| Popis: |
Research Objectives To develop a short form of the Berg Balance Scale (BBS-ML) using a machine learning approach. Design Cross-sectional study. Setting General community. Participants A sample of 408 patients with stroke from the Locomotor Experience Applied Post-stroke (LEAPS) study was used to develop the short form. An independent sample of 226 patients with stroke in Taiwan was used for cross-validation. Interventions Not applicable. Main Outcome Measures Berg Balance Scale (BBS). Results We applied feature selection algorithms to identify the most useful items for developing a short form. The highest-rated items, ranging from top 4 to top 8, were selected to develop 5 initial versions of short forms (i.e., 4-, 5-, 6-, 7-, and 8-item BBS-ML short forms). The R-squares were 0.91, 0.94, 0.97, 0.96, and 0.98 for the 4-, 5-, 6-, 7-, and 8-item BBS-ML, respectively. The 95% limits of agreements (LoAs) were 15.6, 12.5, 9.6, 10.8, and 8.3, respectively. The possible scoring points (PSPs) were 32, 32, 35, 35, and 36, respectively. The 6-item version was determined as the final version of the BBS-ML because it showed relatively high predictive power of the BBS sum scores, relatively small differences with the observed BBS scores, and the highest PSP. Conclusions The BBS-ML appears to be a promising alternative to improve the efficiency of administrations for busy users. Author(s) Disclosures The authors declare that there is no conflict of interest. |
