Keratoconus Progression To Collagen Cross-Linking: Creating Personalised Predictions At Time Of Presentation

Published 2022 - 40th Congress of the ESCRS

Reference: FPM10.10 | Type: Free paper | DOI: 10.82333/b6vz-vq39

Authors: Ji-Peng Olivia Li* ¹ , Howard Maile ² , Mary Fortune ³ , Patrick Royston ⁴ , Marcello Leucci ¹ , Bruce Allan ¹ , Stephen Tuft ¹ , Nikolas Pontikos ² , Daniel Gore ¹

¹Cornea and External Disease,Moorfields Eye Hospital,London,United Kingdom, ²Institute of Ophthalmology,UCL,London,United Kingdom, ³Department of Public Health and Primary Care,University of Cambridge,Cambridge,United Kingdom, ⁴Department of Statistical Science,UCL,London,United Kingdom

Purpose

To develop a prognostic model which can be used in the clinical setting to inform patients and clinicians of an individual’s probability of progression to corneal collagen cross-linking over time. Given the small risk of complications from crosslinking, clinicians usually prefer to demonstrate progression before initiating treatment in early, low risk patients. There is a need to provide patients and clinicians with an predictor of progression to allow for timely treatments without waiting to demonstrate progression. To address this unmet need, we set out to create an algorithm that generates a projected progression curve demonstrating the likelihood of progression of needing crosslinking based on data collected at presentation.

Setting

This was a retrospective cohort study of 9341 eyes from 5025 patients with suspected or confirmed keratoconus from Moorfields Eye Hospital, London.

Methods

3541 eyes received collage crosslinking (CXL). The Royston-Parmar method on the proportional hazards scale was used to generate a prognostic model. Hazard Ratios (HR) for each significant covariate with explained variation and discrimination were calculated for the final model. Internal-external cross validation was performed by splitting the dataset into patient residential regions defined by their postcode and measuring the consistency in discrimination across different regions.

Results

The final dataset used for model fitting comprised 8701 eyes from 4823 patients of which 3232 eyes underwent CXL. Our model explained 33% of the variation in time-to-event with four significant covariates: age HR [95% confidence limits] 0.9 [0.90-0.91], Kmax 1.08 71 [1.07-1.09], Front K1 0.93 [0.91-0.94], and minimum corneal thickness 0.95 [0.93-0.96]. When performing internal-external cross validation, the predicted time-to-event curves generally followed the observed time-to-event curves and differences in discrimination between regions was low, suggesting the model maintained its predictive ability.

Conclusions

We have fitted a prognostic model for progression of keratoconus to CXL that could be used to aid clinical decision making. Age of presentation contributed most of the explained variation. We have identified key variables that contribute to the explained variation. Furthermore, and quite uniquely to our model, genomic risk loci associated with keratoconus for 926 patients were included and their contribution to the explained variation was assessed. We have also built a web application in order to demonstrate our predictive model in practice: https://pontikoslab.com/kcprog .