Abstract:Objective: To construct a predictive model for the risk of myopia in preschool children and evaluate its clinical value.Method: A retrospective analysis was conducted on the data of 550 preschool children (550 eyes, including those with poor vision and those with the same vision in the right eye) who visited hospitals from January 2020 to December 2024. The data were divided into a training set (n=385) and a validation set (n=165) in a 7:3 ratio. The data of the children were collected in detail, and the children ''s spherical equivalent (SE) was used as the dependent variable. The multiple linear regression model was used to screen the indicators that affected the biological indicators related to myopia in preschool children. Subsequently, the children in the training set were divided into a myopia group (n=34) and a normal vision group (n=351) based on whether myopia occurred (SE ≤ -0.50D can be diagnosed as myopia according to refractive power assessment). Subsequently, the influencing factors selected by the multiple linear regression model were included as predictive variables, and a risk prediction equation was constructed using Logistic regression. The discrimination, calibration and clinical application value of the prediction model were verified through the receiver operating characteristic (ROC) curve, Hosmer-Lemeshow test and decision curve.Result: The clinical data between the training set and the validation set were comparable (P>0.05). Multiple linear regression analysis showed that the ratio of axial length (AL) to corneal curvature radius (CR), axial length (AL), choroidal thickness (SFCT), anterior chamber depth (ACD), age, and parental myopia history were risk factors for myopia in preschool children (all P<0.05). Based on the above risk factors, a Logistic regression risk prediction model was constructed to obtain a warning model logit (P)=0.881 × AL/CR-0.318 × SFCT+0.408 × ACD+2.064 × age+1.442 × parental myopia history -29.310. The evaluation of the effectiveness of the myopia risk prediction model for preschool children based on the training set showed that the area under the ROC curve (AUC) was 0.941 (95% CI: 0.916-0.966), and the H-L test was χ2=4.194, P=0.839. The decision curve indicated that within the threshold range of 0.01-0.72 and 0.77-0.98, the net return rate was greater than 0, suggesting that the model had good discrimination, calibration, and clinical application value in predicting the risk of myopia in preschool children compared to the actual risk. Internal validation was conducted in an independent validation set, and the results showed that the AUC of the ROC curve was 0.967 (95% CI: 0.919-1.000), the H-L test was χ2=3.818, P=0.873, and the decision curve indicated that within the threshold range of 0.01-0.99, the net return rate was greater than 0, further suggesting that the model prediction had good consistency with the actual risk.Conclusion: Age, parental history of myopia, AL/CR, SFCT, and ACD are risk factors for myopia in preschool children. The Logistic regression risk prediction model constructed based on these factors has high predictive value and can effectively evaluate the probability of myopia occurrence in individuals, providing reference for individualized prevention of myopia risk in preschool children.
