An important characteristic of major depressive disorder (MDD) is the variation in symptoms from patient to patient (Musliner et al. 2016). Additionally, treatment success is patient-specific, with 20-25% of MDD patients at risk of developing chronic depression (Penninx et al. 2011). Recent research aims to identify biomarkers that can guide treatment decisions and predict treatment outcomes (Gadad et al. 2018). Tailoring treatments to individual patients, particularly those at high risk, could improve the remission rate.
Chronic depression is associated with specific symptoms and characteristics, such as longer duration of symptoms, increased severity, earlier onset, higher levels of neuroticism, lower levels of extraversion and conscientiousness, and various inflammatory markers, low levels of vitamin D, metabolic syndrome, and lower cortisol awakening response (Habets et al. 2023). However, previous attempts to predict individual-level treatment response have been unsuccessful.
To address this, Habets et al. (2023) utilized multi-omics data, demographic, physiological, and clinical data with a non-linear prediction method to capture the complex pathophysiology of MDD. The study included 804 participants from the Netherlands Study of Depression and Anxiety (NESDA). The authors trained prediction models separately with a non-linear technique called XGBoost, using cross validation, and evaluated their performance using area under the receiver operating characteristic curve (AUROC). Additionally, they used SHAP analysis to determine variable importance.
The study found that combining clinical and proteomic data had the highest AUROC of 0.78. Fibrinogen showed the highest variable importance in the proteomics alone model, while symptom severity at baseline was most important in the model with clinical data alone. The study also compared the performance of prediction models with clinician assessments, finding that the models outperformed human raters.
The study concluded that a combined signature of symptom severity, personality traits, and immune and lipid metabolism related proteins at baseline was predictive of remission of MDD within 2 years. However, the study acknowledged that the overall accuracy of the final model is still too low for general clinical practice.
In conclusion, the study suggests that a combination of clinical and proteomic data can improve the accuracy of predicting depression remission after two years. This highlights the potential for machine learning models in clinical trials. However, further research is needed to address limitations and improve the generalizability and practicality of the models for clinical use.
