**4. Discussion**

The objective of this work was to evaluate the validity of the estimations of walked distance by wearable devices in individuals with stroke sequelae. We observed that the best estimations over a 6-min walk at comfortable speed were provided by the pedometer worn at the hip on the non-affected side (MB = 9.7%, RMSE = 10.9%) and by the Actigraph worn at the ankle on the non-affected side (MB = 10.7%, RMSE = 14.6%).

We observed significant differences between the combinations of sensor type and position, which demonstrates the impact of these parameters. For instance, despite being placed in the same location (hip on the non-affected side), the pedometer provided a better estimation of the walked distance than the Actigraph, even though the first is a piezoelectric device and the second is a triaxial accelerometer. On the one hand, the Actigraph's measurement error was considerable (MB = 101.8 ± 60.1 m; RMSE = 60.3 m) compared to the pedometer (MB = 20.5 ± 24.6 m; RMSE = 23.1 m). This could potentially be explained by an issue in the settings of the Actigraph or in its algorithm. The chosen settings were standard, i.e., a standard sampling frequency (30 Hz) and no specific filtering. It is possible that the use of other algorithms or other settings may alter the accuracy of the Actigraph GT3X. The manufacturer of this device recently published an add-on called "low-frequency filter extension" which can be enabled for healthy individuals with low amounts of physical activity [34]. This add-on lowers the detection threshold of the Actigraph to improve the data acquisition sensitivity [34,38]. It would be relevant to evaluate the impact of these settings on the estimation of the subject's walked distance, especially in individuals with limited walking capabilities.

On the other hand, the pedometer might have a lower detection threshold, as it is specifically designed to count the number of steps of an individual. This would, therefore, induce a better sensitivity in this population. However, we are unable to confirm these hypotheses since we could not analyze the raw data used by the sensors. Like the Actigraph, the Sensewear Armband had a

consequential measurement error despite being a multisensor validated for the assessment of energy expenditure in post-stroke populations, with an MB of 127.3 ± 79.8 m and RMSE = 65 m when placed on the affected side, and an MB of 120.6 m ± 83.8 m and RMSE = 80 m when placed on the unaffected side. These results are consistent with those reported by Manns et al., who observed an error of 193.1 ± 168.1 steps in two six-minute walk tests in a sample of 12 post-stroke subjects [30]. Similar results were also reported by Vanroy et al., whose study reported an estimation error between 110 and 190 steps using the Sensewear Armband after walking 120 m at a comfortable speed in a group of 14 post-stroke subjects [39]. Thus, this device does not seem reliable to estimate the number of steps in post-stroke subjects. The cause may be the device's algorithm, which might fail to correctly count the number of steps, but the impossibility of accessing the raw data prevented us from confirming this hypothesis. In post-stroke populations, we would recommend limiting the use of this sensor to its main function, i.e., assessing the energy expenditure.

The sensor's position on the body also had a significant impact, as proven by the differences in the results of the Actigraph between the ankle, hip, and wrist. The estimation of the Actigraph worn at the ankle was the closest to the measured walked distance (MB = 22.6 ± 32.4 m), while the same device worn at the hip or on the wrist had MB values over 95 m (Table 3). It is possible that placing the Actigraph on the ankle provides a better exposure to the accelerations of the limb, which would enable a better acquisition of the number of steps. This hypothesis is supported by Klassen et al., who demonstrated that placing a Fitbit triaxial accelerometer on the ankle provided a lower estimation error, from 84.6% ± 30.5% when placed at the waist to 15.8% ± 22.3% when placed on the ankle in a study including 43 post-stroke subjects [21].
