Harnessing physical activity monitoring and digital biomarkers of frailty from pendant based wearables to predict chemotherapy resilience in veterans with cancer

Abstract

This study evaluated the use of pendant-based wearables for monitoring digital biomarkers of frailty in predicting chemotherapy resilience among 27 veteran cancer patients (average age: 64.6 ± 13.4 years), undergoing bi-weekly chemotherapy. Immediately following their first day of chemotherapy cycle, participants wore a water-resistant pendant sensor for 14 days. This device tracked frailty markers like cadence (slowness), daily steps (inactivity), postural transitions (weakness), and metrics such as longest walk duration and energy expenditure (exhaustion). Participants were divided into resilient and non-resilient groups based on adverse events within 6 months post-chemotherapy, including dose reduction, treatment discontinuation, unplanned hospitalization, or death. A Chemotherapy-Resilience-Index (CRI) ranging from 0 to 1, where higher values indicate poorer resilience, was developed using regression analysis. It combined physical activity data with baseline Eastern Cooperative Oncology Group (ECOG) assessments. The protocol showed a 97% feasibility rate, with sensor metrics effectively differentiating between groups as early as day 6 post-therapy. The CRI, calculated using data up to day 6 and baseline ECOG, significantly distinguished resilient (CRI = 0.2 ± 0.27) from non-resilient (CRI = 0.7 ± 0.26) groups (p < 0.001, Cohen's d = 1.67). This confirms the potential of remote monitoring systems in tracking post-chemotherapy functional capacity changes and aiding early non-resilience detection, subject to validation in larger studies.

Figure 1

A patient wearing the PAMsys pedant sensor. Using validated algorithms different digital biomarkers of mobility including locomotion, cumulative postures, postural transitions, energy expenditure were extracted from the pendant sensor during a 2-week remote monitoring period (24h/7days) during chemotherapy.

Figure 2

Sensor output metrics for a resilient (blue) and non-resilient (red) participants. The mean physical activity parameters between the groups are shown with error bar (standard error) for 2 weeks of recorded activity. The asterisk (*) sign is used to represent significant difference in parameters for the corresponding day between resilient and non-resilient participants.

Figure 3

Area under curve (AUC) of the fitted logistic regression model for 3 different models shows the model that uses both ECOG and MBF can distinguish the resilient vs non-resilient group better than other models.

Figure 4

Chemotherapy resilience index (CRI ≥ 0.54) calculated from the MBF and ECOG can distinguish the patient with resilience significantly (p = 0.0007, effect size = 1.67).