.png){kind=icon}
.png){kind=icon}
Return
Symptom-based machine learning framework for screening hypothyroidism and hyperthyroidism
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.132005.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Screening for adult thyroid disorders remains challenging as conventional symptom-based screening methods lack sufficient diagnostic accuracy. Furthermore, most existing machine learning (ML) approaches rely heavily on invasive laboratory inputs, which makes them impractical for routine use. This study addresses these gaps by proposing a non-invasive symptom-based ML framework that delivers an interpretable risk calculator for accurate screening of thyroid disorders in settings where regular systematic TSH-based screening is not implemented. The framework integrates population-specific reference intervals to carefully label each instance with the appropriate thyroid class. A multidimensional set of input features, including demographics, family history, symptoms, aggregated symptom scores, and predicted thyroid-stimulating hormone (TSH) levels, was utilized to train and test a sequence of binary classifiers to identify patients with hypothyroidism and hyperthyroidism in both subclinical and overt conditions. Among the implemented models, the XGBoost classifier demonstrated superior, consistent performance compared with the others. The Shapley Additive Explanation (SHAP) enhances model interpretability by revealing distinct feature patterns across classification tasks. Aggregated symptom scores stood as the most influential features across all models. The developed risk calculator not only outperformed conventional symptom-based screening methods but also revealed important clinical insights, highlighting the value of considering the aggregated impact of symptoms alongside predicted TSH levels to detect hypothyroidism and hyperthyroidism. The presented methodological framework can be replicated across different populations in the Middle East or worldwide with appropriate local calibration, which will be the next step of our work. This will assist in using this tool as a screening aid until regular TSH screening is implemented for adult populations. Ultimately, the proposed symptom-based screening tool offers a time-efficient complement to laboratory testing that can support thyroid-disorder screening and management.
Keywords:
thyroid disorders
machine learning
symptom-based screening
XGBoost
SHAP解释
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W