Developing machine learning-enhanced WHODAS 2.0 short forms for persons with dementia.

Area of Science:

• Gerontology
• Neuroscience
• Health Informatics

Background:

• Disability assessment in dementia is crucial for care planning.
• The full World Health Organization Disability Assessment Schedule 2.0 (WHODAS 2.0) is time-intensive, limiting clinical utility.
• Developing efficient assessment tools is essential for dementia care.

Purpose of the Study:

• To develop and validate machine learning (ML)-based short forms of the WHODAS 2.0 for dementia disability assessment.
• To evaluate the reliability, concurrent validity, and responsiveness of these ML-based short forms.
• To compare the performance of ML short forms against the full WHODAS 2.0 and a standard short form.

Main Methods:

• Utilized a large dataset of 51,245 persons with dementia for training and validation.
• Developed two ML short forms (ML-WHODAS-16 and ML-WHODAS-10) using Extreme Gradient Boosting and bootstrap-based item selection.
• Compared ML short forms with the full WHODAS-32 and a 12-item short form (Standard-12), assessing anchor-based longitudinal validity.

Main Results:

• Both ML short forms demonstrated high internal consistency (α=0.96 and 0.93) and excellent concurrent validity (r=0.98) with the full WHODAS-32.
• ML short forms showed lower error and negligible bias compared to the Standard-12, meeting equivalence criteria.
• Responsiveness was comparable to the Disability Severity Grade, with identified minimal clinically important differences.

Conclusions:

• ML-WHODAS-16 and ML-WHODAS-10 offer practical, low-burden alternatives for dementia disability assessment.
• These short forms closely reflect full WHODAS-32 scores, suitable for group-level and longitudinal monitoring.
• Further external validation and testing are needed for individual-level interchangeability.