EDIT-B consortium: an external pan-European validation of an innovative machine learning-based test for bipolar...

Differentiating bipolar disorder (BD) from major depressive disorder (MDD) during major depressive episodes remains a significant challenge. EDIT-B is an in vitro diagnostic test based on machine learning (ML) method integrating clinical metadata with adenosine-to-inosine (A-to-I) RNA editing signatures in eight genes (GAB2, IFNAR1, IL17RA, LYN, MDM2, PRKCB, PTPRC, ZNF267) to differentiate BD from MDD.

The objective of this study is to assess and confirm the diagnostic performance of EDIT-B in a new, independent, double-blind, external, multicentre European cohort.

We evaluated EDIT-B results across four European centres (Spain, France, Denmark) compared with physician diagnoses. Explainability and sensitivity analyses were performed to identify the primary drivers of the model.

In 393 patients with current major depressive episode (238 MDD; 155 BD), EDIT-B demonstrated robust performance: area under the curve-receiver operating characteristic of 0.873 (95% CI 0.837 to 0.909), sensitivity of 82.6% (95% CI 75.7% to 88.2%) and specificity of 80.3% (95% CI 74.6% to 85.1%). Performances were consistent across sites and all stratified patient subgroups, corroborating previous results. Explainability analysis pointed to RNA editing biomarkers as the most important variables for EDIT-B diagnosis.

This study confirms the diagnostic performances of EDIT-B showing that A-to-I RNA editing coupled with ML methods is a reliable approach to distinguish BD from MDD.

EDIT-B represents an innovative, reliable and complementary diagnostic tool supporting psychiatric practice and enhancing clinical outcomes with potential impact in reducing diagnostic delays, tailoring the therapeutic strategy and improving therapeutic alliances. These findings may represent a significant advance towards precision psychiatry, with potential to reduce misdiagnosis and inappropriate treatment.

NCT05603819.