Fitbit recognizes bipolar disorder

Fitbit smart bands and smartwatches have been used by scientists to diagnose bipolar disorder, formerly called manic-depressive disorder. The algorithm used allowed for the prediction of upcoming episodes of disease exacerbation.

Research will make life easier for thousands of patients and their loved ones

The researchers used data from popular Fitbit devices, a brand of devices owned by Google. We managed to train on their basis a machine learning algorithm capable of accurately predicting mood episodes associated with bipolar disorder. This disease is characterized by extreme mood swings, between depression and mania, followed by a period of remission.

These episodes have a significant impact on patients’ lives, work, relationships and overall health. Current treatment for ChaAS focuses on limiting the effects of episodes, but this requires rapid recognition and therapy. A breakthrough may come from the research of scientists from Brigham and Women’s Hospital (BWH) in Boston, who have found an effective way to detect upcoming episodes of the disease based on data from Fitbit sports bands.

Most people now use digital devices such as smartphones and smartwatches that collect data every day that can support psychiatric treatment. Our goal was to use these data to identify mood episodes in study participants diagnosed with BD

– says Dr. Jessica Lipschitz from the Department of Psychiatry at BWH and the main author of the study

According to research, most people with bipolar disorder experience characteristic mood polarization at least three times a year. There are two main types of BD: type I (BP-I) i type II (BP-II). BP-I is characterized by episodes of mania lasting at least seven days or requiring hospitalization, as well as separate episodes of depression usually lasting at least two weeks. BP-II is a model of depression and hypomania, a milder form of mania that does not cause disturbances in functioning in society.

The study involved 54 adults with both variants of previously diagnosed bipolar disorder. 17 variables were analyzed, from the number of steps, through the duration of intense activity, sitting time, heart rate, sleep time and efficiency, individual sleep stages to the hour of falling asleep. This data was then used to train an AI model.

The effectiveness turned out to be stunning

The algorithm predicted 89.1% symptoms of hypomania or mania (sensitivity: 80.0%, specificity: 90.1%) and 80.1% of depression symptoms (sensitivity: 71.2%, specificity: 85.6%). And this is a real breakthrough, because previous forms of research used invasive methods, such as geolocation or voice analysis, which raised concerns about privacy. Now all you need is a simple sports band, charged every 5-7 days, and sending the analytical data to a specialist.