wearablevar is a Python package that calculates wearable variability metrics from longitudinal wearable sensors. These features can be used as part of your feature engineering process.

Traditional Human Activity Recognition (HAR) utilizes accelerometry (movement) data to classify activities. This summer, Team #4 examined using physiological sensors to improve HAR accuracy and generalizability. The team developed ML models that are going to be available open source in the Digital Biomarker Discovery Pipeline (DBDP) to enable other researchers and clinicians to make useful insights in the field of HAR. In sum, the goal of the Human Activity Recognition Team is to create a predictive model that: Takes in multimodal data from mechanical sensors (such as accelerometers) and physiological sensors (such as electrodermal sensors and pulse oximeters). Classifies human activity (Rest, Deep Breathing, Walking, Typing) at high accuracy and precision, while being generalizable and adaptable to other HAR datasets.

Objectives: 1. Estimate the resting heart rate biomarker using: Personal data (i.e. different estimates for different individuals) 2. Accessible wearable device data, specifically Fitbit data 3. Transparent and meaningful model/logic based on background literature Organizations: This project is part of the Big Ideas Lab at Duke University. We collaborated with the DISCOVeR Lab at Stanford University, which conducted the Strong-D study. The Strong-D Fitbit data set was used to develop, evaluate and publish this resting heart rate estimation model.

wearablecompute is an open source Python package containing over 50 data and domain-driven features that can be computed from wearables and mHealth sensor data. wearablecompute has been integrated in the Digital Biomarker Discovery Pipeline (DBDP), currently in use in dozens of projects across several institutions. This resource will help make discovering digital biomarkers more accessible and standardize best practices across the field.

Continuous glucose monitoring (CGM) systems provide real-time, dynamic glucose information by tracking interstitial glucose values throughout the day. Glycemic variability, also known as glucose variability, is an established risk factor for hypoglycemia (Kovatchev) and has been shown to be a risk factor in diabetes complications. Over 20 metrics of glycemic variability have been identified. Here, we provide functions to calculate glucose summary metrics, glucose variability metrics (as defined in clinical publications), and visualizations to visualize trends in CGM data.

A Python package to interact with fasting logs from apps like Zero.

Devicely is a Python package for reading, de-identifying and writing data from various health monitoring sensors. With devicely, you can read sensor data and have it easily accessible in dataframes. You can also de-identify data and write them back using their original data format. This makes it convenient to share sensor data with other researchers while mantaining people's privacy.

This repository contains the code artifacts released as part of the following publications: /JMPB-2021 : Application of Convolutional Neural Network Algorithms for Advancing Sedentary and Activity Bout Classification, Journal for the Measurement of Physical Behaviour, DOI|Paper /MSSE-2021 : The CNN Hip Accelerometer Posture (CHAP) Method for Classifying Sitting Patterns from Hip Accelerometers: A Validation Study in Older Adults

Feature Engineering and Extraction Module Tutorial

Sleep & Circadian Rhythm

The DHDR (Digital Health Data Repository) is a repository with sample data for use with the DBDP. We welcome contributions to the DHDR! If your data is too large to upload here, please consider linking a repo here to another data repository (PhysioNet, UCI ML) while we are working on getting the full DHDR up and running!

Code for running the method and replicating results, figures, and tables from 'Adjacency Matrix Decomposition Clustering for Human Activity Data'.

Accompanies Comparing Step Counting Algorithms for High-Resolution Wrist Accelerometry Data in NHANES 2011-2014. Given minute-level step counts from NHANES, run mortality analysis.

A Python implementation of the Mobilise-D algorithm pipeline for gait analysis using IMU worn at the lower back

Functions for Optimizing Missing Observations (FOMO) is a Python module to help you better understand missingness in your data.

An improved Dynamic Time Warping algorithm for aligning biomedical signals of nonuniform sampling frequencies.

Calculates time and frequency domain heart rate variability metrics from RR interval (ECG) or IBI (PPG).

Functions to extract features from wearable data to track and predict infectious disease status.

Tools for exploratory data analysis of wearables and mHealth data.

Feature generation tooLkIt for weaRable daTa such as that from your smartwatch or smart ring.

Functions to calculate glucose summary metrics, glucose variability metrics.