Wearable technology can forecast treatment outcomes for depression, according to research


In recent years, managing one’s mental health has taken on greater importance, with a focus on self-care. More than 300 million people experience depression globally each year. Given this, there is a lot of interest in using common wearable technology to track indicators like activity level, sleep, and heart rate in order to monitor a person’s mental health.

Data from wearable technology was utilised by a group of researchers from Washington University in St. Louis and the University of Illinois at Chicago to forecast the effects of depression treatment on participants in a randomised clinical trial. Instead of creating a distinct model for each group, they created a novel machine learning model that examines data from two sets of patients: those who were randomly chosen to receive therapy and those who were not. A step toward customised medicine, in which doctors create a treatment plan unique to each patient’s needs and forecast outcomes based on an individual’s data, is this integrated multitask model.

The research’s findings were presented at the UbiComp 2022 conference in September and were published in the Proceedings of the ACM on Interactive, Model, Wearable, and Ubiquitous Technologies. The team was led by Chenyang Lu, the Fullgraf Professor at the McKelvey School of Engineering, and included Thomas Kannampallil, an associate professor of anesthesiology and associate chief research information officer at the School of Medicine and associate professor of computer science and engineering at McKelvey Engineering; and Jun Ma, MD, PhD, professor of medicine at the University. Dai previously worked in Lu’s lab as a doctoral student and is currently a software engineer at Google.

Lu remarked that integrated behavioural therapy can be costly and time-consuming. “Patients may continue with treatment only if the model indicates their problems are likely to improve with treatment but less likely without treatment. If we are able to create individualised predictions for individuals on whether a patient will be responsive to a specific treatment. Such individualised estimates of treatment outcome will enable more focused and economical therapy.”

Patients in the trial received Fitbit wristbands and underwent psychological assessment. The remainder patients did not undergo behavioural therapy, although about two thirds of the patients did. In order to determine whether treatment would result in better outcomes based on individual data, the researchers used baseline data from patients in both groups that were statistically similar.

Due to the expense and length of such interventions, clinical studies of behavioural therapies frequently used very small cohorts of participants. A machine learning model had difficulty because there were few patients, despite the fact that more data is usually better for performance. The model could, however, learn from a larger dataset that caught the differences between individuals who had received therapy and those who had not by integrating the data of the two groups. They discovered that their multitask model accurately predicted the development of depression more accurately than a model that considered each group separately.

In order to simultaneously train a unified model to predict the individualised results of an individual with and without therapy, Dai, who received a doctorate in computer science in 2022, combined the intervention group and the control group in a randomised control experiment. “The model included wearable data and clinical features in a layered design. With this method, the study cohorts are not divided into smaller groups for the machine learning models, and a dynamic knowledge transfer can take place across the groups to improve prediction accuracy both with and without intervention.”

According to Ma, “the implications of this data-driven approach extend beyond the implementation in randomised clinical trials to clinical care delivery, where the ability to make personalised predictions of patient outcomes depending on the treatment received, and to do so early and along the treatment course, could meaningfully inform shared-decision making by the patient and the treating physician in order to tailor the treatment plan for that patient.”

Building tailored predictive models using data from randomised controlled trials is made possible by the machine learning technique. The researchers will use machine learning in a new randomised controlled trial of telehealth behavioural therapies employing Fitbit wristbands and weight scales among patients in a weight reduction intervention study in the future.