In the United States, depression affects 16 million people annually, and according to the World Health Organization it’s the leading cause of disability across the globe. Clinicians who treat patients with the disorder understand that different approaches work for different people, but the ability to discern and provide the most effective treatment for a particular individual remains challenging.
This customized-care concept, what’s known as precision medicine, is employed in a number of fields across the health-care spectrum. University of Pennsylvania researchers Robert DeRubeis and Zachary Cohen study it in the context of depression, and in a new paper for the Annual Review of Clinical Psychology, they describe what research has so far revealed about treatment selection for this mood disorder and which areas can still improve.
“There is a tidal wave of interest in and hope for using what we can find out about individuals to direct them to the treatment that’s most helpful,” says DeRubeis, a psychologist whose work focuses on the causes of and treatment for mood disorders.
Part of the problem lies in the number of options available for making treatment decisions, says Cohen, who recently completed his doctoral work at Penn.
“It’s important,” he says, “that as a field we take a step back and say, ‘We’re doing it 100 different ways. We’re all asking a similar question but to get a consistent answer, we should take stock.’”
In the realm of mental health, DeRubeis and Cohen argue that means making better educated guesses upfront about how a person will react to a particular medication or therapy. By their nature, current options can take extended time to prove effective and don’t always lead to patient progress, so “better predicting in advance the most fruitful way to proceed can have three big effects,” DeRubeis says.
One, it increases the chances someone will benefit from the selected treatment. That then leads to a greater likelihood of the person continuing to receive care rather than becoming frustrated or jaded and opting out altogether. Third, it makes the overall health-care system more efficient.
“Efficiency can sound like a bad word,” DeRubeis says, “but the more efficient a system is, getting the right treatment to the right people, the more good it can do. You really can’t not like that promise in the end.”
Since 2011, DeRubeis and colleagues have supported machine learning as one solution to this puzzle, building a tool called the Personalized Advantage Index, which calculates, based on variables specific to an individual, how that person will respond to two distinct care choices. In 2017, the team published promising results showing that the PAI, when indexing a person’s risk for persistent depression, can effectively guide whether that person should start with higher-intensity or lower-intensity treatment.
“‘Depression’ is a term the field has come up with,” Cohen says, “to describe a set of symptoms, but it might not have the same core for every person. Some people might experience hopelessness and feelings of suicidal thoughts; other people’s depression might involve restlessness and anger and an inability to concentrate. The disorder is far more varied in its presentation. And because that’s true, it might be that different treatments can address the underlying problems.”
The time is ripe to hone this process, according to the researchers, partially because of how the field has moved forward. Early in his career, DeRubeis explains, research focused on running clinical trials to determine whether Treatment X helped patients more than Treatment Y. If it did, the idea was that every patient would receive Treatment X.
Today, researchers like DeRubeis are instead analyzing the plethora of information contained in datasets from already-treated patients, from clinical trials and elsewhere. “Now,” DeRubeis says, “we think less about whether Treatment X is better than Treatment Y and more about who should get X as a first option and who should be given Y.”
It’s been an incremental but steady progression, one that’s moving closer to broad applications of treatment selection. The work is especially important in settings like Veterans Affairs facililties, where clinicians have already begun to use data to make treatment decisions regarding post-traumatic stress disorder for patients exhibiting certain characteristics. Though it’s unclear whether this currently improves outcomes, it’s likely that providing information about potential results of different treatments could enrich decision-making.
“When you’re implementing a treatment-selection model in large systems, you can get really meaningful and measurable benefits, where in the past those benefits would have been difficult to see, even if they were there,” Cohen says. “If you treat 500,000 people a year, even if you ‘only’ get a 5 percent increase in the likelihood of having a positive response, that’s 25,000 people. Yes, it’s good to improve efficiency of systems and help clinicians, but at the end of the day this is really about the patients.”
Robert DeRubeis is the Samuel H. Preston Term Professor in the Social Sciences in the School of Arts and Sciences. Zachary Cohen recently completed his doctoral work in the department of Psychology in the School of Arts and Sciences.
