It is common for researchers in spinal cord injury (SCI) to treat limbs from the same individual as independent to increase sample size in limb-based analyses. However, this approach may violate independence assumptions required for traditional statistical tests. This study investigated the dependence of bilateral upper-limb motor and neurophysiological outcomes through a retrospective analysis of participants from the European Multicenter Study about Spinal Cord Injury. Data from 118 participants with acute cervical SCI (236 limbs) were analyzed, including ulnar compound muscle action potentials (CMAPs) recorded in the abductor digiti minimi and manual muscle scores for the flexor digitorum profundus, bilaterally assessed at 12 and 48 weeks postinjury. Our analysis stratified participants by injury severity into two groups: motor complete (n = 55) and motor incomplete (n = 63). Interlimb dependence was assessed using Spearman's rank correlations. To illustrate the implications of such dependence for inference, bootstrap analyses of motor recovery (12 vs. 48-week CMAPs) compared three analytical approaches using a linear mixed-effects model framework: randomly selecting outcomes from one limb per participant, including both limbs but ignoring intraparticipant correlation, and including both limbs but accounting for intraparticipant correlation. Bootstrap resampling was performed across sample sizes ranging from n = 10 to 50. Moderate-to-strong correlations (Spearman's rho = 0.44-0.91) were observed between bilateral CMAPs and bilateral strength scores in both groups at 12 and 48 weeks postinjury. All models yielded similar CMAP recovery estimates (∼1.6 to 1.8 mV), but the precision of these estimates and the empirical power, reflecting the probability of detecting a significant recovery effect given the observed effect size and variability, varied across methods. Treating limbs as independent overestimated precision, resulting in the highest power estimates, whereas including only one limb reduced statistical power, particularly at lower sample sizes. Accounting for intraparticipant dependence yielded intermediate estimates of precision and power. These findings demonstrate that bilateral upper-limb outcomes in SCI are statistically dependent and highlight the need for statistical procedures that account for intraparticipant correlation in SCI research to ensure valid study conclusions.
Missen et al. (Wed,) studied this question.