Multitask Learning Model Outperforms Traditional Machine Learning Models in RCTs, Developers Say

The proposed multitask learning (MTL) model was developed using data from wearable devices worn by individuals in a randomized controlled trial (RCT) to predict outcomes of a depression treatment.

Researchers developed a novel multitask learning (MTL) model designed specifically for randomized controlled trials (RCTs), which analyzes data from both the intervention and control patient groups in RCTs and eliminating the need to develop a separate model for each set of patients.

According to the researchers, this proposed MTL model outperforms the traditional and single-task machine learning models in predictive performance.

These findings were published in the Proceedings of the ACM on Interactive, Model, Wearable and Ubiquitous Technologies, and will be presented at the UbiComp 2022 conference in September.

RCTs traditionally rely on statistical analyses to evaluate differences between treatment and control groups in studies, but generally fail to measure the treatment’s impact at an individual level.

According to the authors, individualized predictions may facilitate more targeted and cost-effective therapy.

“Integrated behavioral therapy can be expensive and time consuming,” Chenyang Lu, PhD, professor at the McKelvey School of Engineering at Washington University in St. Louis and author on the study, said in a news release. “If we can make personalized predictions for individuals on whether it is likely a patient would be responsive to a particular treatment, then patients may continue with treatment only if the model predicts their conditions are likely to improve with treatment but less likely without treatment.”

To develop the proposed MTL model, the authors collected data from wearable devices worn by individuals from an RCT in order to predict outcomes of a depression treatment.

In the RCT, 106 participants with depression wore Fitbit devices and received psychological testing. About two-thirds also received behavioral therapy, while the other third did not. Patients were statistically similar at baseline so the authors could better assess whether behavioral therapy would lead to improved outcomes based on individual data.

The proposed model integrated clinical characteristics and data collected by the devices in a multilayer architecture, and enabled a knowledge transfer between the intervention and control groups to optimize prediction performance for both patient groups.

According to the authors, this proposed MTL model marks a promising development and important step in personalized medicine.

“The implications of this data-driven approach extend beyond randomized clinical trials to implementation in clinical care delivery, where the ability to make personalized prediction 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,” said Jun Ma, MD, PhD, professor of medicine at the University of Illinois Chicago and author on the study.

To further test this proposed MTL model, the authors plan to apply the model in a similar RCT on telehealth behavioral interventions using Fitbit wristbands and weight scales in patients in a weight loss intervention study.

Reference

Dai R, Kannampallil T, Zhang J, Lv N, Ma J, Lu C. Multi-task learning for randomized controlled trials: a case study on predicting depression with wearable data. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. Published online July 7, 2022. doi:10.1145/3534591