Imagine if doctors could predict which type of therapy would work best for you before you even start treatment. A comprehensive study from Finland suggests this might soon be possible, using artificial intelligence to match patients with their optimal therapeutic approach.
Researchers analyzed data from over 2,000 adults who received short term psychotherapy through the Finnish healthcare system. They used advanced machine learning techniques to determine whether matching people to specific therapy types based on their characteristics could improve treatment outcomes compared to the current system where therapy assignment is often based on availability or general preferences.
Four Different Paths to Healing
The study examined four main types of therapy commonly used in Finland. Solution-focused therapy emphasizes building on existing strengths and working toward future goals. Cognitive-behavioural therapy focuses on changing unhelpful thought patterns and behaviours. Psychodynamic therapy examines how past experiences and unconscious processes influence current issues. Integrative and cognitive analytic therapies combine elements from various approaches, adapting flexibly to meet the individual needs of each patient.
Traditionally, research has suggested that these different therapy approaches produce roughly equal results on average, a finding known as the "dodo bird verdict" after the character in Alice in Wonderland who declared that "everyone has won and all must have prizes." However, this new research suggests that while therapies may be equally effective on average, specific individuals might benefit much more from one approach than another.
The Power of Personalized Matching
The researchers used a sophisticated statistical approach called targeted learning, which combines multiple machine learning algorithms to make predictions while accounting for the complex factors that influence both treatment assignment and outcomes. They fed the system information about patients' demographics, symptoms, medical history, social support, and other characteristics to predict which therapy would work best for each person.
When the algorithm assigned each patient to their predicted optimal therapy, treatment outcomes improved by about 0.3 standard deviations across multiple measures of mental health and functioning. To put this in perspective, this improvement would take someone from experiencing moderate benefits to experiencing good benefits from therapy.
This improvement is particularly meaningful because it's comparable to the difference between receiving therapy and receiving no treatment at all in many studies. In other words, getting the right type of therapy might be almost as important as getting therapy in the first place.
Different Goals, Different Recommendations
Interestingly, the optimal therapy recommendation for each person varied depending on what outcome the researchers were trying to maximize. Some patients might be predicted to respond best to one therapy for reducing depression symptoms, but a different treatment for improving their ability to function at work or in relationships.
Only about one-third of patients received the same therapy recommendation across all outcome measures, highlighting that the "best" therapy depends partly on what you're hoping to achieve. This suggests that clear goal setting at the beginning of treatment could be crucial for optimal therapy matching.
Who Benefits from What
The analysis revealed patterns indicating which types of patients were predicted to benefit from specific therapies. Patients recommended for solution focused therapy tended to have fewer social connections and support systems. Those assigned to psychodynamic therapy were typically younger, reported higher alcohol consumption, more loneliness, but fewer anxiety symptoms.
Cognitive behavioural therapy was more often recommended for older patients with more medical and psychiatric conditions, but lower alcohol use. However, the researchers emphasized that the optimal treatment rules were complex and involved intricate interactions between multiple factors that would be difficult for humans to detect or remember.
If implemented in clinical practice, this type of system could potentially help therapists and patients make more informed decisions about treatment approaches. Rather than relying solely on therapist preferences, patient requests, or simple availability, treatment decisions could be informed by data driven predictions about what's most likely to work for each individual.
The researchers found that their algorithm would have assigned more patients to solution focused therapy and fewer to psychodynamic therapy compared to what actually happened in practice. However, every therapy type was still recommended as optimal for some patients, supporting the value of having multiple treatment options available.
Despite these limitations, the study provides compelling evidence that there's room for improvement in how we match patients to therapies. The researchers suggest that even if the real world benefits are smaller than their models predict, population level improvements in mental health treatment could still be substantial.
The most practical application might focus on patients for whom the algorithm makes strong, consistent recommendations across different outcome measures. Rather than trying to optimize therapy assignment for everyone, clinicians could use algorithmic assistance primarily for cases where the predicted benefits are largest and most reliable.
The Human Element Remains
This research doesn't suggest that artificial intelligence should replace clinical judgment or that therapy assignment should be entirely automated. Instead, it points toward a future where data driven insights could complement human expertise to help patients get the most effective treatment for their specific circumstances.
The study also reinforces that having multiple therapy options available is valuable, since different approaches were optimal for different people. This argues against oversimplifying mental health treatment or assuming that one size fits all approaches will work for everyone.
As mental health care continues to evolve, tools like these could help ensure that the right person gets the right treatment at the right time, potentially improving outcomes for the millions of people seeking help for mental health challenges. While we're not there yet, this research provides a compelling glimpse of how technology might enhance the very human work of psychological healing.
Malkki, V. K., Saarni, S. E., Lutz, W., & Rosenstöm, T. H. (2025). Targeted learning for optimal patient assignment to psychotherapy. Psychotherapy Research, 1-15.
