Tackling mental health problems using a heavy dose of data
Figuring out the best ways to prevent and treat mental health disorders—including psychosis in adolescents and young adults, suicide risk, opioid use disorder, and major depressive disorder—is what drives the research of Alejandro Szmulewicz.
Szmulewicz, a newly appointed assistant professor of epidemiology, said he knew early on that he wanted to go into psychiatry. “I was always very puzzled by, and interested by, diseases of the mind,” he said.
In his research, he analyzes large data sources to shed light on the best way to treat mental health disorders, when to start or stop treatment, and how to choose the best treatment for patients in a context where different courses of action exist. He uses a methodology called causal inference, which makes use of data in a way that can emulate a randomized trial.
“When you have different potential courses of action to treat patients with mental health conditions, the best way to understand which course to take is to conduct a randomized experiment—one in which patients are randomized to different treatment strategies,” he explained. “But if you cannot do a randomized experiment—because it takes a lot of time or it isn’t ethical or feasible—the next best thing is to use available data on patients who are receiving different treatments and look at how they fare with those treatments in real life. We can leverage that data to provide timely answers.”
Although Szmulewicz began his new role on Jan. 6, he’s been at Harvard T.H. Chan School of Public Health since 2017, when he enrolled as a student in the MPH program. He first heard about the School during his psychiatry residency in his hometown of Buenos Aires, Argentina. He had begun work on a research project to investigate the cardiovascular risk of certain antipsychotic medications and had collected a large amount of data but was having trouble analyzing it. “I started asking around to find someone who could help me analyze the data to move the research forward,” he recalled. “A friend said they knew someone doing a master’s at a place called the Harvard School of Public Health. So I contacted this person and he was like, ‘Yeah, this is easy.’ And he was able to help me in like three days. I was blown away by that. I knew that this was the place I needed to go.”
Szmulewicz earned an MPH at the School in quantitative methods in 2018, and stayed on for both a master of science in biostatistics and a PhD in epidemiology, completing both degrees in 2022. He was a lecturer for two years before being named an assistant professor. “Seven years later, I am still here,” he said.
Understanding first-episode psychosis
Young patients experiencing their first episodes of psychosis are of particular interest to Szmulewicz. “They have just experienced their first-ever episode of psychosis and they don’t know what’s going on,” he said. “They don’t know if it’s just an episode or a lifelong condition. They want to know how long they have to take medications. And we don’t have good answers for them and their families.”
Since evidence has shown that early intervention is key to improving long-term outcomes for patients experiencing psychosis, Szmulewicz has focused his work on pinpointing what those interventions should be.
One recent study led by Szmulewicz found that two newer antipsychotics, aripiprazole and paliperidone, are better tolerated by young people than several other types. “That’s an important finding,” said Szmulewicz, “because a strong predictor of individuals relapsing after a first episode is discontinuing antipsychotic medications too early. People discontinue it because they don’t like it; they find it has a lot of side effects. And so the finding that these two newer medications are better tolerated indicates that if we chose those agents as first-line treatments, it would be less likely that patents would stop medication early and relapse.”
Another study that Szmulewicz co-authored, currently under review, looked at injectable antipsychotic agents, finding that because patients don’t have to take them every day like oral medications—the shots are given every one to three months—it helps them stay medicated and prevents them from relapsing.
Preventing suicides
For his research on suicide prevention, Szmulewicz conducted a study on whether lithium can help prevent suicide in patients with mood disorders. Lithium is a very safe and affordable medication, and many psychiatrists prescribe it for their patients with suicidal ideation. But studies have found conflicting evidence about whether it’s truly protective.
Szmulewicz and colleagues reanalyzed the data from one recent randomized trial that showed that lithium had no effect in preventing suicide. “We took into account the fact that many patients discontinue lithium because of the side effects,” he said. “The original analysis didn’t account for the fact that many patients took fewer lithium pills than instructed. So when we reanalyzed the data, we found an effect that was compatible with a protective effect from lithium. It was not enough to make a very firm conclusion, but it was a signal that lithium can prevent suicides. We’re continuing that work now with larger sample sizes to get to a firmer understanding about lithium’s benefits.”
Data difficulties
Pulling together high-quality data is the biggest challenge in Szmulewicz’s work. Patients might start with one treatment, then switch to another. They might stop or start taking medications. And patients themselves fluctuate greatly with respect to their symptoms over time. To do proper analysis, Szmulewicz needs to gather and make sense of large amounts of data that need to include these fluctuations in symptoms, what medications they’re receiving, how they’re faring on those medications, and whether they’re relapsing or hospitalized. And the data needs to be regularly updated.
“It takes a lot of work,” he said. “Unlike other clinical areas, there are few standardized measurements in routine clinical care for mental disorders.”
Szmulewicz said that standardized measurements do exist for disease severity, specific symptoms, suicidality, functional status, and social determinants of health, but they are used inconsistently and infrequently in clinical practice. “The information thus required for our type of analyses typically lies in the clinical notes, which are not available in the most frequently used databases,” he said
Potential avenues to improve current data sources include the use of large language models or expert-derived adjudication on variables typically contained in clinical notes to complement electronic health records data, Szmulewicz said.
But the extra effort is no deterrent. “The fact that it’s so challenging is what makes it so interesting,” he added.
