Most patients are offered antidepressant medication (ADM) by default. That’s largely because ADM is more accessible and less costly than therapy. However, research suggests that psychotherapy may actually work better for many people.

Unfortunately, primary care clinics lack tools to determine who will benefit most from which treatment. This is where technology, particularly electronic health records (EHR), can help. Zoobook Systems looked at recent research from the Department of Veteran Affairs to find the latest tools to support the treatment of depression. This study aimed to build a data-based tool to match Veterans with the treatment likely to be most effective for them.

What Problem Does This Study Address?

The central challenge this study tackles is the persistent mismatch between patients with major depressive disorder (MDD) and the treatments they receive. Although psychotherapy—alone or in combination with antidepressant medication (ADM)—has been shown to produce better symptom remission and lower risk of serious negative outcomes (such as suicide attempts, psychiatric hospitalizations, or suicide deaths) compared to ADM alone, the vast majority of MDD patients are still treated with medication only.

Several factors drive this mismatch:

• Greater availability and lower cost of ADM compared to psychotherapy.
• Shortages of trained psychotherapists, making it unrealistic to offer therapy to everyone who might benefit.
• Systemic barriers in healthcare delivery and resource allocation.

Previous attempts to personalize treatment selection for MDD have been limited by:

• Small sample sizes—making it hard to detect meaningful patterns or subgroups
• Limited predictor data—focusing on a narrow range of patient characteristics rather than a comprehensive view
• Suboptimal analytical tools—using traditional statistical approaches that may not capture complex relationships or heterogeneity in treatment response

This study advances the field by addressing all of these limitations:

• Leveraging a large sample of over 43,000 veterans receiving care in the Veterans Health Administration (VHA), which provides access to both ADM and psychotherapy.
• Incorporating 5,865 baseline predictors from electronic health records (EHR) and geospatial social determinants of health databases—covering mental and physical health, medications, socioeconomic status, neighborhood factors, and facility-level indicators.
• Applying advanced machine learning methods (such as Super Learner and Generalized Random Forests) to model risk and optimize individualized treatment assignment.

The study’s primary focus is on reducing the risk of serious and costly mental health events within 12 months of starting treatment, specifically:

• Suicide attempts
• Psychiatric emergency department or urgent care visits
• Psychiatric hospitalizations
• Suicide-related deaths

By developing and testing an individualized treatment rule (ITR), the study aims to better match each patient to the treatment most likely to minimize their risk of these serious outcomes. Improving this match could help reduce both individual suffering and the broader burden on healthcare systems.

How Was the Study Conducted?

The study used data from 43,470 Veterans treated in VA mental health clinics. These patients were split into three treatment groups:

• ADM-only: Medication without therapy
• Psychotherapy-only: Therapy without medication
• Combined treatment: Both medication and therapy

Study Timeline and Setting:

• Timeframe: October 2015 – December 2016
• Setting: VA Primary Care Mental Health Integration (PC-MHI) clinics

Data Sources:

• Electronic Health Records (EHR)
• VA Suicide Prevention Applications Network (SPAN)
• National Death Index (NDI)
• Geospatial databases for neighborhood-level data

Data Included:

• Over 5,800 baseline predictors
• Mental and physical health conditions
• Previous medication and treatments
• Social determinants like housing, economic status, and healthcare access

Analysis Techniques:

• Super Learner model to predict risk
• Generalized Random Forests to capture individual treatment differences
• Targeted Minimum Loss Estimation (TMLE) to estimate treatment effects

EHR-Driven Innovations:

• Risk score prediction models
• Individualized Treatment Rule (ITR) creation
• Subgroup analysis to evaluate treatment impact

What Did the Study Find?

1. Prediction of Serious Negative Outcomes

The study’s aggregate risk model, which used a comprehensive set of 5,865 baseline variables from electronic health records (EHR) and geospatial data, demonstrated moderate predictive accuracy for identifying patients at risk for serious negative outcomes (defined as suicide attempt, psychiatric emergency/urgent care visit, psychiatric hospitalization, or suicide-related death within 12 months of treatment initiation). The model achieved an area under the receiver operating characteristic curve (AU-ROC) of 0.68 (standard error = 0.01).

• Risk Stratification:
  In the test sample, the 5% of patients with the highest predicted risk had a 32.6% prevalence of serious negative outcomes, compared to just 7.1% in the remainder of the sample. This indicates that the model could meaningfully differentiate between higher- and lower-risk patients.

2. Average Treatment Effects

After adjusting for differences in baseline characteristics across treatment groups using advanced statistical methods (propensity score weighting and outcome modeling, combined via targeted minimum loss-based estimation), the study found:

• Psychotherapy-Only Superiority:
  On average, patients who received psychotherapy-only had a significantly lower risk of experiencing a serious negative outcome compared to those who received antidepressant medication (ADM)-only or combined ADM-psychotherapy treatment.
• Magnitude of Effect:
  For the majority of patients, psychotherapy-only was associated with roughly a 20% lower risk of adverse outcomes compared to the other treatment options.

3. Individualized Treatment Rule (ITR) Performance

The individualized treatment rule (ITR) developed from the data identified which treatment (ADM-only, psychotherapy-only, or combined) would be optimal for each patient in terms of minimizing risk for serious negative events.

• Distribution of Optimal Treatment:
  The ITR indicated that for 56% of patients, psychotherapy-only was the optimal treatment (i.e., it would provide the lowest risk of serious negative outcomes).
• Other Patients:
  For the remaining 44% of patients, the type of treatment (ADM-only, psychotherapy-only, or combined) was not significantly related to outcome risk—suggesting these patients could reasonably receive any of the three treatments without a notable difference in risk.

4. Cost Implications

Implementing the ITR in clinical practice would have minimal impact on overall treatment costs:

• Resource Allocation:
  The ITR would result in 16.1% fewer patients being prescribed antidepressant medications and 2.9% more patients receiving psychotherapy.
• Cost Neutrality:
  Despite these shifts, the aggregate treatment costs would remain essentially unchanged, suggesting that targeting psychotherapy to those most likely to benefit is feasible from a budgetary perspective.

5. Clinical and Policy Implications

• Targeted Psychotherapy:
  The findings support the idea that, given current shortages of trained psychotherapists, psychotherapy resources should be prioritized for those patients most likely to benefit—those identified by the ITR.
• Proof of Concept:
  The study demonstrates that it is possible to use large-scale EHR data and advanced machine learning to develop practical, data-driven rules for individualized treatment assignment in mental health care.

6. Need for Further Validation

• Pragmatic Trials Required:
  Although the ITR performed well in this retrospective, observational analysis, the authors emphasize that a prospective, pragmatic trial is necessary to confirm the accuracy and clinical utility of the ITR before it can be widely implemented

The study found that a data-driven, individualized treatment rule can identify a substantial subset of patients with major depressive disorder who would benefit most from psychotherapy-only, leading to a significant reduction in serious negative outcomes without increasing overall costs. For the rest, treatment type made little difference in risk, suggesting flexibility in treatment assignment for these individuals. This approach offers a promising path toward more personalized, efficient, and effective mental health care, pending further validation in clinical trials.

Limitations of the Study

While the study shows promise, several limitations exist:

Observational Design:

• The study wasn’t a randomized trial
• Results could be biased by unknown variables

Missing Data:

• Lacked biomarker or genetic information
• No data on therapy type (e.g., CBT)
• Didn’t account for patient treatment preferences

Excluded Populations:

• Patients using antipsychotic medications were left out
• These may include treatment-resistant or severe MDD cases

Only Focused on Serious Events:

• Did not measure symptom relief or life quality improvements

Generalizability:

• Results may not apply outside of the VA
• Civilian clinics might have less access to psychotherapy

Bottom Line

This study shows how EHR data can be used to create smarter, more personalized depression treatments. For more than half of Veterans, psychotherapy alone could reduce serious mental health risks better than medication.

By adjusting treatment plans based on real-world data, clinics can:

• Improve patient outcomes
• Use fewer resources
• Prevent serious health crises

Key Takeaways:

• Psychotherapy is underused but highly effective for specific groups
• EHR-based tools can guide personalized care
• Results support the need for larger, randomized trials

What’s Next?

• Run a real-world clinical trial to validate findings
• Include more data types like therapy adherence, dosage, and patient feedback
• Expand model use beyond the VA to civilian settings

This research gives us a glimpse into the future of mental health care—personalized, data-driven, and more effective for the people who need it most.

The study highlights the power of electronic health records in shaping individualized treatment strategies for Veterans with depression. By analyzing a vast dataset and applying machine learning techniques, researchers identified which Veterans are most likely to benefit from psychotherapy alone. This finding is crucial, especially considering the overreliance on medication in many clinical settings.

Not only does the ITR reduce the risk of severe outcomes like hospitalization and suicide, but it also helps allocate healthcare resources more efficiently. It offers a promising step forward in transforming mental health care from a one-size-fits-all approach to a more personalized and effective system.

Zoobook Systems seeks to be on the cutting edge of new technologies, we are working to develop partnerships and new features that can support medication management for MDD patients. While more research is needed to validate these findings in diverse populations and real-world settings, this study lays a strong foundation. The integration of data-driven tools in mental health treatment could improve not only outcomes but also the overall quality of care for Veterans struggling with major depressive disorder. Zoobook Systems’ EHR solutions can assist clinics in adopting such procedures.