2026 ECNP Congress

PS01-1109

Magnitude and trial methodology moderators of treatment response variance in major depressive disorder: a pooled analysis across 16 clinical studies

Marius Gruber — N.R. Winter, M. Elsaeßer, E. Schramm, G. Burkhardt, L. Bulubas, F. Padberg, J. Engelmann, K. Lieb, K. Schöninger, R. Redlich, M. Aichholzer, S. Edwin Thanarajah, V.B. Nöhles, C.U. Correll, C. Reif-Leonhard, T. Borgers, M. Klug, V. Enneking, G.R. Dawson, M. Richter, N. Opel, M.P. van den Heuvel, A. Reif, T. Hahn, J. Repple · Goethe University Frankfurt, Department of Psychiatry- Psychosomatic Medicine and Psychotherapy- University Hospital Frankfurt, Frankfurt am Main, Germany

2026 ECNP Congress

Introduction

Major depressive disorder (MDD) affects over 300 million individuals worldwide and remains a leading cause of disability. Despite effective treatments, treatment outcomes remain unsatisfactory: Up to 70% of patients fail to achieve remission with initial treatments, and nearly one-third develop treatment-resistant depression. In response, the ECNP Precision Psychiatry Roadmap calls for individualized treatment matching based on biology-informed predictors rather than symptom-based classification alone. Realizing this vision requires a clear understanding of what symptom-based models can and cannot predict, and how methodological factors such as assessment modality (clinician-rated vs. self-report), instrument choice (HDRS-17, MADRS, BDI-II, QIDS), and study design (RCT vs. observational) shape those conclusions. Without such benchmarks, claims of incremental predictive value from novel biology-informed models cannot be meaningfully evaluated.

Aims

Using harmonized data from clinical studies with available individual patient data spanning diverse study designs, assessment modalities, instrument choices, and treatment modalities, this study examines 1) treatment response patterns across methodological contexts, 2) the predictive value of baseline symptom severity, and 3) the added predictive value of demographic variables.

Methods

Included studies covered diverse treatment modalities and study methodologies, with most patients assessed via both clinician-rated (HDRS-17, MADRS) and self-rated (BDI-II, QIDS) scales. Predictive utility of baseline severity was assessed via reliability-corrected Pearson correlations and 5-fold cross-validated linear regression models. Extended models incorporating age, sex, and treatment duration were compared against baseline-only models. An exploratory logistic regression examined predictors of treatment response, defined as ≥50% symptom reduction from baseline to follow-up.

Results

Across 16 MDD studies (n=2,384, age=42.59±15.87 years, females=60.2%), treatment response varied substantially with assessment modality, study design, and rating scale choice. Pooled response rates were 44.8%, being relatively similar for clinician-rated and self-report measures (44.8% vs 46.3%), but consistently higher in RCTs than in observational studies across both perspectives (clinician-rated: 50.1% vs 35.6%; self-report: 54.1% vs 32.7%). Rating scale-specific differences were also observed (HDRS-17 54.0% vs MADRS 48.8%; QIDS 55.1% vs BDI-II 34.8%). This methodology-dependent heterogeneity was further reflected in predictive analyses: baseline severity explained a meaningful proportion of post-treatment outcome variance both overall (R²=9.90%) and across instruments (HDRS-17: R²=6.10%; MADRS: R²=15.7%), with substantially stronger predictive performance in observational studies (R²=16.77%) than in RCTs (R²=10.21%; Fisher’s z=−2.449, p=0.014). Adding demographic variables (patient age, patient sex, treatment duration) did not significantly improve results across any model or study type. The logistic regression identified study type as the strongest predictor of treatment response: RCT participants were more than twice as likely to respond compared to observational samples (OR=2.31, 95%-CI=[1.90–2.82], p<0.001), without significant effects for age, sex, and rating scale choice.

Conclusions

Baseline depression severity alone explains up to 16% of variance in treatment response, defining a minimum performance threshold that future predictive models must exceed. At the same time, our findings show that methodological choices shape the magnitude and structure of treatment response predictability. These findings have implications for precision psychiatry initiatives: predictive models incorporating novel biomarkers should not only outperform baseline severity but do so across heterogeneous designs and measurement contexts.

References

No references