Toward personalizing treatment for depression: predicting diagnosis and severity

Area of Science:

• Computational psychiatry
• Clinical informatics
• Health data science

Background:

• Depression diagnosis and treatment response prediction remain challenging.
• Personalized treatment approaches are needed for depression.
• Electronic Health Records (EHR) offer rich data for predictive modeling.

Purpose of the Study:

• To develop and evaluate computational models for predicting depression diagnosis and severity using EHR data.
• To predict patient response to depression treatments.
• To enable personalized depression care through data-driven insights.

Main Methods:

• Regression-based models were developed using structured EHR data (diagnosis, medication codes) and unstructured clinical notes.
• Two large datasets were utilized: Palo Alto Medical Foundation (35,000 patients) and Group Health Research Institute (5,651 patients).
• Model performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Main Results:

• Models accurately predicted future depression diagnosis up to 12 months in advance (AUC 0.70-0.80).
• Severe baseline depression was differentiated from minimal/mild depression (AUC 0.72).
• Baseline depression severity emerged as the strongest predictor of treatment response.

Conclusions:

• EHR data can reliably predict depression diagnosis and severity.
• The developed models are portable due to their use of commonly available EHR data.
• Automated severity assessment can facilitate large-scale research into depression treatment moderators.