Accurate prediction of posterior corneal astigmatism in a multi-center study for predictive, preventive and personalised medicine

Aims: To predict posterior corneal astigmatism (PCA) magnitude and flat axis using other corneal parameters. Methods In this multicenter cross-sectional study, sex, age, eye laterality, and eight selected corneal parameters were input, and PCA magnitude and flat axis were output of the ridge regression (RR) models. Data from two centers were used for RR models construction and internal validation, and data from the other three centers were for external validation. Mean absolute error (MAE), root mean square error (RMSE) and area under the receiver operating characteristic curve (AUC) were used to evaluate the prediction accuracy. Results A total of 10448 subjects (17538 eyes) were included (5943 males and 4505 females, mean age: 24.84 ± 5.38 years). An RR equation was generated for prediction of PCA magnitude in all eyes ( R 2  = 0.501) with an MAE and an RMSE of 0.065-0.080D and 0.083-0.102D in external validation, respectively. Another RR equation was generated for prediction of PCA flat axis in all eyes ( R 2  = 0.157) with an MAE and an RMSE of 6.254–8.518 ° and 9.052–12.612 ° in external validation, respectively. Binary prediction of eyes with a PCA ≥ 0.50D was performed with a mean AUC of 0.900-0.928 in external validation. For eyes with a PCA ≥ 0.50D, in external validation an MAE of 0.049-0.059D and an RMSE of 0.058-0.076D were achieved in prediction of PCA magnitude, with an MAE of 2.863–4.010 ° and an RMSE of 3.755–5.263 ° in prediction of PCA flat axis. Conclusions Accurate prediction of PCA magnitude and flat axis based on other corneal parameters can be achieved by RR models. The RR models may be useful in refractive surgery to improve personalized astigmatism correction and prevent residual astigmatism.