Various biomarkers have been identified as being associated with the pathophysiology of major depression, with the potential to be utilised within an objective laboratory test for the diagnosis of depression, based on machine learning techniques.

This study aims to build on previous results by modelling, in a larger and more heterogeneous cohort, the joint diagnostic accuracy of urine and serum-based biomarkers that showed predictive value for depression in our previous work.

A novel, multivariable, machine learning-based diagnostic tool for depression was tested on a combination of 34 urine- and serum-based biomarkers among 160 people diagnosed with major depressive disorder (MDD) and 120 controls, split into 3 different cohorts. Quantile-based prediction was applied to construct a biomarker-based diagnostic model (BDM) yielding a score for each biomarker. The sum score for each participant was used to calculate an area under the receiver operating characteristic curve (AUC) as a measure of discriminatory power.

We demonstrated that the BDM after internal validation had good discriminatory power, with an AUC of 0.81. Further internal–external validation by calculating individual depression probability scores for each separate cohort resulted in an AUC of between 0.62 and 0.72.

In terms of clinical applicability, the present study shows that the combination of biomarkers and a machine learning model can discriminate between MDD and healthy controls with a modest level of diagnostic accuracy. A biomarker test could have potential added value for the future diagnostic toolkit, but this does require further research.