Skip to navigation Skip to main content Skip to footer

Approved Research

Machine Learning for Creating Clinical Biomarkers Base on Wearable Devices

Principal Investigator: Dr Michael Kane
Approved Research ID: 74526
Approval date: May 19th 2022

Lay summary

Fitness devices, like FitBits, Apple Watches, and Garmin Watches are often used to help people track their fitness. The devices capture exercise by recording the gestures and movements people make and then use machine learning to figure out the type and intensity of the exercise. However, there's nothing about the devices that is specific to exercise and they can be used to track other activities like being asleep vs. awake for example.

This project is going to use the low-level recordings captured by these devices to try to understand sleep/awake patterns for healthy people compared to people with narcolepsy. We are focusing on type 1 narcolepsy which is disease that causes fall asleep or pass out and is usually triggered by strong emotions. Currently, this disease is usually treated with stimulants and patients report how well their treatment works. Measuring the efficacy this way is subjective in that we are determining how well a subject is doing by how well they say they are doing. This is a problem because a subject may think they are doing better or worse than they actually are, and it is difficult to compare subjective responses. We would like an objective, quantitative way to determine if a subject is doing well and fitness devices provide a way to do this.

This project is going to take data from fitness devices to quantify how well a narcolepsy drug works through their movement. We will accomplish this in three steps. In the first step We'll do analyze data for a variety of individuals without the disease. This will include individuals with different ages, genders, and possibly other disease conditions. This will essentially let us see what how define "normal," non-narcoleptic activity. In the second part we'll find the differences between individuals with narcolepsy and those without. This lets us understand how narcoleptic individuals are different from non-narcoleptic individuals. In the third part, we will examine narcoleptic patients. At this point we can make two distinct comparisons, first, we can see how therapies change actigraphy for narcoleptics. This tells us the effect of therapy. Second, we can see how narcoleptics receiving therapy compare to non-narcoleptics. This tells us if we have actually made them similar to non-narcoleptics or if they have their own activity profile.