Fig. 1: Skin temperature and thermoception results.

In the left panel, the grey-shaded boxplots display the basal skin temperature (measured in Celsius) distribution across the three different groups of patients, separately by sides (left in light green vs. right in dark green). The central line indicates the median value (50th percentile), while the bounds of the box represent the 25th (lower bound) and 75th percentiles (upper bound), i.e., interquartile range (IQR). The whiskers extend to the minimum and maximum values within 1.5 times the IQR. Jittered points represent the individual data points, while the estimated marginal means for each group and side are shown as larger bold dots (circles for the left side, triangles for the right side), with error bars representing 95% confidence intervals. Solid lines indicate the left side, and dotted lines indicate the right side. To investigate whether the baseline skin temperature of the limb differed between the groups based on the subjective experience of disownership, we performed a General Linear Model with Group (HP+ DSO+ (n = 9), HP+ DSO− (n = 10), HP− DSO− (n = 21)) and Side (left, right) as fixed factors. The temperature was modelled as the dependent variable. Sex and proprioception were included as covariates in the final model. For the hands’ temperature, results showed a main effect of Group (F_(2,79) = 6.126, p = 0.003; η²_p = 0.145). Post hoc Bonferroni-corrected comparisons indicated a difference between HP+ DSO+ and HP+ DSO− (p = 0.007) and between HP+ DSO+ and HP− DSO− (p = 0.005), and no difference between HP+ DSO− and HP− DSO− (p > 0.05). There was no effect of Side (F_(1,79) = 0.583, p = 0.45; η²_p = 0.008) nor a Group by Side interaction (F_(2,79) = 0.093, p = 0.911; η²_p = 0.003). To address the unbalanced group sample size and verify the robustness of the results, non-parametric bootstrap resampling based on distribution’s quintiles with 5000 replicates stratified by group was employed to estimate 95% bootstrap confidence intervals for post hoc pairwise comparisons. Results showed statistically significant differences between HP+ DSO+ and HP+ DSO− (estimate = 1.73, 95% CI [0.71;2.76], p_adj = 0.001) and between HP+ DSO+ and HP− DSO− (estimate = 2.162, 95% CI [0.99;3.40], p_adj = 0.004), and no difference between HP+ DSO− and HP− DSO− (estimate = 0.43, 95% CI [−0.56;1.38], p_adj > 0.05). The right panel shows the average number of stimuli detected in the thermoception task by the three groups of patients, separately by left (light green) and right (dark green) hand. The bold dots represent the estimated marginal means for each group and side, with error bars denoting 95% confidence intervals. Solid lines indicate the left side, and dotted lines indicate the right side. To investigate whether DSO was associated with reduced hand thermoceptive ability, we ran a Generalized Linear Model with Group (HP+ DSO+ (n = 8), HP+ DSO− (n = 7), HP− DSO− (n = 19)), Side (left, right), and Stimulus type (warm, cold) as fixed factors. The thermoception score (ranging from 0 to 3) was modelled as the dependent variable. We adopted a Poisson distribution with a square root link function to optimise the model stability. The results showed a main effect of Group (x²₍₂₎ = 15.993, p < 0.001), indicating that HP+ DSO+ performed worse than the other two groups. There was a main effect of Side (x²₍₁₎ = 21.789, p < 0.001), with an overall lower temperature detection on the left hand. We also found a Group by Side interaction (x²₍₂₎ = 30.880, p < 0.001). Bonferroni-corrected post hoc comparisons showed that HP+ DSO+ were less accurate in perceiving thermal stimuli on the left hand compared to HP+ DSO− (p = 0.008) and HP− DSO− (p < 0.001). Similarly, HP+ DSO− showed less accuracy on the left hand than HP− DSO− (p = 0.009). Furthermore, both HP+ DSO+ (p < 0.001) and HP+ DSO− (p = 0.001) showed a left-right side difference in thermoception, performing worse with the left hand as compared with the right hand, while HP− DSO− did not (p > 0.05). The main effect of Stimulus (cold vs. warm stimuli) was not significant (x²₍₁₎ = 2.728, p = 0.099), as well as the Group by Stimulus (x²₍₂₎ = 0.388, p = 0.824), Side by Stimulus (x²₍₁₎ = 0.121, p = 0.727), and Group by Side by Stimulus (x²₍₂₎ = 0.111, p = 0.946) interactions. Extrapersonal USN was the only covariate that was significant (p = 0.005). To address the unbalanced group sample size and verify the robustness of the results, non-parametric bootstrap resampling based on distribution’s quintiles with 5000 replicates stratified by group was employed to estimate 95% bootstrap confidence intervals for planned post hoc pairwise comparisons. Results from bootstrap resampling showed statistically significant differences for thermoception on the left hand between HP+ DSO+ and HP+ DSO− (estimate = 0.612, 95% CI [0.178;0.921], p_adj = 0.04), HP+ DSO− and HP− DSO− (estimate = 0.522, 95% CI [0.222;0.919], p_adj = 0.008), and between HP+ DSO+ and HP− DSO− (estimate = 1.133, 95% CI [0.835;1.368], p_adj < 0.001). All analyses were two-sided. HP hemiplegia, present (+) or absent (−), DSO Disturbed Sensation of Ownership, present (+) or absent (−), HP+ DSO+ patients with hemiplegia and disturbances in the sense of body ownership, HP+ DSO− patients with hemiplegia and without disturbances in the sense of body ownership, HP− DSO− patients without hemiplegia or disturbances in the sense of body ownership, USN unilateral spatial neglect. Source data are provided as a Source Data file.