A Functional Data Analysis Framework For Modeling Physical Activity Intensity Continuously Using Accelerometer Data

Abstract

PURPOSE: Cutpoints approaches are often used to define physical activity (PA) intensity categories for accelerometry. We proposed methods to characterize the full continuum of PA intensity, free of cutpoints, and to quantify associations between PA accumulated at varying intensity levels and health outcomes. METHODS: During 2012-2014, 6,379 women aged 63-99 wore accelerometers on their waist for 7 consecutive days. Accelerometer counts data were analyzed in 15-second epochs. Complementary cumulative distribution functions (CCDF) were used to characterize how each subject distributes PA across all intensity levels. Functional linear regression models were used to estimate flexible dose-response relationships between PA intensity and health outcomes, including body mass index (BMI) and fasting glucose levels. RESULTS: The CCDFs of PA intensity showed that age-related decline in PA occurred at all intensity levels. The dose-response relationships between PA intensity and BMI and glucose were shown in Figure 1. Neither effects were constant, indicating that intensity mattered after controlling for volume (count-min or MET-min). The relationship between BMI and intensity was complicated, while PA accumulated at higher intensity was found to be associated with lower fasting glucose levels, after controlling for volume. CONCLUSION: We provided a cutpoint-free analytic framework to model PA intensity continuously. The effect of PA on BMI and fasting glucose varied across intensity levels, even when fixing PA volume. The proposed methods are applicable not only to counts data, but also to other acceleration-based metrics calculated using raw data.Figure 1: Association of PA intensity (counts/15-sec or equivalent METs) with BMI and fasting glucose. The adjusted analysis included age, race-ethnicity and education for BMI, and for glucose additionally adjusted for BMI. Corresponding METs were calculated based on an internal calibration study.