Toward Personalized Web-Based Cognitive Rehabilitation for Patients With Ischemic Stroke: Elo Rating Approach

Area of Science:

• Neurorehabilitation
• Cognitive Science
• Data Science in Healthcare

Background:

• Stroke survivors frequently experience cognitive impairments, necessitating specialized inpatient rehabilitation.
• Personalizing web-based cognitive rehabilitation tasks requires balancing patient skill with task difficulty, a persistent challenge.
• Elo ratings are proposed to quantify patient cognitive skills and guide task assignment for improved rehabilitation.

Purpose of the Study:

• To stratify ischemic stroke patients into Elo rating levels during early rehabilitation.
• To examine the relationship between Elo ratings, task difficulty, and rehabilitation outcomes.
• To determine if early Elo ratings predict rehabilitation success.

Main Methods:

• Elo ratings were calculated using the PlayerRatings R library for 288 adult patients.
• Cognitive functions, including working memory (DIGITS) and verbal memory (RAVLT subtests), were assessed.
• Predictive models used logistic and linear regressions, evaluating discrimination with AUC and adjusted R².

Main Results:

• Three Elo levels (low, middle, high) were established, with the middle group showing the highest improvement across all memory tasks (P<.001).
• The middle Elo level demonstrated the best alignment between patient skills and task difficulties across 38,177 task executions (P<.001).
• Elo ratings significantly predicted rehabilitation outcomes for RAVLT075 (R²=0.54) and DIGITS (R²=0.43), with strong predictive accuracy for improvement (AUCs 0.73-0.81).

Conclusions:

• Elo ratings provide a valuable tool for clinicians to identify patient cognitive profiles early in rehabilitation.
• This system supports personalized task difficulty assignments, enhancing cognitive rehabilitation strategies for stroke survivors.