Finding treatment-resistant depression in real-world data: How a data-driven approach compares with expert-based...

Area of Science:

• Psychiatry
• Data Science
• Pharmacoeconomics

Background:

• Treatment-resistant depression (TRD) is diagnosed when depression does not respond to standard antidepressant therapies.
• Defining and identifying TRD in large healthcare databases presents significant challenges due to complexities in treatment response, dosage, and duration assessments.
• Existing TRD definitions are difficult to implement in claims databases, necessitating a more practical, data-driven approach.

Purpose of the Study:

• To develop and evaluate a data-driven definition for identifying treatment-resistant depression (TRD) within large claims databases.
• To assess the performance and transportability of the developed TRD definition.
• To provide a more feasible method for identifying TRD patients for research and clinical purposes.

Main Methods:

• Adult patients with depression were included, excluding those with mania, dementia, or psychosis.
• Decision tree predictive models were fitted using data from three databases, analyzing antidepressant and antipsychotic use, psychotherapy, and expert-based definitions within 3, 6, and 12 months prior to an index date.
• Performance was evaluated using the area under the curve (AUC) and transportability metrics, with subjects stratified by the presence or absence of TRD proxies like electroconvulsive therapy or neuromodulation.

Main Results:

• The analysis included 33,336 subjects without TRD proxies and 3,566 with proxies.
• A model using data from 12 months prior to the index date demonstrated the best performance with an AUC of 0.81.
• The derived rule, identifying TRD as the use of ≥1 antipsychotic or ≥3 antidepressants in the past year, was successfully applied, identifying TRD in 15.8% of depression patients.

Conclusions:

• The most effective data-driven definition for discriminating between subjects with and without TRD involves analyzing the number of distinct antidepressants (≥3) or antipsychotics (≥1) used in the preceding year.
• This validated definition offers a practical and reliable method for identifying TRD in large patient populations.
• The findings facilitate improved identification of patients with TRD, potentially leading to more targeted and effective treatment strategies.