Technical University of Denmark: Digital biomarkers identify bipolar disorder
Digitalisation plays a part in everything we do in our daily lives. Mountains of data collected from all our devices are a key ingredient in the digitalisation process. Health apps on our mobile phones, for example, record our unique behavioural and physiological data, and use that data to help us monitor and understand our own health, and can ultimately lead to improved diagnostics through a much better tracking and understanding of health and disease than ever before. As shown below, psychiatry is an area where digitally collected data has been used to evaluate a patient’s condition.
The voice as a digital biomarker
Psychiatrists, who treat patients with bipolar disorder (also known as manic-depressive illness), can with reasonable accuracy assess a patient’s mood by talking to them and listening to their voice. I.e. they can discriminate between depression, mania and euthymia, with euthymia being a normal stable mental state.
Based on this knowledge, researchers from DTU have together with hospital colleagues investigated if a person’s mood can be assessed using their voice as a digital biomarker. I.e., if they can assess a person’s health condition through recordings of a person’s voice that are transformed to data.
For an extended period of time, the researchers analysed phone conversations with a mobile phone app. They converted each conversation to an average of notes, i.e., depth, volume, speed, and pitch. This transformation of data was done to protect the patients’ privacy by preventing subsequent reconstruction of the conversations.
Bipolar disorder is evident in your voice
Darius Rohani, who was previously employed as a Postdoc at DTU Health Tech and participated in the project, says, ”Our study shows that it is indeed possible to discriminate between patients with bipolar disorder and healthy controls based on an analysis of the voice. And we can also discriminate between patients with bipolar disease and healthy close relatives.”
The second part of the study investigated if it was possible to discriminate between the different moods that a patient with bipolar disorder experiences based on voice data. I.e., is it possible to discriminate between depression and normal mental state, as well as between manic and normal mental state? The results of this part of the study were unfortunately less convincing.
”When we looked at unknown persons, we were unable to discriminate between the different moods”, Darius Rohani explains.
“Our study shows that it is indeed possible to discriminate between patients with bipolar disorder and healthy controls based on an analysis of the voice.”
Postdoc Darius Rohani
“But when we train the model on a known person over a period of two weeks, we can predict if he/she is moving from normal to either a depressive state or a manic state. This means that we cannot give a mobile phone with the app to a newly diagnosed patient and expect to be able to detect when he/she is moving towards a depressive or manic state. But it can be used by a patient, who has had the app before, and where the model has been trained to recognize that person’s voice. In other words, a personalized model, where the app can be used to detect relapses.”
Digital biomarkers and Digital Phenotyping
The scientific background for developing personalised digital health technologies is the research field ‘Digital Phenotyping’, where digital phenotypes are identified for each patient.
”Digital Phenotyping is a new concept within the area of health technology that exploits large amounts of data to identify digital biomarkers”, Professor at DTU Health Tech, Jakob E. Bardram explains.
“A biological marker (biomarker) is a measurable marker that can provide information about a person’s health condition, for example the existence of a disease, physiological change, response to a treatment or a psychological state. A digital biomarker is a marker, which is measured with digital technologies and sensors, and can tell us something about a specific person’s physiological condition and behavioral patterns. They provide us with insights into for example sleep, social interactions, physical mobility, gross motor activity, cognitive function, and speech and language production – factors that can be linked to a person’s present health condition and development in the same.”
Digital biomarkers including our voice has shown great potential within the psychiatric area as shown in the current study. Furthermore, digital biomarkers are applied within e.g. cardio vascular disorders, metabolic diseases and diabetes.