Evaluation of prediction errors in nine intraocular lens calculation formulas using an explainable machine learning
Richul Oh1, Joo Youn Oh1,2,3, Hyuk Jin Choi1,2,3,4
1Department of Ophthalmology, Seoul National University Hospital, 101 Daehak-ro, Jongno-gu, Seoul, 03080, Korea.
Background:
The purpose of the study was to evaluate the relationship between prediction errors (PEs) and ocular biometric variables in cataract surgery using nine intraocular lens (IOL) formulas with an explainable machine learning model.
Methods:
We retrospectively analyzed the medical records of consecutive patients who underwent standard cataract surgery with a Tecnis 1-piece IOL (ZCB00) at a single center. We calculated predicted refraction using the following IOL formulas: Barrett Universal II (BUII), Cooke K6, EVO V2.0, Haigis, Hoffer QST, Holladay 1, Kane, SRK/T, and PEARL-DGS. We used a LightGBM-based machine learning model to evaluate the explanatory power of ocular biometric variables for PEs and assessed the relationship between PEs and ocular biometric variables using Shapley additive explanation (SHAP) values.
Results:
We included 1,430 eyes of 1,430 patients in the analysis. The SRK/T formula exhibited the highest R2 value (0.231) in the test set among the machine-learning models. In contrast, the Kane formula exhibited the lowest R2 value (0.021) in the test set, indicating that the model could explain only 2.1% of the PEs using ocular biometric variables. BUII, Cooke K6, EVO V2.0, Haigis, Hoffer QST, Holladay 1, PEARL-DGS formulas exhibited R2 values of 0.046, 0.025, 0.037, 0.194, 0.106, 0.191, and 0.058, respectively. Lower R2 values for the IOL formulas corresponded to smaller SHAP values.
Conclusion:
The explanatory power of currently used ocular biometric variables for PEs in new-generation formulas such as BUII, Cooke K6, EVO V2.0 and Kane is low, implying that these formulas are already optimized. Therefore, the introduction of new ocular biometric variables into IOL calculation formulas could potentially reduce PEs, enhancing the accuracy of surgical outcomes.
Related Concept Videos
01:16Focusing of Light in the Eye
01:03Prediction Intervals
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.