How to Measure Mobility in the Real World
- Michael McMahon
- Feb 13
- 4 min read
Updated: Apr 15
Note: This is an abridged version our white paper on ‘How to Measure Mobility in the Real World’. To see the full paper with additional results, comprehensive references and more detail on the methodology, please download the paper on our website here.
Measuring Real-World Mobility with the Mobilise-D Method
The Mobilise-D project represented a major step forward in the generation of validated, ecologically relevant and regulatory accepted mobility data. It produced the first simple, unobtrusive and low-cost technologies and methodologies for the measurement of mobility in everyday life that underwent a rigorous process of technical and clinical validation in real world-settings for multiple patient cohorts.
The figure below provides an overview of the Mobilise-D method. This article will focus on the generation of the bout level data, with a future article covering the data at the daily and weekly levels.
The digital mobility outcomes (DMOs) generated by the Mobilise-D method are distinguished from other forms of mobility assessment because they provide insight on mobility performance in the real world rather than mobility capacity in a lab setting. This is an important distinction, as results from supervised tests can be strikingly different to unsupervised tests on the same patient population due to white coat effects (see Warmerdam et al. (2020), for example).
Data is generated from a single sensor placed on the lower back and worn for up to seven days. Feedback from the patients on the comfort of the device was very positive and was evident in the extremely high levels of adherence. The clinical validation study involved 2,366 participants, and an assessment was obtained from 94% of those participants*. We therefore have strong evidence for the feasibility of a digital mobility assessment under the Mobilise-D protocols.
Generating the Bout Level Data
The foundation of the DMOs is a series of gait features (box 1 below) identified from the raw sensor data by a suite of algorithms selected from a comprehensive technical evaluation process (see Micó-Amigo et al., 2023). It is important to note that the algorithms identifying these gait features were tested in real-world conditions as well as in the lab.
The algorithmic pipeline identifies individual walking bouts and evaluates DMOs (such as walking speed) from these walking bouts (box 2 above). The Mobilise-D technical validation study was the most extensive validation of a complex comprehensive multi-stage analytical pipeline for estimation of walking speed from a single wearable device. Participants from 6 cohorts** were monitored in the laboratory and the real-world (2.5 h) using the wearable sensor positioned on the lower back (see Kirk et al. (2024). Results were compared to a reference system that was itself previously validated relative to a gold standard motion capture system.
The study found that in a lab setting the mean absolute error (MAE) for walking speed derived from all walking bouts was 0.10 m/s, which corresponded to a mean absolute relative error of (MARE) of 14.96%. The MAE ranged from 0.06 to 0.12 m/s depending on the cohort, with the MARE ranging from 7.82% to 22.19%. The intraclass correlation coefficient was 0.84 (good) for all walking bouts and ranged from good (0.79) to excellent (0.91) depending on the cohort.
As we can see from the table, errors in real-world testing are similar to lab based tests despite the additional complexity of the real world. This clearly shows the feasibility of mobility assessment in the real world using the Mobilise-D technologies and protocols.
Bout Level Data Processing and Day / Week Level DMOs
As the relevant studies describing bout level data processing and the generation of the day / week level DMOs have not been released yet, we will only briefly describe them here.
Bout Level Data Processing. After generating DMOs at the walking bout level, the next step is to construct DMOs at the daily or weekly level using novel methodologies developed by the Mobilise-D consortium. Any invalid days or unreliable weeks are removed, before the data is aggregated to the daily or weekly level. This aggregation is carried out by applying restrictions (e.g. only including walking bouts > 60 seconds) and a summary statistic (such as a mean, median of sum) to the bout level data for the period in question.
Daily / Weekly DMOs. The process of cleaning, applying restrictions and aggregating the data produces DMOs at the daily and weekly levels. An example would be mean walking speed for walking bouts of greater than 60 seconds over a 7-day period. There is a total of 24 DMOs across five mobility domains, with examples shown in box 3 below.
Conclusion
The Mobilise-D project has brought about a step-change in our ability to measure and understand mobility performance in everyday life. It produced the first technologies and methodologies for the measurement of real-world mobility that have undergone a rigorous process of technical and clinical validation. These studies represent the first steps on the path towards regulatory approval of real-world mobility measures, with the Mobilise-D project receiving two letters of support from the EMA. The Mobilise-D project has therefore overcome a major barrier to the use of DMOs such walking speed in interventional clinical trials.
The team at Enoda were central to the successful delivery of Mobilise-D and can provide a complete mobility data solution, including sensors, algorithms, electronic data capture system and reporting functionality. Our unrivalled team of mobility experts is ready to help you realise the many benefits that the Mobilise-D methods and technologies can bring to clinical trials and clinical care.
Email info@enoda-health.com to find out more.
* See the presentation by Judith Garcia-Aymerich at the Mobilise-D Conference in March 2024 (Edinburgh). https://www.youtube.com/watch?v=M-VUIApEGM0&t=1s
** Parkinson’s Disease, Multiple Sclerosis, Proximal Femoral Fracture, Chronic Obstructive Pulmonary Disease, Congestive Heart Failure and healthy older adults (n = 97).
