Frontiers in cell and developmental biology · 2026
A TabPFN-based prediction system for refractive error and dry eye comorbidity: a retrospective study using large-scale real-world data.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Qin Danyi, Guan Wenying, Wu Shinan, Xu Changsheng, Liu Yuwen, Yan Bing, Lv Jingyao, Li Xiaoxin, Liu Zuguo
Аннотация
INTRODUCTION: Refractive error and dry eye are highly prevalent ocular conditions that significantly impair the quality of life and impose a substantial burden on individuals and society. Growing evidence suggests a correlation between these two conditions. This study aimed to develop and validate a machine learning (ML) model to accurately predict the risk of concurrent dry eye comorbidities in patients with refractive error.
METHODS: Data from Xiamen Eye Center outpatient database (1st January 2024 to 28th February 2025) were analyzed (n = 114,579). Hyperparameter optimization, Spearman correlation analysis, and logistic regression analyses were performed. The final feature set was determined using a Random Forest algorithm with the sequential forward selection technique. Eight ML algorithms were evaluated through ten-fold cross-validation. The optimal model was selected based on a comprehensive assessment of the receiver operating characteristic curve, precision-recall curve, and decision curve analysis. For the best-performing model, SHapley Additive exPlanations and partial dependence plots were utilized to interpret the importance and interactions of risk factors.
RESULTS: Baseline characteristics were comparable between the training set and the internal test set, while significant differences were observed in multiple baseline characteristics between dry eye group and non-dry eye group among subjects with refractive error. Based on ten selected feature variables, the tabular prior-data fitted network (TabPFN) model demonstrated the best performance, showing high screening efficacy with both specificity and accuracy reaching 0.945. The interaction analysis revealed that a longer duration of refractive error was associated with a higher risk of dry eye, a relationship that was particularly pronounced among older and female patients. Furthermore, an online web calculator was developed to deploy this diagnostic prediction model.
DISCUSSION: This study developed a high-performance and interpretable ML system based on a large-scale real-world clinical dataset for the early prediction of concurrent dry eye risk in patients with refractive error. The system holds significant potential as a predictive aid for clinical decision-making, enabling more timely and personalized patient management, thereby offering substantial clinical value and promising application prospects.
METHODS: Data from Xiamen Eye Center outpatient database (1st January 2024 to 28th February 2025) were analyzed (n = 114,579). Hyperparameter optimization, Spearman correlation analysis, and logistic regression analyses were performed. The final feature set was determined using a Random Forest algorithm with the sequential forward selection technique. Eight ML algorithms were evaluated through ten-fold cross-validation. The optimal model was selected based on a comprehensive assessment of the receiver operating characteristic curve, precision-recall curve, and decision curve analysis. For the best-performing model, SHapley Additive exPlanations and partial dependence plots were utilized to interpret the importance and interactions of risk factors.
RESULTS: Baseline characteristics were comparable between the training set and the internal test set, while significant differences were observed in multiple baseline characteristics between dry eye group and non-dry eye group among subjects with refractive error. Based on ten selected feature variables, the tabular prior-data fitted network (TabPFN) model demonstrated the best performance, showing high screening efficacy with both specificity and accuracy reaching 0.945. The interaction analysis revealed that a longer duration of refractive error was associated with a higher risk of dry eye, a relationship that was particularly pronounced among older and female patients. Furthermore, an online web calculator was developed to deploy this diagnostic prediction model.
DISCUSSION: This study developed a high-performance and interpretable ML system based on a large-scale real-world clinical dataset for the early prediction of concurrent dry eye risk in patients with refractive error. The system holds significant potential as a predictive aid for clinical decision-making, enabling more timely and personalized patient management, thereby offering substantial clinical value and promising application prospects.