*3.2. Validity of the Analysis*

Across all recording devices, the Actigraph data of one patient were lost and six armband devices (17% of armbands) had a recording failure. The reasons for these recording issues were not found.

The mean walked distance measured by the examiner at the end of the test was 208.2 ± 109.2 m. No significant differences were found between the distance measured by the examiner and the following devices: the Actigraph worn on the unaffected ankle (d = 185.7 ± 100.6 m; *p* = 0.37), the Actigraph worn on the affected ankle (d = 175.7 m ± 109.3 m; *p* = 0.21), the pedometer worn around the neck (d = 183.4 m ± 147.2 m; *p* = 0.32), and the pedometer worn at the non-affected hip (d = 231.1 m ± 121.2 m; *p* = 0.46). For all other combinations of sensors and locations, there was a significant measurement error with *p* < 0.001 (Figure 2). The step count of each device is illustrated in Table 2.

**Figure 2.** Distance measured by the practitioner and estimated from step counts reported by all devices according to type and placement. nH: non hemiparetic side; H: hemiparetic side. \* *p* < 0.05 at the end of the Wilcoxon test comparing the distance measured by the examiner and that estimated by the device from the number of steps.


**Table 2.** Step count reported by the devices. nH: non hemiparetic side; H: hemiparetic side.

## *3.3. Validity Parameters*

The parameters of validity of each sensor are summarized in Table 3. The most accurate estimations were obtained using the pedometer worn at the hip on the non-affected side (MB = 9.7%, RMSE = 10.9%) and the Actigraph placed at the ankle on the non-affected side (MB = 10.7%, RMSE = 14.6%). On the other hand, the pedometer worn on the hip and the Actigraph worn on the ankle on the affected side showed the best coefficients of correlation (*r* > 0.90) and the lowest limits of agreement. The Actigraph worn on the ankle on the unaffected side had lower correlation (*r* = 0.93) and higher 95% LoA (111.4; −46.4 m) compared to the same device located on the affected limb, even though no statistical difference was observed between the two estimations.

**Table 3.** Validity parameters of distance estimated by wearable devices versus distance measured by examiner. Unit in meters; nH: non hemiparetic side; H: hemiparetic side; percentage difference: mean bias expressed in percentage of distance measured and estimated by device; 95% LoA: limits of agreement of Bland–Altman analysis; *r* = Pearson correlation coefficient; *p* = statistical significance of Pearson correlation coefficient; RMSE: root-mean-square error; percentage RMSE: RMSE expressed in percentage of distance measured by examiner.

