Intraocular Lens Calculation Myth Busters: A-Constants, Surgical Astigmatism, and P-Significance
Purpose
Are calculated A-Constants really constant? Is surgical induced astigmatism predictable? Should we use P-Significance testing in clinical studies?
Methods
Retrospective analysis of over 9000 non-selected, non-randomized postop cataract cases from 7 surgeons, using advanced graphics and statistical techniques - Bayesian Analysis - and new methods of Bayesian significance testing to replace null hypothesis significance testing (NHST - classical P-Significance) - Distributions of A-Constants were fit to normal and T-Distributions, and timed moving averages were graphed. Contour 2 variable graphs of surgical induced astigmatism (SIA) were generated. SIA was measured both strictly on the cornea by changing K values, and by net SIA, including whole eye optics. Bayesian analysis of results was done.
Results
A-Constants are not ‘constant’ but rather are 1) not normally distributed, but fit a T distribution with fat tails 2) vary with axial length, keratometry, and surgeon and 3) vary with time. Surgical induced astigmatism is essentially unpredictable, with bimodal 2.0 D concentrations on incision axis and 90 degrees away. The median is about 0.3 D for most surgeons. The mean is not calculable. NHST should be discarded because of sample size biases that occur with differing observer intentions. Bayesian significance testing relies on pure evaluation of the plausible distributions of data and not an artificial p-significance level.
Conclusion
Reliance on A-Constant based formulas should be tempered by the variance of A-Constant calculations. Continual optimization is necessary. Surgical induced astigmatism is unpredictable, but fortunately small in effect. P-Significance testing in clinical studies is outdated and flawed, and Bayesian Significance testing should replace it.