Accepted for/Published in: JMIR Medical Informatics
Date Submitted: Jan 26, 2026
Date Accepted: Sep 14, 2026
Screening for Cough Variant Asthma Using the TabPFN Model on Routine Laboratory Data in a Multi‑center Cohort: A Prospective Validation Study
ABSTRACT
Background:
Cough variant asthma (CVA) is a frequent cause of chronic cough but remains significantly underdiagnosed in primary care due to limited access to complex pulmonary function testing.
Objective:
This study aimed to develop and prospectively validate an interpretable machine learning model utilizing routine laboratory data to facilitate accessible CVA screening.
Methods:
The study conducted a multi-center study combining retrospective model development and prospective validation using data from January 2021 to August 2025. The study included three sets: a retrospective development set (N=533; 428 training/105 internal testing) and two prospective validation sets (59 internal, 78 external). Fifty-four routine biochemical, coagulation, and hematological markers were analyzed. A foundation model based on the Tabular Prior-Data Fitted Network (TabPFN) was developed and rigorously benchmarked against XGBoost. Performance was evaluated via AUC, accuracy, and specificity. SHapley Additive exPlanations (SHAP) and Decision Curve Analysis (DCA) were employed to assess interpretability and clinical utility.
Results:
In external validation, the TabPFN model demonstrated robust generalization with an AUC of 0.858 (95% CI: 0.764–0.939), an accuracy of 79.5%, and a specificity of 86.0%, This outperformed the XGBoost benchmark, which exhibited severe overfitting (training accuracy 100% vs. external specificity 51.2%). DCA indicated that TabPFN provided a higher net benefit across a broad probability threshold range (0.05–0.95). SHAP analysis identified Serum Sodium (Na), PT-INR, and RDW-SD as key predictors, with Na levels >141 mmol/L indicating increased risk.
Conclusions:
The TabPFN foundation model offers a robust and cost-effective screening tool for CVA, demonstrating superior generalization over traditional algorithms in small-to-medium clinical datasets. By leveraging routine blood markers, this approach has the potential to bridge the diagnostic gap in resource-limited primary care settings.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.