Purpose: To predict the postoperative IOL position and refraction as accurately as possible independent of individualization of the parameters.
Setting: Universitats-Augenklinik, Mainz, Germany, and Vienna, Austria.
Methods: One patient cohort (189 eyes, Vienna) was used to calibrate the prediction method, which was then applied to a second cohort (65 eyes, Mainz). All calculations were based on consistent numerical ray tracing of the pseudophakic eye using the original manufacturer's intraocular lens (IOL) data (radii, thickness, refractive index). A new algorithm to predict IOL position was developed. Ultrasound (US) axial lengths were calibrated relative to partial coherence interferometry (PCI). Corneal radii extracted from topography were checked against radii measured with the IOLMaster (Zeiss) and by Littmann keratometry.
Results: Zero mean prediction errors for IOL position and refraction were obtained without adjusting the parameters and with PCI lengths or US lengths calibrated relative to the PCI values. There was no significant loss of accuracy of US data compared to PCI data. Corneal radii extracted from topography were slightly but statistically significantly different from the Littmann values, and they were more accurate than the latter with respect to prediction error. The measured mean central IOL position (distance from posterior corneal surface) for all IOL types was 4.580 mm, a value very close to the mean recalculated from A-constants (4.587 mm). The difference in the individual central IOL position relative to the mean value depended only linearly (ie, no higher orders such as square or cubic are needed) on axial length, with the mean central IOL position as a free parameter. This parameter should be 4.6 +/- 0.2 mm (the same value as independently measured or recalculated) to obtain zero steepness of the prediction error as a function of axial length, producing zero bias for long and short eyes.
Conclusions: Calculation errors from formulas and confusing adjusting parameters can be avoided if calculations and measurements are performed on a clear and simple physical basis. Nevertheless, an individual prediction error, typically 0.5 to 1.0 diopter, seems to be unavoidable.