Modeling Personalized Heart Rate Response to Exercise and Environmental Factors with Wearables Data

This repository contains the code associated with the research paper Modeling Personalized Heart Rate Response to Exercise and Environmental Factors with Wearables Data (Nazaret et al. NPJ Digital Medicine, 2023). Link forthcoming.

Abstract

Heart rate (HR) response to workout intensity reflects fitness and cardiorespiratory health. Physiological models have been developed to describe such heart rate dynamics and characterize cardiorespiratory fitness. However, these models have been limited to small studies in controlled lab environments and are challenging to apply to noisy — but ubiquitous — data from wearables. We propose a hybrid approach that combines a physiological model with flexible neural network components to learn a personalized, multidimensional representation of fitness. The physiological model describes the evolution of heart rate during exercise using ordinary differential equations (ODEs). ODE parameters are dynamically derived via a neural network connecting personalized representations to external environmental factors, from area topography to weather and instantaneous workout intensity. Our approach efficiently fits the hybrid model to a large set of 270,707 workouts collected from wearables of 7,465 users from the Apple Heart and Movement Study. The resulting model produces fitness representations that accurately predict full HR response to exercise intensity in future workouts, with a per workout median error of 6.1 BPM [4.4 - 8.8 IQR]. We further demonstrate that the learned representations correlate with traditional metrics of cardiorespiratory fitness, such as VO2 max (explained variance 0.81 ± 0.003). Lastly, we illustrate how our model is naturally interpretable and explicitly describes the effects of environmental factors such as temperature and humidity on heart rate, e.g., high temperatures can increase heart rate by 10%. Combining physiological ODEs with flexible neural networks can yield interpretable, robust, and expressive models for health applications.

Getting Started

These instructions will help you get a copy of the hybrid ODE model up and running on your local machine for development and testing purposes. See the Usage section for notes on how to use the code.

Installation

Clone this repository: git clone https://github.com/Apple/repo-name.git
Change to the project directory: cd repo-name
Install the required packages: pip install -r requirements.txt

Dataset

The Apple Heart and Movement Study dataset is not publicly available for privacy reasons. Users can instead use their own data or other publicly available datasets.

For example, the paper Modeling heart rate and activity data for personalized fitness recommendation [1] provides a dataset of user sport records collected from wearables One can use the examples/preprocess.py script to produce the data format required for our algorithm, e.g.,

python examples/preprocess.py --input_path /path/to/dataset_folder --output_path /output/path

which will generate a file endomondo.feather in the output folder.

Code Structure

The code is organized as follows:

.
├── ode
│   ├── data.py                 # Data loading and preprocessing
│   ├── modules_cnn.py          # Modules for the CNN encoder
│   ├── modules_dense_nn.py     # Modules for all scalar dense NNs
│   ├── ode.py                  # Hybrid ODE model
│   └── trainer.py              # Trainer for the ODE model
├── examples
│   ├── preprocess.py            # Script to preprocess the endomondo dataset
│   ├── plotting.py             # Helper functions to plot the data
│   └── train_ode_model.ipynb   # Example notebook to train the ODE model
├── readme.md                   # This file
└── requirements.txt            # Required packages

Usage

Step 1: Prepare the data. The WorkoutDataset class in ode/data.py loads the data from a pandas dataframe and preprocesses it, according to the WorkoutDatasetConfig class. In the current implementation, all the measurements (HR, speed ...) are assumed to be sampled on a uniformly sampled time grid (that we can obtain by interpolating the data). The following data is required in the dataframe:
- the user id that performed the workout
- the id of the workout
- the time grid on which the heart rate and the activity are measured
- a list of heart rate measurements
- a list of heart rate measurements normalized in a reasonable range for neural networks
- one column per activity measurements (e.g. list of horizontal speeds, vertical speed, etc.)
- [optional] one column per weather measurement (e.g. one temperature, humidity, etc.)
Step 2: Define the ODE model.
- The ODEConfig class specifies the hyperparameters of the ODEModel.
Step 3: Train the ODE model.
- The function train_ode_model in ode/trainer.py trains the ODEModel.

An example is provided in examples/train_ode_model.ipynb.

Authors

Achille Nazaret - Columbia University (Work done while at Apple)
Sana Tonekaboni - University of Toronto (Work done while at Apple)
Greg Darnell - Apple
Shirley Ren - Apple
Guillermo Sapiro - Apple
Andrew C. Miller - Apple

Citation

If you use this code or dataset in your research or publications, please cite the original research paper:

Nazaret, A., Tonekaboni, S., Darnell, G., Ren, S., Sapiro, G., & Miller, A.  Modeling
Personalized Heart Rate Response to Exercise and Environmental Factors with Wearables
Data.  NPJ Digital Medicine, 2023.

Bibtex entry

@article{nazaret2023modeling,
  title={Modeling Personalized Heart Rate Response to Exercise and Environmental Factors with Wearables Data},
  author={Nazaret, Achille and Tonekaboni, Sana and Darnell, Gregory and Ren, Shirley and Sapiro, Guillermo and Miller, Andrew C.},
  journal={NPJ Digital Medicine},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

References

[1] Ni, Jianmo, Larry Muhlstein, and Julian McAuley. "Modeling heart rate and activity data for personalized fitness recommendation." The World Wide Web Conference. 2019.

suricforever / ml-heart-rate-models