CardiacProfileR: an R package for extraction and visualisation of heart rate profiles from wearable fitness trackers.

Abstract

A heart rate profile is constructed from continuous measurement of heart rate before, during and after exercise, typically in the context of an exercise stress test. It contains important clinical parameters such as resting heart rate, maximal heart rate during exercise, and decrease of heart rate in the recovery phase. An increase in resting heart rate, a decrease in heart rate elevation in response to exercise, and a delay in heart rate recovery are significant predictors of sudden death (Cole et al. 1999; Jouven et al. 2005). These measurements, when derived from a controlled exercise stress test regime, have high intra-individual reproducibility over 3 years, indicating their usefulness as a diagnostic tool (Orini et al. 2017). Heart rate parameters and profiles have been measured and constructed with exercise stress tests under controlled environments. This is often performed with a treadmill in a laboratory setting. With increasingly widespread availability of wearable wrist-worn fitness trackers, such as Fitbit and Apple Watch, we hypothesise that we can construct heart rate profiles for an individual using heart rate and physical activity (e.g. as measured by global positioning system and accelerator) data from a person’s wearable monitor. The main advantage of this approach is that it opens unprecedented opportunity to continuously and non-invasively monitor a person’s heart rate profile under free-living conditions, which would better match this person’s realistic activity profile. These devices are socially acceptable, generally inexpensive, and are already widely available in many communities. This potential cardiac function monitoring technology can be seamlessly incorporated into a user’s daily life, without the need for clinically based exercise stress testing. A number of recent studies have assessed the accuracy of popular wrist-worn wearable fitness trackers (Dooley et al. 2017; Shcherbina et al. 2017). They found that while energy expenditure estimations are often less accurate, the measurements of heart rate are generally considered accurate. These findings suggest that it should be possible to extract heart rate dynamics information before, during, and after an exercise event, which can be easily defined without accurate measurement of energy expenditure. Here, we present a new open source R package called CardiacProfileR. It is designed to efficiently extract heart rate and physical activity data from a Training Center XML (TCX) file which is generated by the most modern fitness tracking devices. CardiacProfileR can identify periods of active exercise from the data, and construct a heart rate profile for each period of active exercise. This package produces interactive visualisation of one or multiple heart rate profiles in two dimensions or three dimensions. This enables aggregation and comparison of data from multiple periods of exercise or comparison between multiple people. As far as we know, CardiacProfileR is the first open source R package that provides an end-to-end analysis framework for wearable fitness sensor data. CardiacProfileR is available via an MIT license at https://github.com/VCCRI/CardiacProfileR.