Reducing the number of test items of the Action Research Arm Test post stroke: A decision tree analysis

Published:January 05, 2022DOI:


      Objective – The present study aimed to create a shorter version of the Action Research Arm Test (ARAT) without compromising its measurement properties.
      Design – Secondary analysis of stroke recovery cohorts that used the ARAT to measure upper-limb impairment.
      Setting – Rehabilitation centers.
      Participants – Patients with stroke from five different stroke recovery cohorts.
      Interventions – Not applicable.
      Main Outcome Measures – A decision tree version of the ARAT (ARAT-DT) was developed using Chi-squared Automated Interaction Detection (CHAID). In an independent validation subset, criterion validity, agreement of ARAT-DT with original ARAT scores and score categories, and construct validity with the Fugl-Meyer upper extremity (FM-UE) score were determined.
      Results – In total, 3738 ARAT measurements were available involving 1,425 subjects. CHAID analysis in the development subset (n=2803) revealed an optimized decision tree with a maximum of four consecutive items. In the validation dataset (n=935), the ARAT-DT differed by a mean of 0.19 points (0.3% of the total scale) from the original ARAT scores (limits of agreement: -5.67 to 6.05). The ARAT-DT demonstrated excellent criterion validity with the original ARAT scores (ICC=0.99 and ρ=0.99) and scoring categories (κw=0.97). The ARAT-DT showed very good construct validity with the FM-UE motor scale (ρ=0.92).
      Conclusion – A decision tree version of the ARAT was developed, reducing the maximum number of items necessary for ARAT administration from 19 to 4. The scores produced by the decision tree had excellent criterion validity with original ARAT scores.

      Key words

      To read this article in full you will need to make a payment


      Subscribe to Archives of Physical Medicine and Rehabilitation
      Already a print subscriber? Claim online access
      Already an online subscriber? Sign in
      Institutional Access: Sign in to ScienceDirect