Skip to content

Biomarkers and treatment prediction

TL;DR — No blood, imaging, genetic, electrophysiologic, microbiome, or digital marker is validated to diagnose MDD or select a treatment in routine care. Prospective biomarker meta-analysis finds many group-level associations but little reproducible individual prediction (Kennis 2020, PMID 31745238). Inflammation, cortisol, insulin resistance, EEG, imaging, cognition, and polygenic scores are biologically informative but overlap widely across diagnoses and healthy populations. Pharmacogenomic testing can reduce prescriptions with predicted drug–gene interactions, yet PRIME Care showed only a small, nonpersistent remission advantage (Oslin 2022, PMID 35819423). The field's core error is evaluating association in the same data used to build a model rather than external, decision-relevant prediction.

Candidate domains

Domain Candidate Recurrent problem
Inflammation CRP, IL-6, TNF confounding; no universal cutoff
Endocrine cortisol rhythm timing and state dependence
Neurotrophic BDNF peripheral measure ≠ brain mechanism
Metabolic insulin resistance, lipids obesity/medication confounding
Imaging connectivity, cortical thickness scanner/site and analytic instability
EEG alpha/theta, evoked potentials protocol heterogeneity
Genetics polygenic scores, CYP variants weak outcome prediction despite discovery-scale GWAS; ancestry portability (Howard 2019, PMID 30718901)
Digital sleep/activity/voice/typing privacy, drift, missingness, context

Prospective synthesis found no marker ready for clinical use (Kennis 2020, PMID 31745238). Cytokine-response meta-analysis identified candidates without a validated treatment selector (Liu 2020, PMID 31427752). Insulin resistance is elevated at group level but not specific to MDD (Fernandes 2022, PMID 35777578).

The absence claim was re-searched on 2026-08-30. One new exploratory study clustered baseline inflammatory proteins in 54 patients receiving iTBS and found differential response across two clusters, but it did not externally validate a locked classifier or randomize treatment by marker status (Pedraz-Petrozzi 2026, PMID 42364721). The dated evidence gap is therefore clinical utility—a replicated, prespecified treatment-by-marker interaction that improves outcomes—not the absence of any inflammation–stimulation association.

Reviews of treatment-outcome prediction consistently separate plausible clinical predictors from prescriptive biomarkers and find that replication is the bottleneck (Dunlop 2015, PMID 26289221). EEG markers have a long candidate history, but acquisition, preprocessing, and threshold heterogeneity have prevented routine use (Olbrich 2013, PMID 24151805).

Diagnostic versus predictive biomarkers

Type Question Required comparison
Diagnostic Does this person have MDD? MDD vs relevant differential diagnoses
Prognostic What happens regardless of treatment? outcome across care pathways
Predictive Which treatment works better for this person? treatment-by-marker interaction
Monitoring Is state changing? within-person calibrated trajectory

Many “response biomarkers” are prognostic: they predict improvement in every arm rather than preferential benefit from one treatment.

Validation ladder

  1. Prespecify marker, outcome, and threshold.
  2. Lock assay and preprocessing.
  3. Demonstrate technical reliability.
  4. Validate externally across sites and populations.
  5. Compare with simple clinical variables.
  6. Test whether marker-guided care improves outcomes in a randomized utility trial.
  7. Monitor calibration, harms, and subgroup equity after deployment.

Small neuroimaging or omic datasets with thousands of features invite leakage and overfitting. Cross-validation within one dataset is not external validation. Performance should report calibration and decision utility, not accuracy alone.

Case-control metabolomic work can distinguish groups within a dataset yet still fail the clinically harder comparison between unipolar and bipolar depression. A 2024 metabolomic study directly targeted that distinction, illustrating the appropriate differential-diagnosis benchmark while still requiring external utility validation (Tomasik 2024, PMID 37878349).

Pharmacogenomics

PRIME Care randomized pharmacogenomic-guided versus usual care. Testing reduced predicted drug–gene interactions; remission differences were small and not persistent at week 24 (Oslin 2022, PMID 35819423). CYP metabolism can inform exposure for selected drugs, but panel color categories do not validate a disease-mechanism match.

Suicide prediction

Prediction of rare outcomes is especially difficult. A risk model can have good discrimination yet produce many false positives. Cohort meta-analysis identifies prior attempt, severity, hopelessness and other risk factors, but no tool replaces direct assessment or system-level prevention (Li 2022, PMID 35101521).

Active translation studies

The live registry includes infliximab for inflammation/cognitive dysfunction (NCT06136546), precision treatment for first-episode depression (NCT05616559), L-DOPA challenge for psychomotor-response prediction (NCT06626152), exosomal microRNAs for suicidality/outcome (NCT05437588), and clinical utility of the EDIT-B blood test (NCT06507787), re-verified through exact ClinicalTrials.gov v2 records on 2026-08-30.

Prediction evidence deepening

A predictor can correlate with outcome without selecting between treatments. Clinical utility requires a prespecified treatment-by-marker interaction, locked model, external validation, calibration, and a decision showing better outcomes than usual care.

Candidate Recurrent signal Translation barrier
Pharmacogenomics Guided care reduces predicted drug–gene interactions; some meta-analyses find small remission gains Panel content, sponsorship, comparator quality, and persistence of benefit vary
EEG Meta-analyses identify candidate response features Pipelines, preprocessing, and thresholds are rarely externally locked
fMRI/connectomics Multivariate studies report above-chance prediction Small samples and leakage/overfitting reduce replication
Inflammatory markers Elevated markers define plausible subgroups Association does not show that marker-guided anti-inflammatory treatment helps
Clinical + multimodal ML CAN-BIND/STAR*D studies test transport across datasets Performance often falls under external validation and calibration is underreported
Digital phenotyping Passive measures could track within-person change Missingness, device drift, privacy, and false alerts are unresolved

The clinically relevant benchmark is incremental net benefit over symptom history, preference, comorbidity, and prior response—not statistical significance against zero.

Additional live-search evidence ledger

The records below were added after full PubMed E-utilities retrieval on 2026-08-30. The ledger states the evidentiary role of each record and preserves the design limitation that should travel with its citation.

  • Phaterpekar T 2023 — Machine Learning Prediction of Quality of Life Improvement During Antidepressant Treatment of Patients With Major Depressive Disorder: A STAR*D and CAN-BIND-1 Report. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Phaterpekar T 2023, PMID 37967350)

  • Karvelis P 2022 — Computational approaches to treatment response prediction in major depression using brain activity and behavioral data: A systematic review. Systematic review; useful for mapping consistency and gaps, not automatically a pooled causal estimate. (Karvelis P 2022, PMID 38800454)

  • Sajjadian M 2023 — Prediction of depression treatment outcome from multimodal data: a CAN-BIND-1 report. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Sajjadian M 2023, PMID 36004538)

  • He F 2026 — Machine Learning for Comparative Antidepressant Selection in Major Depressive Disorder: Systematic Review. Systematic review; useful for mapping consistency and gaps, not automatically a pooled causal estimate. (He F 2026, PMID 42126586)

  • Hu Y 2026 — Resting-state fMRI-based machine learning for predicting SSRI treatment response in major depressive disorder. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Hu Y 2026, PMID 42552532)

  • Burkhardt G 2026 — Cross-trial prediction of treatment response to transcranial direct current stimulation in patients with major depressive disorder. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Burkhardt G 2026, PMID 41456664)

  • Nunez JJ 2021 — Replication of machine learning methods to predict treatment outcome with antidepressant medications in patients with major depressive disorder from STAR*D and CAN-BIND-1. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Nunez JJ 2021, PMID 34181661)

  • Fan S 2020 — Pretreatment Brain Connectome Fingerprint Predicts Treatment Response in Major Depressive Disorder. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Fan S 2020, PMID 33458556)

  • Cohen SE 2023 — Electroencephalography for predicting antidepressant treatment success: A systematic review and meta-analysis. Meta-analysis; pooled estimates depend on eligibility, heterogeneity, and reporting bias. (Cohen SE 2023, PMID 36341804)

  • Poirot MG 2024 — Treatment Response Prediction in Major Depressive Disorder Using Multimodal MRI and Clinical Data: Secondary Analysis of a Randomized Clinical Trial. Randomized comparison; population, control credibility, duration, and missingness bound transportability. (Poirot MG 2024, PMID 38321916)

  • Esmaeilian Y 2026 — Predictive neuroimaging biomarkers of major depressive disorder treatment response: An umbrella review. Systematic review; useful for mapping consistency and gaps, not automatically a pooled causal estimate. (Esmaeilian Y 2026, PMID 41935973)

  • Ren C 2025 — Transcranial Electrical Stimulation in Treatment of Depression: A Systematic Review and Meta-Analysis. Meta-analysis; pooled estimates depend on eligibility, heterogeneity, and reporting bias. (Ren C 2025, PMID 40531534)

  • Widge AS 2019 — Electroencephalographic Biomarkers for Treatment Response Prediction in Major Depressive Illness: A Meta-Analysis. Meta-analysis; pooled estimates depend on eligibility, heterogeneity, and reporting bias. (Widge AS 2019, PMID 30278789)

  • Monn A 2025 — EEG vigilance and response to oral prolonged-release ketamine in treatment-resistant depression - A double-blind randomized validation study. Randomized comparison; population, control credibility, duration, and missingness bound transportability. (Monn A 2025, PMID 40446631)

  • Liu Y 2026 — Ketamine alters the aperiodic EEG exponent in major depression: implications for cortical E/I balance and treatment prediction. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Liu Y 2026, PMID 42403511)

  • Drevets WC 2022 — Immune targets for therapeutic development in depression: towards precision medicine. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Drevets WC 2022, PMID 35039676)

  • Busch Y 2019 — Blood-based biomarkers predicting response to antidepressants. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Busch Y 2019, PMID 29374800)

  • Nedic Erjavec G 2021 — Depression: Biological markers and treatment. Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Nedic Erjavec G 2021, PMID 33068682)

  • Juárez-Paredes FC 2026 — [Inflammatory depression: a review of advances in pathophysiology, clinical implications, and treatment]. Review-level synthesis; conclusions inherit limitations of the underlying designs. (Juárez-Paredes FC 2026, PMID 42447470)

  • Martin C 2015 — The inflammatory cytokines: molecular biomarkers for major depressive disorder? Primary or secondary empirical evidence; interpretation should follow its design and comparator rather than the headline alone. (Martin C 2015, PMID 24524646)

Open questions

  • Can any marker demonstrate a replicated treatment-by-marker interaction (Kennis 2020, PMID 31745238)?
  • What incremental utility do complex models add beyond symptoms, course, and prior treatment?
  • How can digital models remain calibrated after device and behavior change?
  • Can utility trials show better outcomes rather than better prescribing concordance (Oslin 2022, PMID 35819423)?

References

  1. Kennis M, et al. Prospective biomarkers of major depressive disorder. Molecular Psychiatry. 2020. PMID 31745238
  2. Liu JJ, et al. Peripheral cytokines and antidepressant response. Molecular Psychiatry. 2020. PMID 31427752
  3. Fernandes BS, et al. Insulin resistance in depression. Neuroscience and Biobehavioral Reviews. 2022. PMID 35777578
  4. Oslin DW, et al. PRIME Care pharmacogenomic randomized trial. JAMA. 2022. PMID 35819423
  5. Li X, et al. Predictors of suicidal outcomes in MDD. Journal of Affective Disorders. 2022. PMID 35101521
  6. Howard DM, et al. Genome-wide meta-analysis of depression. Nature Neuroscience. 2019. PMID 30718901
  7. Dunlop BW, Mayberg HS. Prediction of treatment outcomes in MDD. Expert Review of Clinical Pharmacology. 2015. PMID 26289221
  8. Olbrich S, et al. EEG biomarkers in MDD: discriminative power and prediction of treatment response. International Review of Psychiatry. 2013. PMID 24151805
  9. Tomasik J, et al. Metabolomic Biomarker Signatures for Bipolar and Unipolar Depression. JAMA Psychiatry. 2024. PMID 37878349
  10. Phaterpekar T, et al. Machine Learning Prediction of Quality of Life Improvement During Antidepressant Treatment of Patients With Major Depressive Disorder: A STAR*D and CAN-BIND-1 Report. The Journal of clinical psychiatry. 2023;85:23m14864. PMID 37967350
  11. Karvelis P, et al. Computational approaches to treatment response prediction in major depression using brain activity and behavioral data: A systematic review. Network neuroscience (Cambridge, Mass.). 2022;6:1066-1103. PMID 38800454
  12. Sajjadian M, et al. Prediction of depression treatment outcome from multimodal data: a CAN-BIND-1 report. Psychological medicine. 2023;53:5374-5384. PMID 36004538
  13. He F, et al. Machine Learning for Comparative Antidepressant Selection in Major Depressive Disorder: Systematic Review. JMIR mental health. 2026;13:e89352. PMID 42126586
  14. Hu Y, et al. Resting-state fMRI-based machine learning for predicting SSRI treatment response in major depressive disorder. BMC psychiatry. 2026;26:588. PMID 42552532
  15. Burkhardt G, et al. Cross-trial prediction of treatment response to transcranial direct current stimulation in patients with major depressive disorder. Progress in neuro-psychopharmacology & biological psychiatry. 2026;144:111600. PMID 41456664
  16. Nunez JJ, et al. Replication of machine learning methods to predict treatment outcome with antidepressant medications in patients with major depressive disorder from STAR*D and CAN-BIND-1. PloS one. 2021;16:e0253023. PMID 34181661
  17. Fan S, et al. Pretreatment Brain Connectome Fingerprint Predicts Treatment Response in Major Depressive Disorder. Chronic stress (Thousand Oaks, Calif.). 2020;4:2470547020984726. PMID 33458556
  18. Cohen SE, et al. Electroencephalography for predicting antidepressant treatment success: A systematic review and meta-analysis. Journal of affective disorders. 2023;321:201-207. PMID 36341804
  19. Poirot MG, et al. Treatment Response Prediction in Major Depressive Disorder Using Multimodal MRI and Clinical Data: Secondary Analysis of a Randomized Clinical Trial. The American journal of psychiatry. 2024;181:223-233. PMID 38321916
  20. Esmaeilian Y, et al. Predictive neuroimaging biomarkers of major depressive disorder treatment response: An umbrella review. Psychiatry and clinical neurosciences. 2026;80:459-468. PMID 41935973
  21. Ren C, et al. Transcranial Electrical Stimulation in Treatment of Depression: A Systematic Review and Meta-Analysis. JAMA network open. 2025;8:e2516459. PMID 40531534
  22. Widge AS, et al. Electroencephalographic Biomarkers for Treatment Response Prediction in Major Depressive Illness: A Meta-Analysis. The American journal of psychiatry. 2019;176:44-56. PMID 30278789
  23. Monn A, et al. EEG vigilance and response to oral prolonged-release ketamine in treatment-resistant depression - A double-blind randomized validation study. Psychiatry research. Neuroimaging. 2025;350:112001. PMID 40446631
  24. Liu Y, et al. Ketamine alters the aperiodic EEG exponent in major depression: implications for cortical E/I balance and treatment prediction. Therapeutic advances in psychopharmacology. 2026;16:20451253261462240. PMID 42403511
  25. Drevets WC, et al. Immune targets for therapeutic development in depression: towards precision medicine. Nature reviews. Drug discovery. 2022;21:224-244. PMID 35039676
  26. Busch Y, et al. Blood-based biomarkers predicting response to antidepressants. Journal of neural transmission (Vienna, Austria : 1996). 2019;126:47-63. PMID 29374800
  27. Nedic Erjavec G, et al. Depression: Biological markers and treatment. Progress in neuro-psychopharmacology & biological psychiatry. 2021;105:110139. PMID 33068682
  28. Juárez-Paredes FC, et al. [Inflammatory depression: a review of advances in pathophysiology, clinical implications, and treatment]. Revista medica del Instituto Mexicano del Seguro Social. 2026;64:e6996. PMID 42447470
  29. Martin C, et al. The inflammatory cytokines: molecular biomarkers for major depressive disorder? Biomarkers in medicine. 2015;9:169-80. PMID 24524646
  30. Pedraz-Petrozzi B, et al. Baseline inflammatory profiles in moderate-to-severe depression and differential response to intermittent theta-burst stimulation. Progress in Neuro-Psychopharmacology & Biological Psychiatry. 2026;148:111814. PMID 42364721