Accurate capture of real-world gait speed in frail, older adults: insights into performance and behavior from a longitudinal clinical trial (Data for independent validation study)

Digital technologies and advanced analytics have drastically improved our ability to capture and interpret health relevant data from patients. However, to date, limited data and results have been published detailing real-world patient compliance, demonstrating accuracy in target indications or examining what novel insights and clinical value can be derived. Here we present novel, digital mobility data from two studies: an independent, non-interventional validation study with elderly, naturally slow walking subjects, and a global, multi-site phase IIb clinical trial involving patients with age-related muscle loss and slow walking speed (sarcopenia). Based on these data, we validate the accuracy of a novel algorithm for capturing in-clinic and real-world gait speed in frail, slow-walking adults. We demonstrate the feasibility of continuous monitoring with a wearable inertial sensor in elderly adults in real-world settings, and propose minimum thresholds for compliance required for robust capture of gait behaviors in this population. We also show how simple, inferred contextual information, describing the length of a given walking bout, can explain some of the variation in real-world gait speed, and use this information to demonstrate for the first time a relationship between in-clinic performance and real-world gait speed behavior. This work lays a foundation for exploration of the clinical relevance and value of such measures and is a first step in building a more complete chain of evidence between standardized physical performance assessment, real-world behavior, and subjective perceptions of mobility, independence and health. This dataset contains data collected during the independent validation study: derived data from raw accelerometry data, and summary performance data. The full dataset, including raw accelerometry data, is available here: https://s3.console.aws.amazon.com/s3/buckets/mueller-et-al-2019Contact/correspondence for data set: Arne Mueller (arne.mueller@novartis.com) Access/URLs: • Independent validation study o Derived data and metadata: 10.5281/zenodo.2841298 • Interventional clinical trial o Derived and metadata: 10.5281/zenodo.2846014 • Both studies o Full datast: https://s3.console.aws.amazon.com/s3/buckets/mueller-et-al-2019 DOI for publication: TBC Derived data descriptions annotation.csv: Section/parcours and event annotation (se above for details) 1. SubjectId: subject Id 2. RunId: run id 3. Annotation: annotation (see text above for possible annotations) 4. AnnotationType: either 'Section' or 'Event' 5. Comment: optional comment for the annotation 6. start: timestamp of start of the annotation (UTC) 7. end: timestamp of end of the annotation steps.csv: one row per step as reported by the step algorithm 1. subjectId: subject Id 2. runId: run Id 3. running: is this a running step (logicl) 4. speed: step speed in m/sec 5. duration: duration of the step in sec 6. distance: distance of the step im meters 7. side: left or right step (logical) 8. bout: the bout identifier this step belongs to (in a continuous series of steps) 9. freq: indowed step frequency 10. walk.ratio: windowed walking ratio 11. start: timestamp of the heel strike (UTC) 12. end: start + duration of the step (e.g. from heel strike of one foot to heel strike of the other foot) wheel.csv: calculated wheel speed per seconds (reference speed) 1. SubjectId: subject id 2. RunId: run id 3. speed: average wheel speed during this second 4. start: start of a recording chunk (one second apart from the next chunk) 5. mid: mid timestamp of the chunk (usually 0.5 sec) sectionSpeed.csv: speed summaries per subject, run and section 1. SubjectId: subject id 2. RunId: run id 3. Annotation: section annotation 4. subj.speed.sd: speed standard deviation per row 5. wheel.speed.sd: standard deviation of the reference speed for this subject, run and section 6. subj.speed: median speed for the subejct as calculated by the algorithm 7. wheel.speed: median reference speed calculated by the measurement wheel (within this seconds) 8. steps: number of steps in this section 9. rollator: did this subject use rollator for this run (logical)? Raw data and related to raw data HDF5, tar.gzfile, sync, sync.csv: This table reports estimated clock-offsets between the two sensors and the video camera and can be used to correct timestamps for synchronization. This file also links to the raw data recordings in the HDF5 file. 1. SubjectId: subject id 2. RunId: run id 3. wheel.tap.idx: the index (0.01 sec) at which the "tapping" event was detect in the wheel sensor 4. subject.delta: subject tapping offset to the wheel as index (or units of 0.01 sec), can be used to sync clock scew between wheel and subject sensors 5. subject.path: path/key to the raw data in the HDF file 6. wheel.path: path/key to the raw data in the HDF file 7. subject.start: start time stamp of subject sensor in UTC 8. subject.end: end time stamp of subject sensor in UTC 9. subject.duration: duration of subject recording in seconds 10. wheel.start: start time stamp of wheel sensor in UTC 11. wheel.end: end time stamp of wheel sensor in UTC Raw data Each UUID (see meta.csv) is an HDF5 file with each file recording for a dataset for which the x, y and z acceleration is stored as well as the KSS with 100 Hz. To get acceleration gravity units the x, y or z value needs to be divided by 336. The KSS indicates whether the belt buckle was closed (1) or open (0). Raw data is available here: https://s3.console.aws.amazon.com/s3/buckets/mueller-et-al-201