Pharmacogenomics for stratified antidepressant treatment in major depressive disorder: evidence, limits and a roadmap for clinical use Authors: Bernhard T Baune Published: 2026 ABSTRACT ABSTRACTMajor depressive disorder (MDD) is treated largely by trial and error, despite marked interindividual variation in antidepressant exposure, efficacy and tolerability. Pharmacogenomics (PGx) has therefore been proposed as a route to stratified antidepressant treatment. This Review asks whether current evidence supports that claim by distinguishing actionable drug-gene guidance, clinical utility in broad trial populations and subgroup evidence for stratified care. Contemporary guidance supports genotype-informed prescribing for selected antidepressants, most consistently through CYP2D6 and CYP2C19 and, for selected serotonin reuptake inhibitors, CYP2B6, whereas SLC6A4 and HTR2A are not clinically actionable for antidepressant prescribing. Randomized and pragmatic trials show that PGx reduces drug-gene mismatched prescribing more reliably than symptom outcomes in unselected MDD samples. Meta-analyses suggest small-to-moderate short-term response or remission gains, with larger signals in treatment-exposed, severe or difficult-to-treat patients. PGx is therefore not simply weakly useful for everyone; its strongest current role is selective, subgroup-enriched prescribing where exposure mismatch is plausible. A practical roadmap is to restrict current interpretation to guideline-supported CYP2D6/CYP2C19/CYP2B6 gene-drug pairs, prioritize testing before affected antidepressants or after nonresponse, intolerance or polypharmacy, avoid broad proprietary class assignment, and validate stratified claims in prospective enriched trials.KEYWORDS: Major depressive disorder, antidepressants, pharmacogenomics, CYP2D6, CYP2C19, CYP2B6, stratified medicine, precision psychiatry Introduction Major depressive disorder (MDD) remains a paradigmatic example of therapeutic heterogeneity. In routine practice, initial antidepressant selection is often driven by prior exposure, adverse-effect profile, comorbidity, drug-drug interaction burden and clinician preference rather than by validated biological markers. The consequences of this trial-and-error model are familiar: delayed remission, serial switching, cumulative adverse effects, and loss of confidence in treatment. In STAR*D, only about one-third of patients achieved remission after first-step treatment, and outcomes worsened across subsequent treatment steps [1]. The clinical and economic burden is greatest in treatment-resistant depression, where repeated failed trials accumulate costs for patients and health systems [2]. Pharmacogenomics (PGx) has long been proposed as one way to make antidepressant prescribing more rational. The appeal is obvious: inherited variation in drug-metabolizing enzymes and, less consistently, drug targets or transporters may alter exposure, tolerability and response. Psychiatry also contains some of the clearest examples of clinically relevant drug-gene interactions in routine prescribing, particularly through CYP2D6 and CYP2C19 [3–7]. However, commercial and clinical enthusiasm has often exceeded the evidence. Claims that multigene tools can identify the best antidepressant for most patients before treatment begins require a higher evidentiary standard than claims that a CYP phenotype can guide dosing or avoidance for a particular drug. The central question for Pharmacogenomics is therefore not whether genes influence antidepressant pharmacology - they do - but whether current evidence supports a genuinely stratified treatment approach in depression. In this Review, PGx-guided care refers to using genotype information to inform antidepressant selection, dose, monitoring, or avoidance of drug-gene mismatch. PGx-guided prescribing can therefore be clinically useful without constituting a full stratified-treatment model. True stratified treatment requires a stronger claim: a prospectively identifiable subgroup, defined by a biologically interpretable PGx liability and clinical context, follows a different prescribing pathway because it has a higher probability of benefit or lower probability of harm than unstratified care. This Review argues that current evidence supports selective pharmacokinetic stratification rather than universal PGx testing or broad panel-based assignment. The strongest current stratum is the patient in whom genotype, co-medication profile or clinical history makes clinically relevant exposure mismatch plausible for an antidepressant with guideline-supported gene-drug recommendations. Figure 1 summarizes the distinction between what is actionable now, what remains unsupported and the roadmap needed for a more mature stratified model. Pharmacogenomics for stratified antidepressant treatment: actionable evidence, unsupported claims and next-step roadmap. Left: current evidence supports transparent, guideline-anchored interpretation of CYP2D6, CYP2C19 and CYP2B6 for specific antidepressants when pharmacokinetic liability or exposure mismatch is clinically plausible. Middle: current evidence does not support SLC6A4 or HTR2A for clinical antidepressant selection, nor broad proprietary panel categories as class-wide treatment allocation rules. Right: a mature stratified approach requires prospective enriched trials in difficult-to-treat or treatment-resistant MDD, standardized PGx reporting, phenotype-aware interpretation, diverse cohorts and durable clinical endpoints. What would count as evidence for a stratified pharmacogenomic approach? The evidentiary framework used here separates three levels by study design and by the question each level answers. Level 1 comprises drug-gene actionability evidence, including pharmacokinetic studies, replicated exposure or safety data, standardized genotype-to-phenotype translation and CPIC/DPWG-style prescribing guidelines; it asks whether genotype should modify use of a specific drug when known. Level 2 comprises randomized, pragmatic and prospective controlled trials, together with meta-analyses; it asks whether providing PGx information changes prescribing behavior or clinical outcomes in tested populations. Level 3 comprises prospectively specified subgroup, enrichment or treatment-by-subgroup evidence; it asks whether a defined subgroup should follow a different treatment pathway. Level 1 supplies mechanistic prescribing rules, Level 2 tests average clinical utility, and Level 3 is required for a mature stratified-treatment claim. These levels should not be conflated. Strong Level 1 evidence can exist without Level 2 symptom benefit if many trial participants are taking drugs unaffected by the tested genes, if standard clinical care already compensates for adverse effects, or if the test result rarely changes the prescription. Conversely, a modest Level 2 average effect can coexist with a stronger Level 3 signal when benefit is concentrated among patients with actionable mismatch, treatment difficulty or high pretest probability of pharmacokinetic liability. Exploratory Level 3 signals cannot alone establish stratification; they identify subgroups that should be prospectively tested. In depression, the most clinically credible PGx strata are not arbitrary vendor-defined color categories, but biologically interpretable groups: patients with genotypes that predict substantially altered exposure; patients receiving or being considered for antidepressants with guideline-supported gene-drug recommendations; and patients in whom prior nonresponse, intolerance, polypharmacy or likely phenoconversion raises the pretest probability that exposure mismatch is clinically relevant. Table 1 separates these evidence levels and clarifies what each level contributes to the final takeaway. Table 1.Evidentiary levels by design and interpretive contribution for antidepressant PGx stratification.Evidence level and source/designQuestion answeredWhat this evidence addsWhat it cannot showCurrent implicationLevel 1: drug-gene actionabilityPK studies, exposure/safety data, genotype-to-phenotype translation, CPIC/DPWG recommendationsDoes genotype alter exposure, dose, safety or drug choice for a specific antidepressant?Defines transparent drug-gene pairs for prescribing when genotype is known; strongest for CYP2D6/CYP2C19 and selected CYP2B6.Does not prove that testing improves outcomes in broad MDD populations or identify who should be tested.Necessary foundation; supports targeted pharmacokinetic stratification but is not sufficient alone.Level 2: clinical utilityRandomized trials, pragmatic trials, prospective controlled studies and meta-analysesDoes providing PGx information improve prescribing-process or symptom outcomes compared with unguided care?Shows that PGx can reduce drug-gene mismatch and may modestly improve remission or response.Average effects may be diluted; cannot prove subgroup-specific pathways if not prospectively stratified.Supports conditional utility, especially when PGx changes prescribing.Level 3: stratified treatmentProspectively defined enriched subgroups, treatment-by-subgroup interactions and replicated effect-modifier analysesDoes a defined PGx/clinical subgroup benefit enough to follow a different treatment pathway?Identifies whom PGx is most likely to help, such as patients with actionable mismatch, prior intolerance/nonresponse or phenoconversion risk.Post hoc or exploratory subgroup signals cannot establish mature stratified treatment alone.Currently suggestive but incomplete; strongest for pharmacokinetic liability and probable exposure mismatch.Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Evidentiary levels by design and interpretive contribution for antidepressant PGx stratification. Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Table 1 is submitted separately and summarizes the evidence-source/design, question answered, contribution and limitation of each level. Evidence considered in this review This manuscript is a critical narrative Review rather than a formal systematic review or meta-analysis. The evidence base prioritized for discussion comprises consensus statements and prescribing guidelines relevant to antidepressant PGx implementation, prospective controlled clinical trials and pragmatic effectiveness trials of PGx-guided prescribing in depression, recent systematic reviews and meta-analyses addressing efficacy and subgroup effects, and studies comparing commercial panel recommendations with transparent guideline-based interpretation. The emphasis is on adult MDD because that is where most implementation claims are made. Selected adolescent and cross-population studies are included when they clarify generalizability, developmental limitations or the maturity of subgroup evidence. Priority is given to prospective evidence, guideline-supported markers and clinically interpretable drug-gene pairs, because the review question concerns stratified treatment rather than biological association alone. The established evidence base: actionable pharmacokinetic pharmacogenomics The clearest evidence in antidepressant PGx concerns drug metabolism rather than broad symptom-response prediction. Across recent consensus and guideline documents, CYP2D6 and CYP2C19 consistently emerge as the principal genes with sufficient evidence to influence prescribing for selected antidepressants [3–7]. The mechanistic basis is straightforward: altered enzyme activity can meaningfully change parent-drug and metabolite exposure, thereby affecting dose requirements, adverse effects and sometimes the probability that a patient receives an apparently adequate but pharmacologically mismatched trial. For depression, this matters most when clinicians are prescribing antidepressants for which CPIC or DPWG recommendations already exist, including selected serotonin reuptake inhibitors and tricyclic antidepressants [4–7]. In that context, PGx is not functioning as a vague precision-medicine ideal; it is functioning as a drug-exposure management tool. The clinical question becomes whether a given patient is likely to be underexposed, overexposed or at disproportionate risk of dose-related tolerability problems at standard dosing. That is a narrower claim than predicting overall antidepressant response, but it is also the part of the field with the strongest translational footing. The experience with tricyclic antidepressants is instructive even if contemporary prescribing has shifted toward newer agents. For tricyclics, the connection between CYP2D6/CYP2C19 phenotype, exposure and toxicity is sufficiently strong that genotype-informed dosing has become one of the clearest implementation examples in psychiatric PGx [7]. That precedent matters conceptually because it shows that antidepressant PGx can move from association to prescribing rule when the pharmacology is well defined and the clinical consequences of over- or underexposure are tractable. The 2023 CPIC update widened the frame without eroding evidentiary discipline. In addition to CYP2D6 and CYP2C19, it considered CYP2B6 for selected serotonin reuptake inhibitors, while explicitly concluding that SLC6A4 and HTR2A did not currently support clinical antidepressant prescribing recommendations [4]. That conclusion is central to any stratification claim: not every repeatedly studied marker matures into a clinically defensible subgrouping variable. Implementation also depends on standardized genotype-to-phenotype translation. Consensus work on pharmacogenetic terminology, CYP2D6 phenotype assignment and PGx implementation has reduced interpretive ambiguity [8–10]. For stratified care, the relevant unit is therefore not merely the genotype printed on a report, but the effective phenotype in a specific patient at a specific prescribing decision. Co-medications, CYP inhibition or induction, hepatic impairment, adherence, smoking and other clinical factors can produce phenoconversion that makes a genotype-only interpretation incomplete. Table 2 summarizes the principal PGx domains according to clinical actionability, relevance to stratified prescribing and current limitations. Table 2.Pharmacogenomic domains relevant to a stratified approach in depression.Marker/domainRepresentative drugs or contextWhat the evidence supportsRelevance to stratified careMain limitationCYP2D6Paroxetine, fluvoxamine, venlafaxine, vortioxetine and tricyclic antidepressants where relevant [4,7].Substantial evidence that altered phenotype changes exposure and may require dose adjustment, alternative selection or monitoring for selected drugs [4,7].Core current PGx stratum when a contemplated or current drug is CYP2D6-relevant and exposure mismatch is plausible.Does not by itself predict broad antidepressant response across all MDD patients; phenoconversion must be considered.CYP2C19Citalopram, escitalopram, sertraline and other selected antidepressants [4–6]Strong evidence for genotype-informed dosing or avoidance of mismatch for selected drugs [4–6].Core component of narrow pharmacokinetic stratification in depression.Clinical utility is greatest when linked to a specific drug decision rather than broad panel interpretation.CYP2B6Sertraline and selected serotonin reuptake inhibitor contexts considered in the 2023 CPIC update [4].Actionable relevance for selected drugs, but with a narrower implementation history than CYP2D6/CYP2C19.Adjunct to pharmacokinetic stratification when drug-specific CYP2B6 evidence is present.More limited scope and smaller evidence base than the core CYP2D6/CYP2C19 markers.SLC6A4 and HTR2ACommonly marketed pharmacodynamic markers in commercial panels [4,28]Reviewed repeatedly, but current evidence does not support clinical antidepressant prescribing recommendations [4].Not currently validated stratifiers for routine clinical antidepressant selection.Association signals have not translated into consistent, reproducible prescribing rules.Other candidate markersABCB1, CRHBP, PK-PD interaction signals after prior antidepressant failure [12–14]Scientifically interesting and potentially hypothesis-generating.Possible future refinement of stratification if independently replicated.Insufficient replication, standardization and actionability for routine use.Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Pharmacogenomic domains relevant to a stratified approach in depression. Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Why pharmacodynamic markers have not yet matured into stratifiers Much of the early enthusiasm surrounding psychiatric PGx rested on pharmacodynamic candidate genes such as SLC6A4 and HTR2A, or on broader claims that composite algorithms could predict symptom response. However, the strongest contemporary guidance does not support those markers for routine antidepressant selection [4]. The problem is not a total absence of signal, but rather the absence of a sufficiently consistent, clinically actionable signal across replicated studies, drug classes and populations. Other candidate loci remain scientifically interesting. Earlier pharmacogenetic reviews highlighted candidate pathways with potential clinical relevance [11]. Signals involving ABCB1, corticotropin-releasing hormone binding protein and pharmacokinetic-pharmacodynamic interactions after prior antidepressant failure suggest that biology beyond CYP-mediated metabolism could eventually refine treatment assignment [12–14]. Yet most findings remain too context-specific, insufficiently replicated or weakly standardized for current clinical stratification. In practical terms, the field’s evidentiary center of gravity still lies with pharmacokinetic genes. This distinction between actionability and stratification clarifies why the depression PGx field has sometimes talked past itself. Critics of commercial multigene tests are often correct about overstatement, yet that critique does not negate the reality of specific high-confidence pharmacokinetic interactions. Conversely, proponents of PGx are correct that antidepressant response is partly shaped by exposure biology, but that truth does not automatically validate every response-prediction claim layered on top of it. A scientifically serious stratified model must preserve this distinction. Prospective clinical trial evidence Prospective trials anticipated the later controversy because their findings have been directionally positive but heterogeneous. In an early randomized double-blind Spanish study, PGx-guided treatment improved response in MDD at 12 weeks, with benefit dependent on the number of previously failed medication trials [15]. An 8-week single-blinded study also reported improved depressive symptoms and response relative to usual treatment [16]. By contrast, a participant- and rater-blind trial by Perlis and colleagues did not show a significant advantage for test-guided care [17], and a randomized trial in adolescent depression failed to demonstrate clear benefit [18]. Taken together, these studies suggested that PGx could influence outcomes under some circumstances, but also that any effect was unlikely to be large, universal or independent of study design. The GUIDED trial remains pivotal because it was large, patient- and rater-blinded, and explicitly designed around combinatorial PGx-guided treatment in adults with difficult-to-treat depression [19]. Its main result is often simplified in ways that distort the evidence. The predefined primary endpoint – mean symptom improvement at 8 weeks – was not significantly different between guided care and treatment as usual [19]. However, secondary endpoints for response and remission favored guided care, and post hoc analyses suggested that advantage was more pronounced in patients taking medications with actionable gene-drug interactions [20,21]. GUIDED therefore supports two narrower propositions more strongly than one broad proposition. It supports that PGx information can be clinically useful in treatment-exposed outpatients, and that benefit is biologically more plausible when there is a definable drug-gene mismatch. It does not support the stronger claim that combinatorial PGx uniformly improves depressive symptoms across all patients with MDD. For the present Review, GUIDED is best read as evidence for conditional rather than universal utility. The Canadian GAPP-MDD trial strengthened the impression that PGx-guided care might have signal in treatment-exposed adults, but also illustrated the evidentiary fragility of the field [22]. The study was patient- and rater-blinded and directionally favored guided prescribing across several clinical outcomes, yet recruitment constraints left it underpowered for definitive inference. GAPP-MDD is therefore consistent with GUIDED without resolving the central uncertainty: whether observed benefits reflect a reproducible subgroup effect or a more diffuse and context-dependent average effect. PRIME Care is arguably the most policy-relevant antidepressant PGx trial because it tested pharmacogenomic guidance in a large pragmatic Veterans Affairs setting rather than an efficacy-style trial environment [23]. Its most robust finding was not a dramatic symptom advantage, but a prescribing-process effect: PGx testing reduced the likelihood that clinicians selected medications with substantial predicted drug-gene interactions. Over 24 weeks, remission was somewhat more likely among tested patients overall, but the difference was small and was not significant at the principal 24-week endpoint [23]. This pattern is exactly what one would expect if PGx is most valuable in a subset of patients with actionable mismatch rather than as a universal front-end treatment selector. A more recent single-center randomized trial in depressive disorder reported higher remission and response rates and fewer adverse drug reactions with PGx-guided care at 8 and 12 weeks [24]. This finding adds contemporary positive evidence, but its open-label clinical design, single-center setting and use of a specific testing approach limit its ability to settle the stratification question. The larger evidentiary pattern remains that PGx guidance has a clearer and more reproducible effect on prescribing-process measures than on average symptom outcomes across all comers. Table 3 maps the main prospective trials according to design, key finding and what each adds to the evidence hierarchy, emphasizing the difference between prescribing-process effects, symptom outcomes and subgroup evidence. Table 3.Prospective controlled trials most relevant to stratified antidepressant PGx.StudyDesign/populationKey findingWhat this adds to stratification evidencePérez et al. [15]12-week randomized double-blind adult MDD trial of PGx-guided versus usual treatment.Reported improved response, with benefit dependent on the number of previously failed medication trials.Supports the hypothesis that treatment-exposed patients may be a higher-yield PGx subgroup.Han et al. [16]8-week single-blinded adult MDD trial.Reported improved symptoms and response in the guided arm.Supports proof of concept, but design limits confidence and generalizability.Perlis et al. [17]Participant- and rater-blind adult MDD trial.Did not show a significant clinical advantage for PGx-guided care.Demonstrates that benefit is not universal across blinded studies.GUIDED [19]Large patient- and rater-blinded adult outpatient trial in difficult-to-treat depression.Primary symptom-change endpoint negative; response and remission secondary outcomes favored guided care; larger benefit in actionable gene-drug interaction subgroup [20,21].Best interpreted as evidence for conditional utility concentrated in biologically plausible subgroups.GAPP-MDD [22]Patient- and rater-blinded Canadian randomized trial.Directionally favored guided care, but underpowered for definitive inference.Consistent with subgroup-sensitive benefit, but not conclusive.PRIME Care [23]Large pragmatic Veterans Affairs randomized clinical trial.Reduced prescribing of medications with substantial predicted drug-gene interactions; remission benefit small overall and not significant at the principal 24-week endpoint.Strongest real-world evidence that PGx changes prescribing, while average symptom gains remain modest.Vande Voort et al. [18]Randomized adolescent depression trial of combinatorial PGx testing.Did not demonstrate clear clinical benefit.Cautions against assuming generalizability across age groups and settings.Xu et al. [24]Single-center randomized clinical trial in depressive disorder.Reported higher remission and response rates and fewer adverse drug reactions with PGx-guided treatment at 8 and 12 weeks.Adds contemporary positive evidence, but single-center and nonblinded features limit stratification inference.Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Prospective controlled trials most relevant to stratified antidepressant PGx. Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Meta-analytic evidence and why reviews disagree Recent evidence syntheses generally point in the same direction, but not to the same level of confidence. In a meta-analysis restricted to prospective controlled trials, Brown and colleagues reported a pooled remission benefit for PGx-guided care, with larger effects in studies enrolling patients with greater baseline severity or more previous antidepressant failures [25]. Those observations are important because they imply that average effects may be diluted when trials enroll many patients with low pretest probability of actionable mismatch. An updated umbrella review and meta-analysis by Tesfamicael and colleagues likewise concluded that PGx-guided antidepressant prescribing improved remission and response outcomes, while noting material heterogeneity across platforms, outcomes and study designs [26]. A 2025 subgroup and cumulative meta-analysis extended that picture by reporting moderate benefits at 8 and 12 weeks, uncertain long-term remission advantage, and subgroup signals suggesting stronger effects in studies conducted in Asian countries and in difficult-to-treat or treatment-resistant samples [27]. Those more optimistic quantitative syntheses coexist with more skeptical interpretive reviews. An American Psychiatric Association workgroup review concluded that available evidence does not support currently marketed combinatorial PGx tools for broad treatment selection in MDD [28]. Frye and Nemeroff similarly argued that the field has learned important lessons but still lacks the prospectively enriched, phenotype-aware studies needed to justify strong clinical claims [29]. Earlier reviews reached analogous cautionary conclusions, especially when cost-effectiveness claims outpaced the consistency of blinded trial data [30,31]. The apparent contradiction between positive meta-analyses and cautious expert interpretation is less paradoxical than it seems. Meta-analyses ask whether an average effect exists across included trials; they do not automatically validate the interchangeability of assays, the transportability of proprietary algorithms or the clinical importance of modest short-term gains. In antidepressant PGx, the magnitude and durability of benefit are shaped by what is being aggregated: blinded versus unblinded designs, actionable versus nonactionable panels, all-comer versus treatment-exposed populations, and symptom versus prescribing-process outcomes. The question for stratification is therefore narrower than ‘does PGx ever help?’ and more demanding than ‘is the pooled effect estimate greater than 1?’ Comparator quality also matters. A trial in which usual care consists of minimally structured prescribing is not asking the same clinical question as one in which both arms receive careful measurement-based management with expert dose adjustment. PGx is more likely to appear beneficial when it compensates for otherwise inconsistent pharmacologic decision-making, and less likely to appear transformative when usual care is already rigorous. Stratification claims should therefore be judged against the quality of the comparator, not only the nominal randomization design. Does current evidence justify a stratified approach? Table 4 translates the review question into operational, evidence-based criteria for calling antidepressant PGx a stratified approach rather than merely an aid to prescribing. The most important criteria are a biologically actionable drug-gene mechanism, a prospectively identifiable subgroup with high prior probability of PGx relevance, replicated evidence of differential benefit within that subgroup, clinically meaningful outcome gains, transparent guideline-anchored interpretation, and reproducibility across populations and care settings [4,23,27,28]. Table 4.Evidence-based criteria supporting a stratified PGx approach in depression.CriterionWhy it mattersCurrent evidence in depression PGxCurrent status and implicationDrug-specific actionable pharmacokinetic mechanismStratification is most credible when genotype plausibly alters exposure or tolerability for a specific antidepressant.CYP2D6 and CYP2C19, and selected CYP2B6 pairs, support prescribing changes for several antidepressants; SLC6A4 and HTR2A do not support prescribing use [4,6,28].SupportedProspectively identifiable high-liability subgroupThe putative stratum should be recognizable before prescribing or switching.Patients with actionable drug-gene mismatch or substantial predicted drug-gene interactions are identifiable and appear to have the highest prior probability of benefit [19,20,23].Partially supportedReplicated differential benefit within the subgroupA stratum requires more than an overall mean effect; subgroup-by-treatment interaction should recur across studies.Signals exist in actionable-mismatch and difficult-to-treat subgroups, but most are post hoc or meta-analytic rather than prospectively validated [20,23,25,27].Partially supportedClinically meaningful benefit beyond prescribing-process measuresThe subgroup should show improved remission, tolerability, switching, adverse-effect burden or durability, not only fewer mismatched prescriptions.PGx more consistently reduces drug-gene mismatch than it produces large or durable symptom benefits; remission gains are modest and often early [23,26,27].Partially supportedTransparent, guideline-anchored and reproducible interpretationA clinically usable stratum should be interpretable and portable, not dependent on noninterchangeable vendor algorithms.The strongest support is for CPIC/DPWG-based CYP interpretation; commercial combinatorial tools remain heterogeneous and noninterchangeable [28,35–40].Supported narrowlyExternal validity across populations/settings and confirmation in enriched trialsA stratifier should reproduce across ancestries, care settings and trials enriched for the putative subgroup.Stronger effects in some Asian-country and difficult-to-treat analyses require validation; representation remains limited, and enriched randomized trials are still largely absent [27,32–34].Not yet supportedAbbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. Evidence-based criteria supporting a stratified PGx approach in depression. Abbreviations: CPIC, Clinical Pharmacogenetics Implementation Consortium; DPWG, Dutch Pharmacogenetics Working Group; MDD, major depressive disorder; PGx, pharmacogenomics; PK-PD, pharmacokinetic-pharmacodynamic. When the current literature is judged against these criteria, the case for stratification is narrow but real. The best-supported stratum remains patients whose contemplated or current antidepressant creates an actionable pharmacokinetic liability, most commonly through CYP2D6 or CYP2C19 and, in selected contexts, CYP2B6. In this setting, the PGx result can plausibly change dose, monitoring, avoidance of mismatch or selection among otherwise reasonable antidepressants. Table 4 operationalizes the criteria for judging whether antidepressant PGx evidence supports a true stratified-treatment claim rather than merely a general prescribing aid. The modest average effects in unselected MDD samples and the larger signals in severe, treatment-exposed or difficult-to-treat groups are not contradictory. All-comer trials dilute benefit because many patients have no actionable genotype, are not being prescribed an exposure-sensitive antidepressant, have drug choices unaffected by the panel, or already receive close clinical management. The more coherent interpretation is subgroup-dependent utility: PGx is most likely to matter when prior nonresponse, dose-limiting intolerance, multiple medication trials, polypharmacy, phenoconversion or a planned affected antidepressant increases the probability of exposure mismatch. This subgroup is not limited to textbook poor or ultrarapid metabolizers. In practice, it also includes patients whose effective phenotype may be altered by co-medications, patients with marked intolerance at standard doses, and patients who appear not to have received a pharmacologically adequate exposure despite nominally adequate dosing. Real-world stratification therefore requires phenotype-aware interpretation rather than simple genotype lookup alone [29]. A second plausible stratification axis is prior treatment difficulty. Both clinical intuition and meta-analytic data suggest that PGx has a higher pretest probability of clinical usefulness after repeated nonresponse, dose-limiting adverse effects or multiple medication switches. Brown and colleagues observed larger effects in more treatment-exposed and more severely ill samples, and the 2025 subgroup meta-analysis reported stronger signals in difficult-to-treat and treatment-resistant populations [25,27]. Clinically, prior failures can justify a lower threshold for using PGx, but they do not yet define a fully validated pharmacogenomic treatment pathway on their own. Population differences in pharmacogene frequencies are real, and antidepressant outcomes can vary with ancestry, prescribing context and environment [32,33]. Yet the evidentiary base in depression PGx remains unbalanced. A recent systematic review of sample representation found substantial overrepresentation of White and Asian participants and underrepresentation of Black, Hispanic/Latino and Indigenous populations [34]. For that reason, subgroup findings suggesting larger effects in studies from Asian countries should be interpreted cautiously [27,34]. They may reflect true population pharmacology, but they may also reflect differences in trial design, available drugs, background prescribing patterns or platform mix. The current literature does not justify using ancestry or geography as a stand-alone clinical stratifier for antidepressant PGx. One of the most important practical barriers to a stratified PGx framework is that currently marketed panels are heterogeneous in gene content, phenotype translation, weighting rules and recommendation logic [35–38]. Recent comprehensive reviews likewise show that reported antidepressant PGx associations far outnumber findings that are replicated, clinically interpretable and ready for prescribing [39]. Even comparisons between CPIC and DPWG guidance emphasize that implementation requires careful attention to guideline scope and harmonization across systems [40]. In some comparisons, commercial recommendations diverge meaningfully from transparent CPIC-based interpretations [35,36]. That matters because a positive trial of one proprietary tool cannot be assumed to validate another. Stratification should rest on reproducible biology and transparent interpretive rules, not on the brand recognition of a commercial assay. The most defensible answer to the central question is therefore a qualified yes. Current evidence supports selective PGx use for actionable pharmacokinetic strata, especially when CYP2D6, CYP2C19 or CYP2B6 status is likely to change exposure to the antidepressant under consideration. It does not support broad proprietary panel categories as class-wide allocation rules or first-line universal treatment selection for all patients with MDD. Why the evidence remains incomplete Several recurring design features explain why evidence for stratified PGx in depression remains incomplete even though drug-gene actionability is real. First, many trials have enrolled broad all-comer MDD samples rather than patients prospectively enriched for high PGx relevance. If a substantial fraction of participants are not taking drugs materially affected by the tested genes, or would do well regardless of genotype, true subgroup benefit will be diluted toward the null [23,28,29]. Outcome selection also matters. Many studies prioritize symptom score change at 8 weeks, but a stratified prescribing tool may exert its most meaningful effects on different outcomes: fewer intolerable trials, less premature discontinuation, faster arrival at a pharmacologically adequate regimen, or reduced exposure to high-risk drug-gene combinations. Those outcomes are clinically important, yet they do not always map neatly onto conventional antidepressant efficacy endpoints. Future trials should therefore prespecify both pharmacologic process outcomes and patient-centered clinical outcomes. Second, symptomatic endpoints have often been assessed over relatively short time horizons. That is useful for detecting early tolerability or exposure advantages, but insufficient for proving durable remission, relapse prevention or cumulative reductions in ineffective prescribing. PRIME Care exemplified this problem: the prescribing signal was clear, whereas the remission signal was small and not sustained at the principal endpoint [23]. Third, real-world antidepressant pharmacokinetics are shaped by factors that many PGx studies only partially capture, including adherence, smoking, hepatic function, therapeutic drug monitoring, renal function for some compounds, and phenoconversion due to CYP inhibitors or inducers [29]. A genotype-only model can therefore misclassify functional exposure risk unless clinical context is integrated into interpretation. Finally, evidence generalizability remains a major concern. Depression PGx studies still underrepresent several populations most likely to be harmed by overconfident implementation of poorly calibrated tools [34]. A stratified approach that is valid only in selectively studied groups is not yet a sufficiently mature stratified approach for routine global practice. Clinical implications for current practice Current practice should avoid two opposite errors: overselling PGx as a universal antidepressant chooser, and dismissing it because broad combinatorial claims remain unproven. The evidence supports a middle position. PGx is most defensible when clinicians are considering antidepressants with guideline-supported gene-drug recommendations, when patients have shown unusual sensitivity or nonresponse at standard doses, when multiple prior trials have failed, or when polypharmacy raises the likelihood that genotype-informed prescribing will prevent avoidable mismatch [4,6,7,23,29]. A practical way to implement this evidence is to think in terms of pretest probability. The expected value of PGx is highest when the chance that exposure mismatch contributed to prior treatment failure is already elevated. Examples include recurrent intolerance to standard doses, apparent nonresponse despite adequate adherence, switching among multiple CYP2D6- or CYP2C19-relevant drugs, and clinical scenarios in which dose escalation is being considered but toxicity risk is uncertain. In low-complexity first-episode cases without such features, the incremental value of testing is likely smaller even if the assay is analytically valid. In that setting, the preferred interpretive model is transparent and guideline anchored. Results should be translated through CPIC- or DPWG-consistent logic, with explicit attention to phenotype assignment, co-medications, drug specificity and whether the implicated gene has demonstrated clinical actionability for the drug under consideration [35,40]. Vendor color categories may be clinically convenient, but they should not substitute for transparent pharmacologic reasoning. Clinician-facing action statement In patients with prior nonresponse, dose-limiting intolerance, polypharmacy or suspected exposure mismatch, clinicians should test or retrieve CYP2D6 and CYP2C19 results before initiating or switching to antidepressants with guideline-supported recommendations, such as citalopram, escitalopram, sertraline, paroxetine, fluvoxamine, venlafaxine, vortioxetine and tricyclic antidepressants; CYP2B6 should be considered where sertraline-specific interpretation is relevant. Results should be interpreted with CPIC/DPWG-consistent logic. Broad proprietary panel categories should not be used as class-wide antidepressant allocation rules. What should not be done is equally important. PGx should not replace measurement-based care, careful diagnostic assessment, shared decision-making, adherence assessment or attention to comorbidity. Pharmacodynamic candidate genes without current prescribing guidance should not be treated as though they carry the same evidentiary status as CYP2D6 or CYP2C19 [4]. Positive data from one combinatorial panel should not be generalized to every commercial test on the market [28,35–38]. For clinicians treating depression today, the pragmatic conclusion is that PGx has crossed the threshold from speculative biology to selective clinical tool, but not the higher threshold required for universal first-line stratification. Research agenda The next phase of antidepressant PGx research should be designed explicitly around stratification rather than around all-comer average effects. As summarized in the right panel of Figure 1, trials should enrich for patients with high prior probability of PGx relevance: prior intolerance, repeated nonresponse despite apparent adherence, polypharmacy with likely phenoconversion, or planned exposure to antidepressants with existing CPIC/DPWG recommendations. Subgroup hypotheses should be prospectively specified and statistically powered; trial reports should disclose panel content, genotype-to-phenotype translation, recommendation logic and whether interpretations are aligned with transparent standards. Outcome sets should also be broadened beyond short-term symptom change. Stratified PGx should be judged by remission and response, but also by time to adequate exposure, adverse-effect burden, discontinuation, need for switching, therapeutic drug monitoring concordance and durability of remission. Trial populations should be deliberately multi-ancestry, and commercial platforms should be compared directly with guideline-based modular PGx decision support to determine whether proprietary combinatorial logic adds value beyond validated drug-gene pairs. Conclusion The evidence base in antidepressant pharmacogenomics is no longer at the stage of simple plausibility, but neither has it reached the stage of universally validated stratified psychiatry. The strongest support is for pharmacokinetic stratification: identifying patients whose CYP2D6, CYP2C19 or selected CYP2B6 status makes standard exposure unlikely or unsafe for particular antidepressants. That support is reinforced by contemporary guidelines, by the reproducible effect of testing on prescribing-process measures and by subgroup signals in more treatment-exposed or difficult-to-treat patients. What remains unproven is the broader claim that current multigene commercial tools can reliably sort most patients with MDD into superior treatment pathways before treatment begins. Thus, the answer to the review question is deliberately asymmetric: yes for selective, pharmacokinetically grounded PGx use in high-yield contexts; not yet for broad panel-based treatment assignment across unselected depression populations. Future perspective Over the next 5–10 years, antidepressant PGx is unlikely to succeed as a stand-alone genomic oracle. Its more plausible future is as one layer in a transparent prescribing framework that combines actionable CYP results with co-medication review, clinical phenotype, adverse-effect history and, where available, therapeutic drug monitoring. The roadmap in Figure 1 therefore points toward modular, auditable decision support rather than opaque class-wide panel assignment. The most informative next-generation trials should enrich for patients with high pretest probability of pharmacokinetic mismatch and should prospectively test whether PGx changes clinically durable outcomes. More diverse cohorts, standardized PGx reporting and explicit phenoconversion assessment are required if stratified antidepressant PGx is to be equitable, reproducible and clinically transportable. A mature stratified model may eventually incorporate pharmacodynamic, transcriptomic, inflammatory, digital or therapeutic-drug-monitoring data. Until those layers are independently replicated and clinically interpretable, they should complement – not replace – the narrower evidence base supporting transparent CYP-guided prescribing. Funding Statement This paper was not funded. Article highlights Current antidepressant PGx evidence is strongest for pharmacokinetic genes, especially CYP2D6 and CYP2C19, and, for selected serotonin reuptake inhibitors, CYP2B6. Current antidepressant PGx evidence is strongest for pharmacokinetic genes, especially CYP2D6 and CYP2C19, and, for selected serotonin reuptake inhibitors, CYP2B6. PGx-guided care differs from true stratified treatment: the latter requires a prospectively identifiable subgroup that benefits from a different treatment pathway. PGx-guided care differs from true stratified treatment: the latter requires a prospectively identifiable subgroup that benefits from a different treatment pathway. CPIC and DPWG support genotype-informed prescribing for selected antidepressants, whereas SLC6A4 and HTR2A do not currently support clinical prescribing recommendations. CPIC and DPWG support genotype-informed prescribing for selected antidepressants, whereas SLC6A4 and HTR2A do not currently support clinical prescribing recommendations. Prospective trials show PGx testing more consistently reduces drug-gene mismatch than improves average symptom outcomes across unselected MDD populations. Prospective trials show PGx testing more consistently reduces drug-gene mismatch than improves average symptom outcomes across unselected MDD populations. Modest all-comer effects and larger signals in severe or treatment-exposed patients are compatible with subgroup-dependent rather than universal utility. Modest all-comer effects and larger signals in severe or treatment-exposed patients are compatible with subgroup-dependent rather than universal utility. Evidence for stratified PGx is strongest when the subgroup is defined by actionable pharmacokinetic liability and probable antidepressant exposure mismatch. Evidence for stratified PGx is strongest when the subgroup is defined by actionable pharmacokinetic liability and probable antidepressant exposure mismatch. Commercial combinatorial PGx panels are heterogeneous and not interchangeable; evidence supporting one platform should not be generalized across all marketed tests. Commercial combinatorial PGx panels are heterogeneous and not interchangeable; evidence supporting one platform should not be generalized across all marketed tests. Clinically, PGx should be used selectively through transparent CYP2D6/CYP2C19/CYP2B6 interpretation for affected antidepressants, not broad panel class assignment. Clinically, PGx should be used selectively through transparent CYP2D6/CYP2C19/CYP2B6 interpretation for affected antidepressants, not broad panel class assignment. Future progress requires prospective enriched trials, transparent algorithms, standardized reporting, phenotype-aware interpretation and diverse cohorts. Future progress requires prospective enriched trials, transparent algorithms, standardized reporting, phenotype-aware interpretation and diverse cohorts. Author contributions Bernhard T Baune: Conceptualization, methodology, investigation, writing – original draft, visualization, writing – review and editing. All authors approved the final version of the manuscript. Disclosure statement Bernhard T. Baune: Consultant – National Health and Medical Research Council (Australia). Honoraria – Angelini, AstraZeneca, Biogen, BMS, Boehringer Ingelheim, Johnson & Johnson, LivaNova, Lundbeck, Medscape, Novartis, Otsuka, Pfizer, Roche, Servier, Sumitomo Pharma, Sunovion, Takeda, Teva, Viatris and Wyeth. Advisory boards – Biogen, Boehringer Ingelheim, GH Research, Janssen-Cilag, LivaNova, Lundbeck, Medscape, Novartis, Otsuka, Teva and Viatris. Research grants from private industries or nonprofit funds – AstraZeneca, BMBF (Germany), BMG (Germany), DFG (Germany), ERA PerMed, Fay Fuller Foundation, Horizon Europe (European Union), James & Diana Ramsay Foundation (Adelaide), Johnson & Johnson, Lundbeck, La Marató de TV3, National Health and Medical Research Council (Australia), Sanofi-Synthélabo and Wellcome Trust (UK). The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed. References Papers of special note have been highlighted as either of interest (•) or of considerable interest (••) to readers.