Objective physical activity measures to predict hospitalizations among patients with cancer receiving chemoradiotherapy.

Abstract

592 Background: There is an unmet need for measures that predict adverse events during cancer therapy. Previous studies have demonstrated feasibility and value of collecting physical activity (PA) data from cancer patients undergoing curative-intent chemoradiotherapy using a commercial fitness tracker. Here, we compare the predictive value of objective PA measures versus standard clinical information and assess whether additional PA measures beyond daily step count help predict hospitalization events during treatment. Methods: Cancer patients enrolled in NCT03115398 (2017-2020) wore a Garmin VivoFit continuously during chemoradiotherapy for a variety of solid tumors. ECOG performance status (PS) was assessed at baseline. PA measures tabulated for each day included total step count, peak 30-minute cadence (average step rate over the 30 most active minutes [not necessarily consecutive]), and sedentary time. Baseline PA measures were defined as mean values from up to 14 days prior to start of chemoradiotherapy. Cox regression models with time-dependent (3-day rolling averages of PA measures and recent EORTC QLQ-C30 quality of life [QoL] score) and time-fixed (baseline PA variables, age, sex, cancer diagnosis) covariates were used to identify predictors of hospitalization. The incremental predictive value of PA measures was assessed using logistic regression models and five-fold cross validation. Results: One hundred thirty-eight patients were included in the analysis. The most common cancer diagnoses were gastrointestinal cancer (29%) and head and neck cancer (28%). Twenty-four patients (17%) were hospitalized during chemoradiotherapy. Univariate Cox regression models identified baseline step count (hazard ratio [HR] = 0.85 per 1000 steps, 95% confidence interval [CI]: 0.74-0.98, p = 0.03) and baseline peak 30-minute cadence (HR = 0.86 per 10 steps/min, 95% CI: 0.74-0.99, p = 0.03) as the most significant prognostic predictors of hospitalization. Neither baseline PS nor baseline sedentary time were significant predictors of hospitalization risk. Including both baseline step count and peak 30-minute cadence in multivariable models did not improve predictive power. During treatment, peak 30-minute cadence was identified as the most powerful dynamic predictor of short-term hospitalization risk (HR = 0.83 per 10 steps/min, 95% CI: 0.75-0.92, p < 0.001). Conclusions: This study examined the relationship between a suite of PA measures and risk of hospitalization during concurrent chemoradiotherapy. Our results suggest that before treatment, daily step count average can serve as a simple and effective metric for identifying high-risk patients. During treatment, measures of peak activity may have additional utility for identifying patients in need of enhanced supportive care.