Abstract
*Correspondence, Sanjay Patel: patel.sanjay@mayo.edu
Introduction We previously developed a model to predict improvement in central corneal thickness (CCT) after Descemet’s membrane endothelial keratoplasty (DMEK) for FECD from Scheimpflug images. The model incorporated parameters of pachymetry map isopach regularity and posterior corneal radius. In this study we assessed if adding corneal backscatter and pachymetric progression indices improved the predictive power of the existing model.
Methods The additional 37 parameters of interest were exported from the Scheimpflug camera software for images of eyes undergoing DMEK and were combined with all previous 180 parameters originally considered for the predictive model. Gradient boosting machine (GBM) models were used to determine the 5 parameters with highest relative influence. A regression model was derived from the 5 highest relative influencers and goodness-of-fit of predicted vs. observed improvement in CCT was assessed in derivation and validation groups.
Results Anterior and mid-corneal backscatter were high influencers along with isopach regularity parameters whereas pachymetric progression indices were not. After incorporating corneal backscatter, the predictive power (from R2) of the model in the derivation group was 79% (n=48). When the derivation model coefficients were applied to the validation group, the predictive power in the validation group was 72% (n=45).
Conclusions Combining anterior and mid-corneal backscatter with isopach regularity parameters creates a strong predictive model of CCT improvement after DMEK. However, the predictive power of this model did not improve the predictive power of the original model (derivation group, 80%; validation group, 78%). The predictive model could provide important ancillary test information to help inform clinical decision-making for FECD.
