Predicting Relapse Onset in Bipolar Disorder

Bipolar disorder (BD) is characterized by extreme fluctuations in mood, energy, and cognition resulting in manic and depressive episodes. There is no cure for BD; traditional treatment steps involve lifelong illness management and constant vigilance against relapse. Indeed, preventing relapse is a key clinical focus since it results in better patient outcomes, improves quality of life, and reduces the need for hospitalization1 — a major source of national health care cost.

Mania and depression episodes are often accompanied by behavioral anomalies and significant changes in social interactions. Hyperactivity, decreased need of sleep, being more talkative, and irritability often precedes mania; symptoms including low energy, hypersomnia, and lack of interest are common before a depressive episode2. Identifying and managing these early symptoms is crucial to prevent and minimize the impact of relapse in BD. However, current clinical methods for detecting early-warning signs are inadequate for granular and long-term monitoring.

Our social relationships and behavioral routines are now often embedded in online activities. We use mails and social media to communicate and collaborate, type into search engines to retrieve information, and utilize online shopping services to buy things. This is true for individuals with serious mental illnesses as well — approximately 81% of patients with BD use online technologies. Online activities provide unique insights into behavior and social interactions, and thus could be used to identify early- warning signs in BD.

In this project, we aim to address this gap. Our key goal here is to develop a prediction framework that will identify behavioral anomalies and early-warning signs in BD using continuous streams of online behavioral data. Towards this goal, we are collecting online activity data (e.g., email, social media, and search queries) as well as clinical history from patients with BD. This will enable us to associate personalized trends and changes in online activities with illness states — leading to a computational framework to identify early-warning signs in BD.

This project is a collaboration with Erika Saunders and Dahlia Mukherjee.