Nonimpaired literature-derived scaling does not accurately represent the shoulder and elbow strength of children and adults with paraplegia.

This study evaluated whether participant characteristics, literature-derived regression equations, and a generic musculoskeletal model developed for nonimpaired populations could accurately predict maximal shoulder and elbow strength in children and adults with paraplegia. Thirty individuals with paraplegia (22 adults, 8 children) completed maximal voluntary isometric contractions (MVICs) in seven upper extremity postures. Predicted MVICs were generated using a scaled OpenSim upper extremity model with three muscle activation inputs (Full, Separated, and EMG-informed). Linear mixed-effects models evaluated relationships between predicted and experimentally measured MVICs across postures. Model performance was assessed using marginal and conditional R2, root mean square error (RMSE), mean absolute error (MAE), Lin's concordance correlation coefficient (CCC), and repeated-measures Bland-Altman analyses. Agreement between measured and height-predicted upper extremity segment lengths was also evaluated. Joint posture significantly influenced measured MVIC, and significant posture-by-predicted MVIC interactions demonstrated that prediction accuracy was posture dependent (all p < 0.001). The Full activation input explained the greatest proportion of variance in measured MVIC (marginal R2 = 0.770; conditional R2 = 0.931), whereas the EMG-informed activation input produced the lowest prediction error (RMSE = 22.38% MVIC; MAE = 17.14% MVIC). However, agreement between predicted and measured MVIC remained modest across all activation inputs (CCC = 0.259-0.312), with systematic bias evident in Bland-Altman analyses. Height-based anthropometric scaling overestimated forearm and humerus lengths by 1.25 and 4.96 cm, respectively, with greater error for the humerus (RMSE = 5.82 cm). Regression-based scaling and generic musculoskeletal models derived from nonimpaired populations provided only modest estimates of maximal upper extremity strength in individuals with paraplegia. Improving strength prediction will require innovative scaling approaches that incorporate participant-specific anthropometry, population-specific muscle properties, and individualized neuromuscular activation while balancing model accuracy with practical clinical feasibility.