The criterion for keeping a variable in the forward stepwise regression was a significant contribution to the model (P≤.05).

The criterion for removing a variable was if it was not making a significant contribution to the model (P≥0.1). Paired t tests were used to compare the ARAT, FMA, and MAL scores before and after TST. Significance was set at alpha=.05. Thirty-three patients (13 women; mean age, 61.5y) were included. Participant characteristics and assessment scores are selleck presented in table 1. There were no significant differences in function or MAL scores between those who received active (n=16) or sham (n=17) somatosensory stimulation at baseline or for the changes 3 months after TST (independent samples t test; P>.05); therefore, all participants were grouped together for the analyses. The mean time since stroke ± SD was 37.7±36.7 months, baseline ARAT score was 29.5±11.9, and FMA score was 40.0±10.5. All participants were right handed prior to stroke, and 19 had their right arm affected. Three participants failed to attend the 3-month follow-up assessment; therefore, their data are not included for the prediction of change in MAL amount of use. The results of the Spearman correlations

are presented in table 2. There was a significant negative correlation between the amount of use and the MAS (P=.001), and there were positive correlations with the ARAT and FMA (P<.01) ( fig 1). The baseline ARAT score predicted 47% of the variability in baseline MAL amount of use (F1,31=27.457; Dasatinib in vitro P<.001). In using the equation for the regression model, an ARAT score of 54 is required to reach an amount of use score of 2.5 (half the maximum value, described as between rarely and half as much as before the stroke). All other

clinical variables were excluded, not significantly adding to the predictive power of the model (all P>.19). If participants were examined separately based on which hand was affected, the baseline ARAT score still strongly predicted the amount of use for those with the dominant hand PAK6 affected (R2=0.6; F1,17=25.518; P<.001). The equation for this regression model calculates that an ARAT score of 46 is required for an amount of use score of 2.5. For participants with the nondominant hand affected, the ARAT gross component score predicted 56.8% of the variability in the amount of use (F1,12=15.806; P=.002). The equation for the regression model calculates that patients will not score ≥2.5 even if they reach a maximum score on the grasp component of the ARAT. The predictive power of the model was further increased when the FMA wrist component score was added (R2=0.7; F2,11=13.069; P=.001). ARAT, FMA, and MAL scores increased significantly after TST (P<.01) (see table 1). Changes in the ARAT score predicted 30.8% of the variability in change in MAL amount of use (F1,28=12.486; P=.001). The relation between change in ARAT score and change in the amount of use is presented in figure 2.