Blood-based biomarkers of antidepressant response to ketamine and esketamine: A systematic review and meta-analysis

Major depressive disorder (MDD) and bipolar depression are common, disabling illnesses for which conventional antidepressant treatments are often slow to act and ineffective in a substantial subset of patients. Ketamine (R,S‑ketamine) and its S‑enantiomer esketamine produce rapid antidepressant effects in many patients, but clinical response is variable and both agents have potential adverse effects. Given the heterogeneity of mood disorders and the trial‑and‑error nature of current prescribing, there is intense interest in biomarkers that might predict who will benefit from specific treatments. Medeiros and colleagues therefore conducted a preregistered systematic review and meta‑analysis to examine whether blood‑based biomarkers are associated with antidepressant response to ketamine and esketamine. The review addressed two related questions: whether baseline (pre‑treatment) biomarker levels predict later clinical response, and whether longitudinal changes in blood biomarkers after treatment are associated with clinical improvement. The authors focused on peripheral, blood‑based markers because these are more accessible and potentially cost‑effective than brain‑based measures for clinical use.

The review was registered on PROSPERO (ID: CRD42020210941) and followed PRISMA guidelines. Electronic searches were run in MEDLINE (PubMed) and Embase, supplemented by backward and forward manual searches, Google Scholar citation searches, grey literature, and selected studies obtained via personal communication. Studies were eligible if they met all of the following criteria: adult human participants (≥18 years); a current major depressive episode in MDD or bipolar disorder; English language; administration of at least one dose of ketamine or esketamine in a clinical trial (randomised or open‑label); measurement of depressive symptoms with standardised scales; and assessment of associations between blood‑based biomarkers and symptom improvement after treatment. Studies were excluded if they included current psychosis, depression attributable to other disorders, concurrent neuromodulatory treatments (except concurrent medications/psychotherapy were allowed), serious comorbid medical/neurological conditions as specified, perioperative/surgical settings, or naturalistic non‑treatment studies. Two investigators independently screened titles/abstracts and full texts; disagreements were resolved by consensus. Data extraction was performed by two independent investigators and verified. Risk of bias for included studies was assessed using the QUIPS tool across six domains, with each study classified as low, moderate, or high risk of bias. Primary outcomes comprised two association types: (1) baseline biomarker levels versus responder status, and (2) longitudinal changes (pre‑ versus post‑treatment) in biomarkers versus response. For quantitative synthesis the authors used responder versus non‑responder status as the primary outcome (responder = ≥50% improvement). Meta‑analyses were performed only when at least three comparable studies existed for the same biomarker. Random‑effects models (inverse‑variance weighting) were used to pool standardized mean differences (SMD) with 95% confidence intervals; SMD thresholds of 0.2, 0.5 and 0.8 were cited for small, medium and large effects. Heterogeneity was assessed with I2 and publication bias with funnel plots and Egger's test. Meta‑regressions and sensitivity analyses were planned to explore moderators (number of treatments, blood fraction, diagnosis, risk of bias) and the influence of individual studies; meta‑regressions were only conducted when there were ≥10 studies in a pooled analysis. Analyses were performed using R (metafor) and Review Manager 5.4.

The search retrieved 5611 citations; after screening 56 manuscripts (N = 2801 participants) met inclusion for qualitative synthesis and 26 provided data suitable for meta‑analysis. Fifty‑four studies (96%) examined racemic ketamine (all intravenous), while two studies (4%) examined intranasal esketamine. Across the included literature the authors identified data on 464 baseline biomarkers and 470 longitudinal biomarker changes. The most frequently studied groups were neurotrophic factors (15 studies), ketamine metabolites (13), inflammatory markers (12), tryptophan‑kynurenine metabolites (10), genetic markers (7) and amino acids (5). Most studies (74%) examined a single 0.5 mg/kg infusion delivered over 40 minutes; the most common response assessment timepoint was 1 day post‑infusion. After excluding overlapping samples, the combined response rate after a single 0.5 mg/kg infusion at 1 day was 46% (95/208), with no significant difference between MDD and bipolar depression at that timepoint. Risk of bias was rated moderate in 21 studies (37%) and high in 35 studies (63%). Common methodological problems were inadequate adjustment for confounders (concurrent medications, smoking, fasting status, time of day), incomplete participant description and missing biomarker data, and variability or under‑reporting of laboratory procedures. Neurotrophic factors: Fifteen studies addressed neurotrophic factors, especially BDNF (15 studies) and VEGF (5). In qualitative syntheses only two studies reported baseline BDNF predicting improvement; longitudinally, three studies reported nominally significant BDNF increases in association with clinical improvement but findings were inconsistent. Twelve studies were suitable for meta‑analysis (ten for BDNF, N = 332; three for VEGF, N = 154). Meta‑analysis found no significant association between baseline BDNF or VEGF and responder status. However, pooled longitudinal data (11 studies, N = 331) showed that responders had a small but statistically significant increase in peripheral BDNF post‑treatment compared with baseline (SMD = 0.26, 95% CI 0.03 to 0.48, p = 0.02), whereas non‑responders did not show a significant change (SMD = 0.05, 95% CI −0.19 to 0.28, p = 0.70). There was no detectable between‑study heterogeneity for the BDNF longitudinal analysis (I2 = 0) and no evidence of publication bias by Egger's test (p = 0.58). Subgroup inspection suggested larger BDNF increases in studies with multiple infusions (SMD = 0.35, 3 studies) than single‑infusion studies (SMD = 0.16, 8 studies), but the difference was not statistically significant (β = 0.19, p = 0.41). Ketamine and metabolites: Thirteen studies measured ketamine and metabolites (ketamine, norketamine, DHNK, selected HNKs). Ten studies reported no significant associations; two small sets of studies suggested possible associations (higher ketamine in two studies, lower (2R‑6R)‑HNK in two studies), but meta‑analysis of nine studies (N = 286) did not find a significant relationship between blood levels of ketamine or norketamine and responder status. Inflammatory markers: Twelve studies evaluated immune‑related biomarkers (interleukins, TNF‑α, CRP). Eleven reported baseline associations with inconsistent findings; only three of ten longitudinal studies found an association between symptom improvement and reductions in pro‑inflammatory markers, and no single inflammatory marker showed replication across multiple studies. Meta‑analysis of six studies (N = 260) showed a non‑significant tendency for responders to have lower baseline pro‑inflammatory levels; CRP had the strongest baseline association but was non‑significant (SMD = −0.28, 95% CI −0.67 to 0.10, p = 0.15). No significant longitudinal changes in inflammatory markers were observed in pooled analyses. Tryptophan‑kynurenine pathway: Ten studies examined metabolites (tryptophan, kynurenine, kynurenic acid, quinolinic acid and ratios). Only one of seven baseline studies reported a significant association—lower baseline KynA/QA ratio associated with greater improvement (reported p = 0.016). No consistent longitudinal patterns emerged. Genetic markers, amino acids and others: Seven genetic studies (including candidate SNP and small GWAS samples) produced mixed results: one small study reported greater improvement in Val/Val BDNF carriers (F = 5.59, p = 0.0007) while another failed to replicate this. GWAS samples were underpowered and yielded no genome‑wide significant hits. Five studies of amino acids and various miscellaneous biomarker studies (micronutrients, gene expression, targeted proteins, metabolomics) produced inconsistent findings without replication across studies. Esketamine: Only two studies addressed esketamine biomarkers. Rotroff et al. used metabolomics in 20 participants and reported no significant metabolite correlations with response. Li et al. conducted a GWAS in 527 participants receiving multiple esketamine infusions and reported associations with IRAK3 and NME7, but did not observe an association with the BDNF Val66Met polymorphism.

Medeiros and colleagues interpret their results as indicating limited and inconsistent evidence that blood‑based biomarkers predict antidepressant response to ketamine or esketamine. Across a large and growing literature (>460 distinct biomarkers investigated), no baseline peripheral biomarker showed consistent predictive value. The principal positive finding from pooled longitudinal data was a small increase in peripheral BDNF among responders but not non‑responders, which is congruent with preclinical hypotheses that ketamine enhances synaptogenesis via BDNF‑related pathways. The authors note, however, that it remains uncertain how well peripheral BDNF reflects central nervous system BDNF or synaptic changes in humans. The pooled analysis did not support a relationship between measured blood levels of ketamine or norketamine and clinical response, suggesting limited immediate clinical utility in therapeutic drug monitoring of these analytes. Similarly, the review did not find reproducible evidence that baseline or post‑treatment pro‑inflammatory markers reliably predict benefit from ketamine, despite preclinical rationale for anti‑inflammatory effects. Key limitations acknowledged by the authors include generally small sample sizes (median N = 33), incomplete control for confounders (concurrent medications, smoking, fasting, time of day), inconsistent reporting of assay reliability and laboratory procedures, and marked heterogeneity across studies in design, participant characteristics, treatment schedules and outcome timepoints. The QUIPS assessment found that most studies had moderate to high risk of bias. The authors therefore call for larger, better controlled studies, standardised laboratory and clinical protocols, and increased data sharing to enable individual‑level pooled analyses. They caution that current evidence does not support clinical implementation of blood‑based biomarkers for selecting patients for ketamine or esketamine treatment, although peripheral BDNF increases merit further investigation as a potential dynamic marker of treatment effect.

The authors conclude that despite extensive investigation, there is limited evidence that blood‑based biomarkers can predict ketamine or esketamine treatment efficacy in depression. The single replicated finding from pooled longitudinal data is a small post‑treatment increase in peripheral BDNF among responders; no other biomarker, including ketamine metabolites and inflammatory markers, showed consistent associations across studies. Given the high risk of bias and methodological heterogeneity in the existing literature, the authors recommend larger, more rigorous trials with comprehensive control of confounders before blood‑based biomarkers can be considered clinically useful for guiding ketamine or esketamine treatment.