Development of a 16-item Short Form for Fugl-Meyer Assessment of Upper Extremity (FMA-UE) Scale

Research Objectives: This study was undertaken to develop a short form of the Fugl-Meyer Assessment of Upper Extremity (FMA-UE) Scale using a machine learning approach.

Design: Cross-sectional observational study.

Setting: Hospitals.

Participants: Data of 408 individuals post-stroke were extracted from a published database.

Interventions: N/A

Main Outcome Measures: Fugl-Meyer Assessment of Upper Extremity (FMA-UE).

Results: We selected short-form items using the feature importance and feature selection in the Gradient Boosted Tree (XGBoost) machine learning (ML) algorithm. The validity of the short form was examined by computing the (1) residuals, R-squared value, and mean absolute error, (2) intraclass correlation coefficient (ICC), (3) standardized response mean (SRM), and (4) kappa statistics between the predicted scores derived from the short form and the observed scores from the original FMA-UE. The 16-item version was selected as the final short form (FMA-UE ML-16). Regrading differences/associations between the predicted and observed FMA-UE scores, the mean (SD) residuals ranged from 0.0 (0.4) to -0.2 (0.3), R-squared values were all 0.98, and the mean absolute errors ranged from 1.9 to 2.1. The ML-16 correlated highly with the original FMA-UE (all ICCs = 0.99), suggesting excellent concurrent validity. The SRMs derived from the ML-16 were 0.58, 0.50, and 0.28, which were similar to SRMs derived from the original FMA-UE scale (0.65, 0.57, and 0.35, respectively). The kappa statistics ranged from 0.60 to 1.00.

Conclusions: The ML-16 seems a promising alternative to improve administrative efficiency in busy clinical settings and research studies.

Author(s) Disclosures: The authors have no conflicts of interest to declare that are relevant to the content of this submission.