Speaker
Description
Self-report questionnaires are widely used in healthcare to assess disease risk and symptom severity. However, their length can burden respondents and compromise data quality. While machine learning models have enabled the development of shortened questionnaires with high predictive performance, they often operate as black boxes, limiting transparency and requiring specialized expertise that hinders clinical adoption.
To address this, we have developed the Symbolic Regression-Based Clinical Score Generator (SymScore), a framework designed to produce score tables for shortened questionnaires while maintaining accuracy comparable to machine learning approaches. SymScore employs symbolic regression to optimize response grouping and assign predictive weights that capture the relationship between questionnaire responses and disease severity. The resulting score tables provide a transparent and practical tool for clinical use.
SymScore achieves performance comparable to high-accuracy machine learning-based instruments, including MCQI-6 (MAE = 9.94, R² = 0.82) and SLEEPS (AUROC = 0.88–0.94), developed for assessing sleep disorders. Beyond these applications, SymScore has also been applied to questionnaires evaluating sleep-related cognitive dysfunction in patients with cancer.
By combining predictive performance with interpretability, SymScore offers a practical pathway for translating advanced computational methods into trustworthy and accessible tools for healthcare professionals.
