Related Experiment Video
Updated: Aug 20, 2025
Computerized Adaptive Testing System of Functional Assessment of Stroke
Published on: January 7, 2019
Development of a Short-Form Stroke Impact Scale Using a Machine Learning Algorithm for Patients at the Subacute Stage
Shih-Chieh Lee1, Inga Wang2, Gong-Hong Lin3
1Shih-Chieh Lee, PhD, is Postdoctoral Researcher, Department of Occupational Therapy, College of Medicine, National Cheng Kung University, Tainan City, Taiwan; Adjunct Assistant Professor, School of Occupational Therapy, College of Medicine, National Taiwan University, Taipei, Taiwan; and Adjunct Assistant Professor, Institute of Long-Term Care, MacKay Medical College, New Taipei City, Taiwan. At the time this article was submitted, Lee was Postdoctoral Researcher, School of Occupational Therapy, College of Medicine, National Taiwan University, Taipei, Taiwan.
A new machine learning-based short form of the Stroke Impact Scale (ML-SIS) was developed. This efficient ML-SIS provides valid and comparable scores to the original measure, reducing administration time.
Area of Science:
- Neuroscience
- Rehabilitation Medicine
- Health Informatics
Background:
- The original Stroke Impact Scale (SIS 3.0) takes approximately 20 minutes to administer.
- Existing short forms of the SIS 3.0 lack comparability with the original measure's scores.
Purpose of the Study:
- To develop a machine learning-derived short form of the Stroke Impact Scale 3.0 (ML-SIS).
- To ensure the ML-SIS provides valid scores comparable to the original SIS 3.0.
Main Methods:
- A machine learning algorithm, specifically deep neural networks, was used to identify key items.
- Items were iteratively selected based on prediction accuracy (R² ≥ .90) for original domain scores.
- The ML-SIS underwent validation for comparability, concurrent validity, and convergent validity.
Main Results:
- The ML-SIS comprises 28 items, utilizing approximately half the items of the original SIS 3.0.
- High R² values (.90-.96) and low residuals confirmed good comparability.
- Strong correlations (r = .95-.98) demonstrated sufficient concurrent and convergent validity.
Conclusions:
- The ML-SIS offers an efficient alternative to the SIS 3.0, with an estimated 10-minute administration time.
- The ML-SIS provides valid and comparable scores, making it suitable for clinical settings.
- This study introduces a novel, data-driven approach to developing shorter, yet reliable, patient-reported outcome measures.
