Objective To construct the risk prediction model for pathologic myopia in patients with high myopia in fundus tessellation by employing machine learning and Logistic regression model. Methods Clinical data of patients with high myopia in fundus tessellation, who were diagnosed at the Ophthalmology Department of a hospital between May 2023 and April 2024 (n=418, development dataset), and between September 2024 and November 2024 (n=98, external validation set), including demographic characteristics, eye usage habits, family history of myopia, and ophthalmic examination results, were collected. The development dataset was randomly divided into training set (n=333) or internal validation set (n=85) at a ratio of 8∶2. Based on the training set data, three machine learning algorithms (random forest, gradient boosting decision tree, and extreme gradient boosting) were used to screen variables, and combined with variable clustering analysis, the final candidate predictor variables were determined, which were included in the Logistic regression model to construct the pathologic myopia risk prediction model. Predictive performance of the model was evaluated using the internal and external validation sets. Results Seven candidate predictor variables, including axial length, corneal curvature, peripapillary atrophy area, age, body mass index, spherical equivalent, and optic disc tilt ratio, were finally screened. The formula for predicting the risk of pathologic myopia was 

P=[e-12.390-4.040×optic disc tilt ratio+0.657×axial length+0.148×peripapillary atrophy area-0.212×spherical equivalent-0.127×corneal curvature+0.020×age+0.055×body mass index1+ee-12. 390-4.040×optic disc tilt ratio+0.657×axial length+0.148×peripapillary atrophy area-0.212×spherical equivalent-0.127×corneal curvature+0.020×age+0.055× body mass index] Axial length, spherical equivalent and optic disc tilt ratio were independently associated with pathologic myopia in patients with high myopia in fundus tessellation. The area under the receiver operating characteristic curve of the model in the internal validation set and the external validation set was 0.877 and 0.760, respectively, which exhibited good prediction performance. Conclusion Axial length, spherical equivalent, and optic disc tilt ratio are independent influencing factors for pathologic myopia in highly myopic patients with fundus tessellation. The pathologic myopia risk prediction model based on 7 candidate predictor variables screened by machine learning combined with Logistic regression model has good discriminative ability and interpretability, which can provide auxiliary reference for the risk assessment and early screening of pathologic myopia in patients with high myopia in fundus tessellation.

无