A predictive model based on random forest for shoulder-hand syndrome

Area of Science:

• Neurology
• Rehabilitation Medicine
• Data Science in Healthcare

Background:

• Shoulder-hand syndrome (SHS) significantly hinders post-stroke functional recovery.
• Current understanding lacks high-risk factors and effective treatments for SHS.
• Early identification of SHS risk is crucial for stroke patient management.

Purpose of the Study:

• To develop a predictive model for SHS occurrence post-stroke using random forest (RF) algorithm.
• To identify demographic, clinical, and laboratory factors associated with high risk of SHS.
• To explore potential therapeutic strategies based on identified risk factors.

Main Methods:

• Retrospective analysis of 36 first-onset stroke patients with hemiplegia.
• Application of RF algorithms for predicting SHS occurrence.
• Model validation using confusion matrix and ROC curve analysis.

Main Results:

• A binary classification model achieved an Area Under the ROC Curve of 0.8 and 72.73% out-of-bag accuracy.
• The model demonstrated a sensitivity of 0.8 and specificity of 0.5.
• Top predictive features for SHS were D-dimer, C-reactive protein (CRP), and hemoglobin.

Conclusions:

• A reliable predictive model for SHS can be built using patient data.
• D-dimer, CRP, and hemoglobin are significant factors influencing SHS occurrence post-stroke.
• Findings support targeted interventions for high-risk SHS patients.