Correlation between step count and clinical outcomes for patients with cancer during treatment: A feasibility study.

Abstract

e18804 Background: Step count, as measured by a wearable accelerometer, has been shown to have a relationship to premature death, cardiovascular events, functional decline, longer stay, and higher rehospitalization rates. However, few cancer studies or trials have incorporated accelerometers to measure response to active treatment. We developed the DigiBioMarC™ smartphone application for cancer patients to enable participation in decentralized clinical trials and remote cancer care by collecting informed consent, ePROs and accelerometer data using an Apple Watch. This analysis assessed whether daily step data were associated with participants clinical events. Methods: We tested the feasibility of the DigiBioMarC application along with the Apple Watch for approximately 4 weeks with 50 cancer patients undergoing IV chemotherapy or immunotherapy recruited in a fully decentralized study through Kaiser Permanente Northern California. Participants used the app for at least 28 days and were provided with an Apple Watch if they did not already have one. Data pre-processing was performed to identify periods of missing data and non-wear time. Step count was calculated for each calendar day and days that included at least ten hours of wear time while awake were considered sufficient and included in the analysis. Specific clinical events were collected from the patients’ electronic health records (EHR) up to six months following the study. Results: Thirteen participants experienced at least one clinical event, and there were 5 deaths. Using Cox regression, patients with more sufficient days were less likely to die during follow up (p = 0.122) than patients with fewer sufficient days. On sufficient days, median daily steps < = 2,510 were associated with one or more adverse clinical events, while daily steps > 2,510 were associated with no clinical events and had a longer time to adverse clinical event (p = 0.068) compared to those with less than or equal to 2,510 median daily steps on sufficient days. Daily median step count on sufficient days predicted clinical event occurrence with an accuracy of 0.833. Conclusions: Findings from this feasibility study support the hypothesis that daily stepping behavior is a valid real-world digital measure to predict clinical events in patients undergoing cancer treatment. Although the predictive models did not reach statistical significance (p < 0.05), this is likely due to the low frequency of clinical events in the dataset. These findings indicate that future investigations with larger sample sizes are warranted as this may be a beneficial tool for decentralized trials or care when patients have longer periods of time between clinical visits. In patients undergoing cancer treatment, real-world based step data extracted from wearables can provide early indication of poor or declining health.