, Austria, involving 49 patients undergoing cataract surgery with implantation of a hydrophobic single-piece IOL (Clareon CNA0T0). Preoperative biometric data were collected using swept-source optical coherence tomography. Prediction models were developed using partial least squares regression to identify key predictive variables, followed by validation with a random forest model. Postoperative outcomes were assessed 8 weeks after surgery. RESULTS: The mean out-of-bag (OOB) error was 0.07 mm for ACD, 1.17° for IOL tilt, and 0.02 mm for IOL decentration. Key predictive variables for ACD included preoperative ACD, axial eye length (AL), and preoperative lens tilt. IOL tilt prediction was mainly influenced by preoperative lens tilt, thickness, and decentration. For IOL decentration, preoperative lens tilt, AL, and keratometry were significant predictors. CONCLUSIONS: The study demonstrated excellent predictability of postoperative ACD and IOL tilt and good predictability of IOL decentration using machine learning models. These findings support the clinical applicability of predictive modeling in optimizing IOL positioning for improved visual outcomes in cataract surgery. TRIAL REGISTRATION: ClinicalTrials. gov identifier, NCT06595693.