Researchers Analyze 10,000 Depression Cases to Develop Personalized Treatment Tool

Depression arises from a complex mix of psychological patterns, biological vulnerabilities, and social stressors, so its symptoms and underlying causes can look very different from one person to the next. Treating it is equally complex and often requires an individualized combination of medication, psychotherapy, and lifestyle changes.

In a decade-long, multi-institutional project, psychologists from the University of Alberta collaborated with Radboud University in the Netherlands to develop a precision approach to depression treatment. The goal is to generate tailored recommendations for patients based on multiple personal characteristics, including factors such as age and gender. The team described its work in PLOS One.

Zachary Cohen, an assistant professor in the University of Alberta’s Department of Psychology and senior author of the paper, said first-line depression care should not follow a one-size-fits-all model. In many clinics, he noted, the current standard still resembles trial and error: clinicians try different medications or therapy approaches until something works.

According to Cohen, roughly half of patients do not respond to initial treatment. That uneven response highlights how much treatment outcomes can vary across individuals.

The project focused on adults with depression and pooled participant-level data from randomized clinical trials worldwide. The researchers included trials that evaluated five commonly used treatments: antidepressant medication, cognitive therapy, behavioral therapy, interpersonal therapy, and short-term psychodynamic therapy.

Before treatment, trial participants had been assessed across a range of dimensions, including the presence of additional mental health conditions such as anxiety disorders and personality disorders. Ellen Driessen, the study’s lead researcher and an assistant professor of clinical psychology at Radboud University, said the team examined whether certain features—such as comorbid conditions—were linked to better outcomes with one type of treatment compared with another.

The researchers ultimately aim to create a clinical decision support tool: an algorithm that can consider many variables at once and produce a single, personalized recommendation. Instead of relying on broad guidelines that offer general options, clinicians would enter an individual’s characteristics into the tool and receive a targeted suggestion based on patterns observed across the combined evidence.

Cohen said earlier attempts at treatment selection often relied on single clinical trials, where limited sample sizes make it difficult to build reliable prediction models. To overcome that limitation, the team spent about 10 years collecting and processing data from more than 60 trials involving nearly 10,000 patients. Researchers worldwide contributed by sharing trial data, and the project brought together an international, multidisciplinary group to design the analysis strategy.

Cleaning and merging the datasets took years, Cohen said, but the aim is to ensure the eventual model reflects the full weight of available evidence. Driessen added that the published paper sets out the protocol and analytical plan in detail, while the tool itself is expected to be developed over the next one to two years.

Looking ahead, the researchers plan to run a clinical trial to test whether using such a decision support tool improves outcomes by better matching patients to their optimal treatment. If the results are positive, the tool could be scaled for real-world use, potentially as a simple computer program or web application where clinicians input patient information.

The team hopes the approach will help clinicians and patients make more efficient use of existing treatment options and reduce the heavy personal and societal burden associated with depression. Cohen noted that the variables the model would use are generally easy to collect through standard questionnaires and basic demographic information, which could make implementation relatively low-cost and widely applicable.