Analysis of the risk factors for the occurrence of dry eye disease in patients with chronic rhinitis: Construction and validation of the DE-Risk prediction model.

Chronic rhinitis and dry eye disease (DED) are common chronic diseases, but the potential association and influencing factors between the 2 are not yet clear. The aim of this study is to explore the influencing factors of the development of DED in patients with chronic rhinitis, and to construct a model for predicting the risk of DED through machine learning, in order to provide reference for clinical prediction and intervention. This study included 530 patients with chronic rhinitis, of whom 318 did not have DED and 212 had DED. Using 3 machine learning models (multi factor logistic regression (LR) model, support vector machine model, and extreme gradient boosting model) to explore the factors that affect the occurrence of DED, and visualizing them using Shapley Additive Explanations values. On this basis, combining the important features and contribution values of 3 models, a Dry Eye Risk model (DE-Risk) was constructed, and 30% of the data was used as the internal validation set to evaluate the performance of the DE-Risk model. Among the 3 machine learning models, extreme gradient boosting had the highest area under the curve value (0.684), support vector machine showed the best accuracy (77.215%) and precision (76.471%), while the LR model performed better in specificity (0.729). The results of the 3 models indicated that symptom score, body mass index, clear nasal secretion, and duration of chronic rhinitis were important characteristics affecting the occurrence of DED. The DE-Risk model constructed based on the standardized average of the first 3 important features had an area under the curve value of 0.706 on the internal validation set, which was higher than all single machine learning models. The DE-Risk model constructed based on the important features of multi factor LR and machine learning models in this study showed good predictive performance, high accuracy and robustness, and could provide strong support for early identification of high-risk patients in clinical practice, offering new insights into the prevention and treatment of DED.