From high-resolution time series to a single, clinically-interpretable value - considerations for the aggregation of real world walking speed assessed by wearable sensors in patients with chronic obstructive pulmonary disease (COPD)

Le résumé fourni par la source

Introduction: Using wearables to measure real-world walking speed (RWS) at high sampling rates poses a challenge when aggregating measures into a single interpretable value. Aim: Identify the optimal aggregation of daily RWS in patients with COPD. Methods: RWS was assessed in 18 COPD patients and 15 healthy controls over 7 days with the DynaPort 7 accelerometer. We assessed within-patient day-to-day stability of RWS with intraclass correlation coefficients (ICC) of various aggregation approaches (i.e., different summary statistics applied to different levels of restrictions [Walking bouts (WB) of 10-30s; WB>10seconds (s), WB 30-60s, WB>30s, and WB>60s]). Best discriminatory ability in RWS between COPD patients and healthy controls was determined using area under the curve (AUC). Results: ICCs and AUCs for central, extreme, and variability measures of RWS are presented in Fig 1. For central measures, highest ICC was observed for the daily median of WBs of 10-30s, and highest AUC for (trimmed) mean with WBs>30s. Maximal values for extreme, and closed restrictions for variability measures were removed as conceptually not useful. Best ICC and AUC were found for 90th percentile of WB>30s. Highest ICC and AUC for variability measures were found for standard deviation of WBs>10s. Conclusion: In COPD, RWS with WB>30s using (trimmed)mean and 90th percentile results in the best combination of within-patient day-to-day stability and discriminatory ability to healthy controls for central and extreme measures. Standard deviation for WB>10s best aggregates variability.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.