ILSHIP: An interpretable and predictive model for hypothyroidism.

Area of Science:

• Endocrinology
• Medical Informatics
• Machine Learning

Background:

• Hypothyroidism is a prevalent endocrine disorder with increasing incidence.
• Its insidious nature often results in delayed diagnosis and misdiagnosis, complicating treatment.
• Existing diagnostic models lack comprehensive performance and interpretability.

Purpose of the Study:

• To introduce ILSHIP, an interpretable hypothyroidism prediction model.
• To enhance predictive performance and interpretability compared to existing methods.
• To reduce diagnostic complexity and improve hypothyroidism management.

Main Methods:

• The ILSHIP model was developed using label encoding, missing value imputation, feature selection, and data enhancement.
• Performance was evaluated against twelve related study models and eleven mainstream machine learning models (e.g., XGBoost, MLP).
• The SHAP framework was integrated for feature importance analysis.

Main Results:

• ILSHIP achieved high performance metrics: 99.392% accuracy, 99.437% recall, 99.348% specificity, 99.381% F1-score, and 99.960% AUC.
• ILSHIP's accuracy surpassed existing models by 0.7%-15.4%.
• Key hypothyroidism indicators, including thyroid-stimulating hormone (TSH) and free thyroxine index (FTI), were identified as crucial predictors.

Conclusions:

• ILSHIP offers superior predictive accuracy and interpretability for hypothyroidism diagnosis.
• The model provides insights into individual influencing factors, supporting clinical decision-making.
• ILSHIP facilitates earlier and more accurate detection, potentially improving patient outcomes.