PURPOSE: To quantify the current accuracy limits, analyze the residual errors, and propose the next steps for prediction accuracy improvements. SETTING: Eye hospitals in Germany, Denmark, and Austria. METHOD: Numerical ray tracing using manufacturer's intraocular lens (IOL) data (vertex radii, central thickness, refractive index) was used for all calculations. Postoperative lens position was predicted by a simple scaling model based on measurements in 1 patient collective. The model was compared with 2 other approaches in 2 patient collectives at 2 hospitals (1121 eyes with 13 IOL models; 936 eyes with 2 models). Axial lengths were measured optically (IOLMaster, Zeiss). No parameter adjustments or individualization of IOL types or of surgeons/localizations were done. The prediction errors and measures of systematic bias for short or long eyes were used to quantify the outcome. RESULTS: The mean prediction errors in the 2 collectives were +0.13 diopter (D) and -0.13 D and the mean absolute errors were 0.44 D and 0.50 D without bias for long or short eyes, but depending on the IOL position model approach. The differences in the mean prediction errors for the IOL types were below the allowed manufacturing tolerances and below human recognition thresholds. CONCLUSIONS: The need to individualize and fudge parameters decreases with better physical models of the pseudophakic eye. Further improvements are possible by individual topography to extract corneal asphericity and measured pupil size to calculate the best focus, by improved position predictions based on individual measurements of the crystalline lens and by smaller tolerances for IOL manufacturing.
PURPOSE: To quantify the current accuracy limits, analyze the residual errors, and propose the next steps for prediction accuracy improvements. SETTING: Eye hospitals in Germany, Denmark, and Austria. METHOD: Numerical ray tracing using manufacturer's intraocular lens (IOL) data (vertex radii, central thickness, refractive index) was used for all calculations. Postoperative lens position was predicted by a simple scaling model based on measurements in 1 patient collective. The model was compared with 2 other approaches in 2 patient collectives at 2 hospitals (1121 eyes with 13 IOL models; 936 eyes with 2 models). Axial lengths were measured optically (IOLMaster, Zeiss). No parameter adjustments or individualization of IOL types or of surgeons/localizations were done. The prediction errors and measures of systematic bias for short or long eyes were used to quantify the outcome. RESULTS: The mean prediction errors in the 2 collectives were +0.13 diopter (D) and -0.13 D and the mean absolute errors were 0.44 D and 0.50 D without bias for long or short eyes, but depending on the IOL position model approach. The differences in the mean prediction errors for the IOL types were below the allowed manufacturing tolerances and below human recognition thresholds. CONCLUSIONS: The need to individualize and fudge parameters decreases with better physical models of the pseudophakic eye. Further improvements are possible by individual topography to extract corneal asphericity and measured pupil size to calculate the best focus, by improved position predictions based on individual measurements of the crystalline lens and by smaller tolerances for IOL manufacturing.