Accuracy and reproducibility of accelerometer-assessed walking distance in heart failure patients using Mobilise-D algorithms

Abstract Background Patients with heart failure (HF) experience physical functioning limitations and symptom burdens, leading to some of the lowest levels of physical activity among chronic disease populations. Current methods of testing physical function include Cardiopulmonary Exercise Testing (CPET) and the Six-Minute Walk Test (6MWT) offer snapshots at clinic visits, but they leave room for improvement in patient relatability and clinical trial implementation. Actigraphy, utilising wearable devices with accelerometers, offers an objective method for measuring physical activity outside the clinic if technical and clinical validity requirements are met. Mobilise-D, a European public-private partnership consortium, successfully completed a thorough technical validation for digital mobility assessment in a small population of individuals with HF. However, further evaluation is needed to demonstrate scalability, particularly across the full spectrum of ejection fraction in HF patients and in a larger population. Walking distance emerged as the mobility factor most strongly linked to physical and general limitations, according to HF patients' descriptions of their daily physical activity challenges. Purpose The purpose of this study was to evaluate the accuracy and reproducibility of walking distance assessed using Mobilise-D algorithms in a broad HF (HFpEF and HFrEF) patient population. The evaluation was based on lower back accelerometer data collected from two clinical trials during a 6MWT, using the manually recorded distance as the reference. Methods Data from a total of 491 HF patients with up to 4 visits each, resulting in 1395 6MWT assessments were analysed using the Mobilise-D walking algorithm pipeline and compared to the distance assessed manually during the same 6MWT as the reference standard. Within-subject variability of the error was also analysed. Results The Mean Absolute Percentage Error (MAPE) between the calculated walking distances using Mobilise-D and the manual test was 11.24% (Figure 1). The mean within-subject variability of the error across visits was 5.66% (Figure 2). Conclusion The Mobilise-D algorithm demonstrated strong performance in a diverse population of HF patients. Some overestimation is expected as, unlike the manual 6MWT, the Mobilise-D calculation includes the distance walked during turns in the 6MWT. One limitation is that sensor height is needed as input for Mobilise-D algorithm, and it was not directly available in our study. Our findings support further investigation into using Mobilise-D algorithms to assess walking distance and other related walking metrics in HF patients.