A retrospective analysis of data from 42 chronic stroke survivors (n = 21 left-hemisphere damage, LHD) was conducted. Participants were enrolled as part of a larger phase-IIb clinical trial (Dose Optimization for Stroke Evaluation, ClinicalTrials.gov ID: NCT01749358) (21) and provided informed consent in accordance with the 1964 Declaration of Helsinki and the guidelines of the Institutional Review Board for the Health Science Campus of the University of Southern California. Participants were at least 150 days post-stroke, pre-morbidly right-handed with mostly resolved upper extremity paresis. For a complete description of the inclusionary and exclusionary criteria, please refer to Supplementary Materials.

All analyses were conducted in the R statistical computing package version 3.5.1 (25). To test the hypothesis that the inter-limb relationship of motor capacity is modified by the side of stroke lesion, we used the coefficient of determination (R²) and compared the covariances between LHD and RHD using the Fisher's Z-test. We then performed a simple linear regression to determine the slope of the relationship between contralesional (CL) UEFM and ipsilesional (IL) dWMFT (Model 1), and, CL dWMFT and IL dWMFT (Model 2). We used t-tests to compare slope between LHD and RHD.

To supplement these primary analyses and as a more robust assessment of the interaction between the side of lesion and contralesional motor capacity, we used multiple linear regression of the following form:

In both models, y is the average time score on the distal WMFT of the ipsilesional hand. Using this multiple model, our hypotheses were that β₁ ≠ 0 and β₃ ≠ 0 (see Figure 1). Any statistically significant interaction was resolved post-hoc using a t-test comparison of estimated marginal trends between LHD and RHD.

All continuous variables were assessed for normality using Lilliefors test (modified Kolmogorov–Smirnov test). Distributions for chronicity and average time-score for the distal WMFT were positively skewed and were therefore log-transformed. Welch's t-tests were used to compare age, chronicity, and Upper Extremity Fugl-Meyer scores between LHD and RHD, whereas Chi-square test was used to compare the proportion of females and males between the two groups. Each group was standardized to its own unit variance (z-scored) to equalize range and for subsequent linear regression analysis. Outliers were identified by visual inspection of scatterplots. Any value of IL dWMFT more extreme than ±1.5 log-SD was examined carefully for their influence on interlimb covariance and slope. If removal of these observations did not change the direction or significance of the effect in the simple model, we included them in the final model. Residuals of the final model were analyzed to confirm that all necessary assumptions for multiple regression were met. Significance level (α) was set at p < 0.05.

In order to select the predictor variables that best explain the response, we used a backward selection approach, in which we began by adding all predictor variables in each of the two above models to explain the response variable y. This included our hypothesized predictors, CL UEFM (or CL dWMFT) and the side of stroke lesion (LHD or RHD), and, potential known confounders (age, chronicity, and sex). In a combined full model, those predictors that met a liberal cut-off of p = 0.2 were preserved in the final reduced model. Based on this selection process, we found sex to be a significant confounder (p = 0.08) in Model 1, and therefore included it as a predictor in the reduced Model 1. For Model 2, none of the confounders met the cut-off p-value, except our hypothesized predictors. Additional information on model selection and model diagnostics is included as Supplementary Materials. Standard errors and 95% CI of the estimates of regression coefficients were confirmed by performing 1,000 bootstrap replicates.

Contralesional UEFM explained 42% of the variance in ipsilesional hand motor capacity in LHD (p = 0.001), but <1% in RHD (p > 0.05). The slope of this relationship was −0.65 ± 0.17 (p = 0.001) in LHD and −0.066 ± 0.23 (p = 0.78) in RHD. Compared to RHD, the covariance between contralesional UEFM and ipsilesional hand motor capacity was significantly stronger (Fisher's z = 2.12, p = 0.03) and the slope was steeper in LHD (t = −2.03, p = 0.04).

Four observations (two each in LHD and RHD) were identified as potential outliers. After removal of these outliers, contralesional UEFM explained 44.3% of the variance in ipsilesional hand motor capacity in LHD (p = 0.001), and 2.26% in RHD (p > 0.05). The slope of this relationship changed to −0.42 ± 0.11 (p = 0.001) in LHD and 0.13 ± 0.20 (p = 0.54) in RHD. Again, a comparison of the covariances and slope between the groups revealed that compared to RHD, the relationship between contralesional UEFM and ipsilesional hand motor capacity was significantly stronger (Fisher's z = 2.7, p = 0.006) and the slope was steeper in LHD (t = −2.41, p = 0.02).

Since these observations did not significantly change the strength of covariance nor the slope of the relationship, they were preserved in the final multiple model. Analysis of residuals of the final model did not indicate violations of necessary assumptions in multiple regression in terms of linearity, equality of variance, independence and normality of errors, and multicollinearity of independent variables, nor the presence of unduly influential observations. Nonetheless, estimates below are reported both with and without suspected outliers.

After adjusting for main effects and significant confounders using multiple regression, the final reduced form of Model 1 was statistically different from a null model (F = 3.47, p = 0.016, adjusted R² = 0.19). Based on estimates from Model 1, CL impairment (UEFM) was significantly associated with IL hand motor capacity, i.e., dWMFT (β₁ = −0.72 ± 0.21, p = 0.001; without outliers: −0.44 ± 0.17, p = 0.01) (Figure 2A). There was no significant effect of the side of lesion (β₂ = 0.026 ± 0.27, p = 0.92; without outliers: 0.22 ± 0.23, p = 0.33). There was a significant interaction between the side of lesion and CL impairment (β₃ = 0.66 ± 0.30, p = 0.024; without outliers: 0.56 ± 0.24, p = 0.024). Post-hoc contrasts of estimated marginal trends indicated that the slope of the relationship between CL UEFM and IL dWMFT was significantly more negative in LHD compared to RHD (t = −2.34, p = 0.02; without outliers: −2.37, p = 0.02). Figures 2B,C illustrates the interaction.

Contralesional dWMFT explained 65% of the variance in ipsilesional hand motor capacity in LHD (p < 0.001), but only 9% in RHD (p > 0.05). The slope of this relationship was 0.81 ± 0.13 (p < 0.001) in LHD and 0.29 ± 0.22 (p = 0.19) in RHD. A comparison of the covariances and slope between LHD and RHD revealed that compared to RHD, the relationship between CL dWMFT and IL motor capacity was significantly stronger (Fisher's z = 2.45, p = 0.01) and the slope was steeper in LHD (t = 2.03, p = 0.04).

After removing the outlying observations, contralesional dWMFT explained 62% of the variance in ipsilesional hand motor capacity in LHD (p < 0.001), and <1% in RHD (p > 0.05). The slope of this relationship changed to 0.54 ± 0.1 (p < 0.001) in LHD and 0.05 ± 0.21 (p = 0.81) in RHD. Compared to RHD, the relationship between CL dWMFT and IL motor capacity was significantly stronger (Fisher's z = 2.85, p = 0.004) and the slope was steeper in LHD (t = 2.11, p = 0.04).

Since these observations did not significantly change the strength of covariance or the slope of the relationship, they were preserved in the final multiple model. Once again, analysis of residuals did not indicate violations of necessary assumptions in multiple regression nor the presence of unduly influential observations. Nonetheless, estimates below are reported both with and without suspected outliers.

After adjusting for main effects and significant confounders using multiple regression, the final reduced form of Model 2 was statistically different from a null model (F = 7.48, p < 0.001, adjusted R² = 0.32). Based on estimates from Model 2, CL hand motor capacity was significantly associated with IL hand motor capacity (β₁ = −0.81± 0.16, p < 0.001; without outliers: 0.54 ± 0.17, p = 0.003) (Figure 3A). There was no significant effect of the side of lesion (β₂ = 0.12 ± 0.26, p = 0.66; without outliers: 0.19 ± 0.22, p = 0.39). There was an interaction between the side of lesion and CL hand motor capacity, but it only approached statistical significance (β₃ = −0.51 ± 0.34, p = 0.05; without outliers: −0.49 ± 0.24, p = 0.046). Figures 3B,C illustrates the interaction.

Finally, we note here that we conducted similar analyses for the proximal component of the WMFT (not reported) but did not observe the same relationships (model adj. R² = ~3% for both models, p > 0.25). One possibility is that unlike the distal component, the proximal WMFT is a much less sensitive metric of motor performance, especially in individuals with mild-to-moderate impairment.

Hints of this interaction were implicit in a few previous studies (2, 3, 14); however, categorical reporting of the Upper Extremity Fugl-Meyer (UEFM) masked this interesting effect. By preserving the continuity of the UEFM and utilizing continuous standardized z-scores, our statistical approach allowed a direct comparison of our regression estimates, thus reflecting effect sizes of each of the candidate predictors. Unlike previous studies, we did not observe a significant effect of the side of lesion on ipsilesional motor capacity (16, 17) or on contralesional UEFM or dWMFT (11). One reason might be that the effect of the side of lesion observed in those previous studies may have arisen from its interaction with contralesional impairment. However, because an interaction effect was not explicitly tested and because contralesional impairment was either collapsed across the groups (2, 3) or categorical (8, 14), variance in the ipsilesional capacity may have been conflated with the effect of the side of lesion, or, remained unexplained, especially for UEFM scores that fell at the boundaries of the pre-defined categories. Previously, various cut-off scores for the UEFM have been used to define impairment categories (8, 14). Of these, our data suggest that a UEFM score of 42, which occurs at the intersection of the linear fits for LHD and RHD, would best reflect the change in the direction of effect—that is, for UEFM scores <42, the ipsilesional limb would be slower in LHD compared to RHD whereas for UEFM scores >42, those with LHD would be slightly faster compared to RHD. Interestingly, at this score of 42, there would appear to be no differences in motor capacity of either hand between LHD and RHD, which might explain why a number of large clinical trials [e.g., EXCITE (26), ICARE (27)], designed for mild to moderately impaired stroke survivors (mean UEFM scores in these studies were 42.5 and 41.6, respectively) may not have observed, on average, any differences in motor capacity based on the side of stroke.

On a related note, in Model 1, we observed that the relationship between contralesional UEFM and ipsilesional motor performance was (mildly) confounded by sex. Specifically, males showed slightly faster performance times (mean difference = 0.3 s, β = −0.56, p = 0.08) compared to females. Given that this difference was very small and may have been exaggerated by the unequal sample sizes, i.e., there were about 3 times more males than females, in LHD, RHD and overall, we suspect this was an artifact of the unequal sample sizes rather than a true difference between males and females.

A common link between our study and past reports is that ipsilesional deficits were found to be most pronounced for distal (dexterous) tasks. These tasks—lift can, lift pencil, lift paper clip, stack checkers, flip cards, turn a key in a lock, fold towel, and lift basket—nearly always involve object manipulation and inherently require dexterous motor control of the hands. Sunderland et al. (4) demonstrated that early on after a stroke, spatial accuracy in dexterity tasks performed with the ipsilesional hand correlated with cognitive deficits, such as apraxia, in individuals with LHD. While individuals included in this study did not exhibit severe apraxia and were ~5 years post-stroke, it is possible that mild cognitive deficits, including apraxia, may have impacted dexterous task performance in those with LHD, especially in the more severe ranges of UEFM. Furthermore, we note that our evaluation of dexterous task performance was through timed tests, and not quality of movement or accuracy. It has been suggested that the left hemisphere plays an important role in regulating the timing and speed of movements (13), and thus, injury to the left hemisphere, such as to premotor and fronto-parietal networks (28, 29) may impair planning and sequencing required for smooth and rapid performance of dexterous motor tasks.

Our main observation that deficits in ipsilesional hand motor capacity scale with contralesional impairment only in LHD is qualitatively similar to previous clinico-behavioral [e.g., (16, 17, 30, 31)] and phenomenological evidence [e.g., (32–34)]. These findings are consistent with a rather simplified organizational model of the nervous system in which certain aspects of motor and/or cognitive control are lateralized to the left (or dominant) hemisphere, such that damage to the left hemisphere results in deficits in skilled motor actions of both upper extremities. For example, using EMG recordings of homologous muscles in the arm, Cernacek (35) demonstrated that the frequency of motor irradiations, i.e., unintended motor output in the ipsilateral hand, were significantly higher from the dominant to the non-dominant extremity. Similarly, Wyke (17) reported that while individuals with left-sided cerebral lesions exhibited bilateral motor deficits in speed and limb postural control, deficits in those with right cerebral lesions were restricted to the contralateral limb. Even within the LHD patient group, the nature and extent of ipsilesional deficits has been shown to be modulated by the degree of (clinical) paresis. For example, Haaland et al. (36) demonstrated that deficits in ipsilesional torque amplitude specification were statistically significant in LHD patients with contralesional upper extremity paresis compared to those with no paresis, despite all other features of the movement and lesion (e.g., error, speed, lesion volume) being similar.

Lastly, in one of the earliest experiments using functional MRI, Kim et al. (37) showed that the task-evoked activation of the left hemisphere was substantially greater for ipsilateral movements compared to the right hemisphere. In later years, a number of neuroimaging (30, 31) and neurophysiologic (32–34) studies have provided confirmatory evidence for the role of the dominant hemisphere in organizing motor outputs to both hands. Our results of co-varying deficits between the contralesional and ipsilesional hand in LHD provides further empirical support for the role of the left hemisphere (in our pre-morbidly right-handed group) in the control of both hands.

In interpreting our findings, it is important to consider that this study is a retrospective analysis of a relatively small dataset. Therefore, sample size, and characteristics such as the range of impairment, chronicity, age, were limited to what was available. To this point, we conducted analyses with and without outliers, and found that while exclusion of outliers affected the strength of the overall model, it did not affect the probability associated with rejecting the null hypothesis. A prospective study or independent validation in a separate cohort would be ideal, if larger samples were available. A larger sample would render more robust findings that are less sensitive to distortions from outlying values.

Along this line, UEFM scores for the RHD group were restricted toward the more severe range, with the most severely impaired individual's score being 28. This restriction, however, was less so in the LHD group (min. UEFM = 19). Although this limitation in range was circumvented by using group-wise z-scores, we are cautious in generalizing our observations regarding the interaction effect to more severe ranges of motor impairment in RHD. This is quite apparent in the variability around our estimated linear fits especially toward the extreme ranges of predictor values for RHD. Indeed, it is possible that for the more severe range in RHD, there exists a linear relationship between contralesional and ipsilesional motor deficits as illustrated in Figure 1C. Thus, while we can, with some confidence, reject the null hypothesis, our data are insufficient to differentiate between the two alternate hypotheses, and warrant a follow-up study.

Finally, it must be emphasized that the absence of a relationship with contralesional impairment in RHD should not be taken to mean that ipsilesional deficits are absent in this group. In fact, there is substantial evidence to the contrary (38–41). Comparison with an appropriate control group would be necessary to demonstrate the presence of ipsilesional deficits in RHD and the functional implications of these deficits. As alluded to earlier, measuring the speed of performance, as in the case of timed functional tasks assessed here, does not provide specific information about perceptual errors, spatial accuracy or visuomotor deficits, which, based on previous evidence (42, 43), might be a more important component of motor performance in RHD.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.