MedElevator
Войти в исследования JMIR medical informatics · 2026
Screening for Cough Variant Asthma Using the TabPFN Model on Routine Laboratory Data in a Multicenter Cohort: Prospective Validation Study.
СтатьяМногоцентровое исследованиеВалидационное исследованиеНазвание 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.
Yang Li-Jie, Wang Shi-Ning, Shi Xiu, Deng Ni-Shan, Zhu Miao-Juan, Wang Yun-Fan, Nie Han-Xiang
Аннотация
BACKGROUND: Cough variant asthma (CVA) is a prevalent but underdiagnosed subtype of chronic cough, largely due to the limited accessibility of pulmonary function tests in primary care.
OBJECTIVE: This study aimed to develop and prospectively validate an interpretable machine learning model using routine laboratory data to facilitate accessible CVA screening.
METHODS: We conducted a multicenter study combining retrospective model development and prospective validation using data from January 2021 to August 2025. The study included 3 sets-a retrospective development set (n=533; 428 training and 105 internal testing) and 2 prospective validation sets (internal validation: n=59 and external validation: n=78). In total, 54 routine biochemical, coagulation, and hematological markers were analyzed. A Tabular Prior-Data Fitted Network (TabPFN) model was developed and benchmarked against extreme gradient boosting (XGBoost). Performance was evaluated via area under the curve (AUC), accuracy, and specificity. Shapley Additive Explanations (SHAP) and decision curve analysis (DCA) were used to assess interpretability and clinical utility.
RESULTS: In external validation, TabPFN achieved an AUC of 0.858 (95% CI 0.763-0.943), accuracy of 79.5%, sensitivity of 71.4%, and specificity of 86%. XGBoost achieved a higher external AUC of 0.912 (95% CI 0.842-0.982) and sensitivity of 91.4%, but lower accuracy (75.6%) and specificity (62.8%) at the fixed 0.5 probability threshold. TabPFN retained higher AUCs than XGBoost in internal testing (0.964 vs 0.890) and temporal validation (0.925 vs 0.846). DCA demonstrated that TabPFN provided higher net benefit across clinically relevant threshold probabilities, with the range varying across validation cohorts. SHAP analysis identified serum sodium (Na) and prothrombin time international normalized ratio as the leading predictors, followed jointly by mean corpuscular hemoglobin concentration and red cell distribution width-SD, which showed the same rounded mean absolute SHAP value at the reported 2-decimal precision, and elevated sodium levels (>141 mmol/L) were associated with increased predicted risk.
CONCLUSIONS: The TabPFN model demonstrated robust prospective performance as an accessible screening tool for CVA, with balanced threshold-dependent classification performance across validation cohorts. Although TabPFN showed higher discrimination than XGBoost in internal and temporal validation, XGBoost achieved a higher AUC in the external cohort, indicating that neither model uniformly outperformed the other across all performance dimensions.
OBJECTIVE: This study aimed to develop and prospectively validate an interpretable machine learning model using routine laboratory data to facilitate accessible CVA screening.
METHODS: We conducted a multicenter study combining retrospective model development and prospective validation using data from January 2021 to August 2025. The study included 3 sets-a retrospective development set (n=533; 428 training and 105 internal testing) and 2 prospective validation sets (internal validation: n=59 and external validation: n=78). In total, 54 routine biochemical, coagulation, and hematological markers were analyzed. A Tabular Prior-Data Fitted Network (TabPFN) model was developed and benchmarked against extreme gradient boosting (XGBoost). Performance was evaluated via area under the curve (AUC), accuracy, and specificity. Shapley Additive Explanations (SHAP) and decision curve analysis (DCA) were used to assess interpretability and clinical utility.
RESULTS: In external validation, TabPFN achieved an AUC of 0.858 (95% CI 0.763-0.943), accuracy of 79.5%, sensitivity of 71.4%, and specificity of 86%. XGBoost achieved a higher external AUC of 0.912 (95% CI 0.842-0.982) and sensitivity of 91.4%, but lower accuracy (75.6%) and specificity (62.8%) at the fixed 0.5 probability threshold. TabPFN retained higher AUCs than XGBoost in internal testing (0.964 vs 0.890) and temporal validation (0.925 vs 0.846). DCA demonstrated that TabPFN provided higher net benefit across clinically relevant threshold probabilities, with the range varying across validation cohorts. SHAP analysis identified serum sodium (Na) and prothrombin time international normalized ratio as the leading predictors, followed jointly by mean corpuscular hemoglobin concentration and red cell distribution width-SD, which showed the same rounded mean absolute SHAP value at the reported 2-decimal precision, and elevated sodium levels (>141 mmol/L) were associated with increased predicted risk.
CONCLUSIONS: The TabPFN model demonstrated robust prospective performance as an accessible screening tool for CVA, with balanced threshold-dependent classification performance across validation cohorts. Although TabPFN showed higher discrimination than XGBoost in internal and temporal validation, XGBoost achieved a higher AUC in the external cohort, indicating that neither model uniformly outperformed the other across all performance dimensions.