Validation Of A Novel Online Tool For Individual Prediction Error Recalculation After Constant Optimization

To validate a novel, freely available online tool that provides an accurate and simplified method for determining the individual shift in prediction error (PE) for every patient following modification of an Intraocular Lens (IOL) formula constant. The central objective is to offer a validated resource that facilitates IOL constant optimization and research of IOL power formula comparisons studies, particularly for new-generation formulas which are unpublished.

Private practice, Buenos Aires, Argentina.

A retrospective, single-IOL model study was conducted using a dataset of 519 eyes . The expected postoperative refraction was calculated using six modern IOL formulas (Barrett Universal II, Cooke K6, EVO, Hill RBF, Kane, Pearl DGS, and Hoffer QST) under two conditions: using a standardized, initial A-constant and subsequently using the formula authors published optimized constant. The reference PE shift was established by comparing the expected refractions from a web-scraping process applied to the formula authors’ online calculators. The results from our novel online tool, which requires only the initial IOL constant, the implanted IOL power, and the initial patient prediction error were then compared against the web scraped results.

The tool demonstrated exceptional accuracy, with the absolute difference in prediction errors consistently measuring less than 0.025 diopters (D) across 100% of the entire patient cohort for all five formulas. Furthermore, the difference was less than 0.015 D in over 90% of cases for all formulas. Measures of central tendency for the error were minimal: the mean average error, mean prediction error, and mean absolute error were all less than 0.01 D. Subgroup analysis of high-power and low-power IOLs found no statistically significant compromise in the tool's accuracy..

The validated online tool offers a = precise, fast, and efficient method for calculating corrected prediction error following IOL constant adjustment. Its performance has no statistical differences to more elaborated methods such as web-scraping. By simplifying constant optimization and individual patient prediction error shift accordingly, this tool serves as a significant resource for researchers, facilitating IOL power calculation research and
contributing to more reliable comparisons of modern IOL formulas.