Trend Errors With Modern And Classical Iol Calculation Formulas

To evaluate and quantify trend errors in popular IOL formulas. A formula with low trend errors is suitable for any type of eye or IOL model independent of input variables – liberating the user from having to use multiple formulas for different ranges of input values. The concept of the Castrop formula and an upgraded version of Haigis (Haigis neo) is disclosed in detail.

Private eye clinic in Germany

Retrospectively, we analyzed prediction errors for four different IOL types (Hoya Vivinex n=888, B&L MX60 n=467, Alcon SN60AT n=822, J&J Tecnis n=613) and 17 different Gaussian IOL formulas and 1 ray-tracing software. This cohort is not only quite large but has also very good quality of postop refraction which highlights the formula properties. Besides the usual descriptive statistics like median, mean, standard deviation, we used squared prediction errors and polynomial regression to detect trend errors with the variables axial length (AL), corneal power (K), anterior chamber depth (ACD), lens thickness (LT) and IOL power (IOLP).

All classical formulas have significant trend errors with most input variables tested. Eg Haigis and SRK/T have pronounced trend errors with K, Hoffer Q and Holladay with AL, all of them with IOLP. Of the modern formulas, Castrop, EVO2, K6, Kane, Olsen and Pearl-DGS yield comparable results and have low trend errors. Same is true for Okulix raytracing. The popular Barrett Universal2 formula does have severe trend errors in very high IOL powers (≥ 28 D) but is otherwise quite neutral. K6 tends to be the best performer overall (standard deviation 0.361 D), it is statistically significantly better than all formulas except Castrop, EVO2, Kane, Haigis neo, Okulix and Olsen stand-alone.

The aforementioned modern formulas work with any type of eye without the necessity of selecting formulas for individual eyes. Results vary with IOL types, shape factor variability, asphericity and selection bias being the most probable reasons. Differences between the classics and the best performing modern formulas (standard deviation ≈ 0.44 D) are quite high. Unfortunately, most modern formulas are neither published nor disclosed and therefore very hard to comprehend.