Predictors of Treatment Resistance Across Disorders: Biomarkers, Clinical Features, and Prognostic Models

Treatment resistance is one of the most consequential challenges in clinical psychiatry. Despite the availability of multiple evidence-based interventions for major psychiatric disorders, a substantial proportion of patients fail to achieve adequate response or sustained remission with standard treatments. The clinical and economic burden of treatment resistance is immense — it accounts for disproportionate healthcare utilization, higher rates of disability, increased suicide risk, and markedly diminished quality of life.

The concept of treatment resistance lacks a single, universal definition across disorders, which itself introduces complexity. In major depressive disorder (MDD), treatment-resistant depression (TRD) is most commonly defined as failure to respond to two or more adequate antidepressant trials of different pharmacological classes, each at adequate dose and duration (typically ≥ 6 weeks at therapeutic dose). In schizophrenia, treatment resistance is defined by the failure of two adequate trials of antipsychotic medication, at least one of which should be a non-clozapine second-generation antipsychotic. In obsessive-compulsive disorder (OCD), treatment resistance typically refers to inadequate response to at least two adequate trials of serotonin reuptake inhibitors (SRIs) at maximum tolerated doses for ≥ 12 weeks, plus failure to respond to an adequate course of exposure and response prevention (ERP) therapy.

The prevalence of treatment resistance varies across disorders but is consistently substantial. Approximately 30–35% of patients with MDD develop TRD. In schizophrenia, roughly 20–30% of patients meet criteria for treatment-resistant schizophrenia (TRS). In OCD, approximately 40–60% of patients show incomplete response to first-line SRI treatment, and about 20–30% remain significantly symptomatic despite optimized multimodal intervention. For bipolar disorder, treatment resistance patterns are less formally codified, but approximately 30–40% of patients experience persistent mood instability despite guideline-concordant pharmacotherapy.

Understanding the predictors of treatment resistance — spanning genetic, neurobiological, clinical, and psychosocial domains — is critical for early identification of high-risk patients, personalized treatment planning, and the development of novel interventions targeting the mechanisms underlying refractory illness. This article provides a transdiagnostic and disorder-specific review of the evidence base for predictors of treatment resistance, including biomarkers, clinical features, and emerging prognostic models.

The Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial remains the most influential study informing our understanding of treatment resistance in MDD. This NIMH-funded, multi-step, sequentially randomized trial enrolled 4,041 outpatients with nonpsychotic MDD across 41 clinical sites. The results were sobering: only approximately 37% of patients achieved remission (defined as a Hamilton Rating Scale for Depression [HRSD-17] score ≤ 7) with the first adequate antidepressant trial (citalopram). With each subsequent treatment step, remission rates declined progressively — approximately 31% in Step 2, 14% in Step 3, and 13% in Step 4. Cumulatively, after up to four sequential treatment attempts, the overall remission rate was approximately 67%, meaning that roughly one-third of patients with MDD failed to remit despite multiple pharmacological strategies.

STAR*D also revealed critical prognostic factors. Clinical features predicting poorer outcomes across treatment steps included: longer current episode duration, greater baseline severity, comorbid anxiety disorders (present in approximately 50–60% of the sample), comorbid substance use disorders, comorbid general medical conditions, earlier age of onset, and the presence of suicidal ideation. Each successive treatment failure was associated with higher relapse rates among those who did eventually remit — the relapse rate was approximately 40% after Step 1 remission but rose to approximately 70% after Step 3 or 4 remission.

These findings underscored a fundamental clinical reality: early treatment failure is itself the strongest predictor of subsequent treatment resistance. The number of prior failed trials is the single most robust clinical predictor of non-response to the next intervention, a finding replicated across multiple datasets including the European Group for the Study of Resistant Depression (GSRD) studies and the CRESCEND cohort.

Neurobiological predictors of TRD have been increasingly characterized. Structural neuroimaging studies consistently implicate reduced hippocampal volume as a predictor of poor antidepressant response, with a meta-analysis by Schmaal et al. (2016) in the ENIGMA Major Depressive Disorder Working Group demonstrating hippocampal volume reductions in recurrent MDD. Functional connectivity studies identify anterior cingulate cortex (ACC) hyperactivity — specifically, elevated pretreatment activity in the subgenual ACC (Brodmann area 25) — as a potential predictor of response to certain treatments (e.g., deep brain stimulation), while rostral ACC hyperactivity has been associated with better response to antidepressants in several studies. The default mode network (DMN), which shows hyperconnectivity in depression, demonstrates persistent dysconnectivity patterns in TRD patients compared to treatment-responsive patients.

At the molecular level, hypothalamic-pituitary-adrenal (HPA) axis dysregulation — evidenced by elevated cortisol levels and non-suppression on the dexamethasone suppression test — has been associated with poorer antidepressant outcomes. Elevated inflammatory markers, particularly C-reactive protein (CRP), interleukin-6 (IL-6), and tumor necrosis factor-alpha (TNF-α), predict poorer response to monoaminergic antidepressants. Notably, a secondary analysis of the GUIDED trial and data from Jha et al. (2017) suggest that patients with CRP levels > 1 mg/L may respond preferentially to agents with anti-inflammatory or glutamatergic mechanisms. The glutamatergic system has emerged as particularly relevant to TRD, given the rapid antidepressant effects of ketamine and esketamine — the latter approved by the FDA in 2019 for TRD — with response rates of approximately 50–70% in acute trials, though durability remains a clinical concern.

Treatment-resistant schizophrenia (TRS) affects approximately 20–30% of patients with schizophrenia and is formally defined by persistent positive symptoms despite adequate trials of at least two non-clozapine antipsychotic medications at appropriate doses for adequate durations (typically ≥ 6 weeks each). The Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study, a landmark NIMH-funded trial comparing first- and second-generation antipsychotics in 1,493 patients, demonstrated high discontinuation rates across all antipsychotics (approximately 74% of patients discontinued their assigned medication within 18 months), underscoring the broader challenges of treatment optimization in schizophrenia.

The neurobiological basis of TRS appears to involve distinct pathophysiology compared to treatment-responsive schizophrenia. The prevailing model distinguishes between dopaminergic (type 1) and non-dopaminergic (type 2) subtypes. PET imaging studies by Howes et al. (2012) and the Psychiatric Imaging Group at the MRC London Institute of Medical Sciences demonstrated that treatment-responsive schizophrenia patients show elevated presynaptic dopamine synthesis capacity in the striatum, consistent with the classical dopamine hypothesis. In contrast, patients with TRS show normal or near-normal striatal dopamine synthesis capacity, suggesting that their psychotic symptoms may be driven by non-dopaminergic mechanisms — potentially involving glutamatergic dysfunction, GABAergic deficits, or cortical rather than subcortical pathology.

Magnetic resonance spectroscopy (MRS) studies support this distinction, with TRS patients showing elevated glutamate and glutamine (Glx) levels in the anterior cingulate cortex compared to treatment-responsive patients. This glutamatergic excess may reflect NMDA receptor hypofunction, which is consistent with the glutamate hypothesis of schizophrenia and provides a neurochemical rationale for why standard D2-blocking antipsychotics are ineffective in this subgroup.

Clozapine remains the only medication with robust evidence for TRS and is considered the gold standard treatment. Its unique efficacy is likely attributable to its broad receptor-binding profile, including relatively weak D2 antagonism, potent 5-HT2A antagonism, muscarinic M1/M4 agonism, and effects on glutamatergic transmission. Meta-analytic evidence indicates clozapine's superiority over other antipsychotics for TRS, with a number needed to treat (NNT) of approximately 6–8 for meaningful clinical response versus other second-generation antipsychotics. However, approximately 40–60% of TRS patients fail to respond adequately to clozapine, a condition sometimes termed ultra-treatment-resistant schizophrenia (UTRS).

Clinical predictors of TRS include early onset of illness (particularly onset before age 20), male sex, longer duration of untreated psychosis (DUP), prominent negative symptoms, poor premorbid functioning, substance use comorbidity (particularly cannabis use), and family history of schizophrenia. The DUP is a particularly robust predictor: meta-analytic data from Perkins et al. (2005) and subsequent reviews demonstrate that longer DUP is associated with poorer treatment outcomes, greater residual symptoms, and increased risk of developing treatment resistance. This finding has been a cornerstone argument for early intervention services in psychosis.

Genetic predictors remain an active area of research. Common genetic variants in the DRD2 gene and the CYP enzyme family (particularly CYP1A2, CYP2D6, and CYP3A4) influence antipsychotic pharmacokinetics and may contribute to apparent treatment resistance that is actually pharmacokinetic in nature. Genome-wide association studies (GWAS) have not yet identified robust single-gene predictors of treatment resistance, though polygenic risk scores (PRS) are being investigated as potential composite biomarkers.

Obsessive-compulsive disorder (OCD) presents particularly high rates of partial or non-response to first-line treatments. Meta-analytic data indicate that approximately 40–60% of OCD patients fail to achieve ≥ 35% reduction in Yale-Brown Obsessive Compulsive Scale (Y-BOCS) scores — the typical threshold for treatment response — with an adequate SRI trial. Even among responders, full remission is achieved by only approximately 25% of patients. The combination of SRI pharmacotherapy and exposure and response prevention (ERP) therapy yields the best outcomes, but approximately 20–30% of patients remain significantly symptomatic despite optimized multimodal treatment.

Predictors of treatment resistance in OCD are well-characterized. Clinical predictors of poor outcome include: hoarding symptoms (which show the poorest SRI response rates, with response rates approximately 18–31% compared to 50–60% for contamination/washing subtypes), early onset (before age 10), longer illness duration, greater baseline severity, comorbid tic disorders, poor insight or overvalued ideation, symmetry/ordering obsessions, and comorbid personality disorders (particularly schizotypal and cluster B disorders). Comorbid MDD is present in approximately 60–70% of OCD patients and is associated with greater functional impairment but does not consistently predict SRI non-response; however, severe depression may interfere with engagement in ERP.

Neurobiologically, treatment resistance in OCD is associated with dysfunction in the cortico-striato-thalamo-cortical (CSTC) circuits, particularly hyperactivity in the orbitofrontal cortex (OFC), caudate nucleus, and anterior cingulate cortex. PET and fMRI studies demonstrate that treatment responders show normalization of OFC and caudate hyperactivity with successful SRI treatment or CBT, while non-responders show persistent hyperactivity. Glutamatergic dysregulation is also implicated in treatment-resistant OCD, with MRS studies showing elevated glutamate levels in the caudate in refractory patients. This has led to trials of glutamate-modulating agents, including memantine, riluzole, and N-acetylcysteine, as augmentation strategies, with modest but variable evidence.

For treatment-resistant OCD, augmentation strategies include low-dose antipsychotic augmentation (particularly aripiprazole or risperidone), which shows a response rate of approximately 30–35% and an NNT of approximately 4–5 based on meta-analytic data. For truly refractory cases, neurosurgical interventions including deep brain stimulation (DBS) targeting the ventral capsule/ventral striatum and anterior capsulotomy have demonstrated response rates of approximately 40–60% in carefully selected patients, though the evidence base consists primarily of small case series and open-label trials.

In posttraumatic stress disorder (PTSD), approximately 30–50% of patients fail to respond adequately to evidence-based psychotherapies (prolonged exposure, cognitive processing therapy) or pharmacotherapy (SSRIs, SNRIs). Predictors of treatment resistance include: combat-related trauma (which shows lower response rates than other trauma types), comorbid traumatic brain injury, dissociative subtype PTSD, comorbid substance use disorders, ongoing exposure to trauma or threat, and childhood trauma history. The VA Cooperative Studies Program trials and the RESPECT-PTSD trial demonstrated that SSRI response rates in veterans with PTSD are notably lower (approximately 25–40%) than in civilian samples, highlighting the importance of trauma type as a prognostic factor.

For generalized anxiety disorder (GAD), treatment resistance is less formally codified, but approximately 30–40% of patients remain symptomatic after adequate SSRI/SNRI treatment. Comorbid depression, personality pathology, chronic medical illness, and psychosocial stressors predict poorer outcomes. Neurobiologically, amygdala hyperreactivity and impaired prefrontal-amygdala connectivity have been associated with poor anxiolytic treatment response.

One of the most active frontiers in psychiatric research is the identification of transdiagnostic biomarkers that predict treatment resistance across diagnostic boundaries. Several candidate biomarker domains have shown promise, though none has yet achieved the sensitivity and specificity required for routine clinical implementation.

Inflammatory Biomarkers

Chronic low-grade inflammation is one of the most consistently replicated biological correlates of treatment resistance across multiple disorders. Elevated CRP (> 3 mg/L), IL-6, and TNF-α are associated with poorer antidepressant response in MDD, poorer antipsychotic response in schizophrenia, and greater symptom severity in PTSD and bipolar disorder. The Insight study by Chamberlain et al. (2019) and data from the Biomarkers of Depression (BioDep) consortium demonstrated that approximately 25–30% of patients with TRD show elevated inflammatory markers, and this subgroup may benefit differentially from anti-inflammatory augmentation strategies or glutamatergic agents. However, meta-analyses of anti-inflammatory augmentation (e.g., celecoxib, minocycline) in depression show modest effect sizes (Cohen's d ≈ 0.30–0.55) and significant heterogeneity.

Neuroimaging Biomarkers

Functional neuroimaging has yielded several promising predictive signals. The landmark EMBARC (Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care) trial, a multi-site RCT comparing sertraline to placebo, investigated neuroimaging predictors of treatment response. Key findings included that rostral ACC activity on fMRI predicted differential response to sertraline versus placebo, and that reward-related neural activity in the ventral striatum predicted response trajectory. The iSPOT-D (International Study to Predict Optimized Treatment in Depression) study similarly demonstrated that EEG-based biomarkers, specifically frontal theta cordance, predicted differential response to escitalopram, sertraline, and venlafaxine with modest accuracy.

Structural MRI findings have been less specific but more consistent. Reduced hippocampal volume, reduced anterior cingulate cortex volume, and white matter hyperintensities are transdiagnostic markers associated with poorer treatment outcomes across depression, schizophrenia, and anxiety disorders. White matter hyperintensities in late-life depression are particularly predictive of treatment resistance, forming the basis of the vascular depression hypothesis proposed by Alexopoulos and Krishnan.

Genetic and Pharmacogenomic Biomarkers

Pharmacogenomic testing has become increasingly commercially available, particularly for CYP450 enzyme genotyping (CYP2D6, CYP2C19, CYP2B6, CYP3A4) and serotonin transporter (SLC6A4) polymorphisms. The GUIDED trial (Genomics Used to Improve DEpression Decisions) demonstrated that pharmacogenomics-guided treatment led to modestly higher remission rates compared to treatment as usual (15.3% vs. 10.1% at Week 8, though the primary endpoint did not reach significance). The IMPACT study similarly found marginal benefits. The clinical utility of combinatorial pharmacogenomic testing remains debated, with the American Psychiatric Association and the Clinical Pharmacogenetics Implementation Consortium (CPIC) recommending CYP2D6 and CYP2C19 genotyping primarily for dosing guidance rather than drug selection.

Beyond pharmacokinetic genes, genome-wide approaches have identified polygenic signals associated with treatment resistance. A GWAS by the Psychiatric Genomics Consortium (PGC) identified common variant associations with antidepressant response, but individual SNP effect sizes are small (odds ratios typically 1.02–1.10), and polygenic risk scores currently explain only approximately 1–3% of variance in treatment outcomes — insufficient for individual-level prediction.

HPA Axis and Neuroendocrine Markers

Cortisol dysregulation, measured via diurnal cortisol slope, cortisol awakening response, or dexamethasone-CRH test non-suppression, has been associated with poorer treatment outcomes in depression and PTSD. However, these measures show substantial within-individual variability and have not achieved clinical biomarker status. Brain-derived neurotrophic factor (BDNF) levels in serum have been investigated extensively; while lower BDNF levels are associated with depression and some evidence suggests that treatment response involves BDNF normalization, the sensitivity and specificity of BDNF as a predictive biomarker remain inadequate for clinical use.

While biomarker research is advancing, clinical features remain the most immediately accessible and practically useful predictors of treatment resistance. Several clinical variables show robust associations with poor treatment outcomes across multiple disorders.

Comorbidity

Psychiatric comorbidity is the norm rather than the exception, and its presence consistently predicts treatment resistance. In MDD, comorbid anxiety disorders are present in approximately 50–60% of patients and are associated with poorer antidepressant response, longer time to remission, and higher relapse rates — a finding consistently replicated in STAR*D and the GSRD dataset. Comorbid personality disorders, particularly borderline personality disorder, predict poorer outcomes across depression, anxiety, and PTSD treatment trials, with some studies suggesting the relationship is mediated by reduced treatment adherence and interpersonal difficulties in the therapeutic alliance. Substance use disorders co-occur with schizophrenia in approximately 40–50% of cases, with MDD in approximately 20–30%, and with PTSD in approximately 30–50%, and consistently predict treatment resistance, partly through pharmacokinetic interference and partly through psychosocial destabilization.

Episode Characteristics

Across mood and anxiety disorders, greater baseline symptom severity and longer episode duration predict poorer treatment response. In depression, episodes lasting > 2 years (chronic depression) show lower response rates to pharmacotherapy alone compared to acute episodes, though the CBASP (Cognitive Behavioral Analysis System of Psychotherapy) developed by McCullough specifically targets chronic depression and showed response rates of approximately 48% (combined with nefazodone: 73%) in the landmark Keller et al. (2000) trial. The presence of psychotic features in depression predicts poor response to antidepressant monotherapy (response rates approximately 20–30% compared to 50–60% for non-psychotic MDD) but good response to combination antidepressant-antipsychotic therapy or ECT.

Childhood Adversity and Trauma History

A history of childhood trauma and adversity — particularly childhood sexual abuse, physical abuse, emotional neglect, and household dysfunction — is one of the most consistently replicated predictors of treatment resistance across disorders. Patients with MDD and a history of childhood maltreatment show antidepressant response rates approximately 15–25% lower than those without such history. This association may be mediated by persistent neurobiological changes including HPA axis hyperactivation, epigenetic modifications (particularly methylation of the glucocorticoid receptor gene NR3C1 and the FKBP5 gene), reduced hippocampal volume, and amplified inflammatory signaling. The ACE (Adverse Childhood Experiences) study data demonstrate a dose-response relationship between childhood adversity and adult mental health outcomes, with higher ACE scores predicting greater treatment resistance.

Age of Onset and Duration of Illness

Earlier age of onset is associated with treatment resistance in depression, schizophrenia, OCD, and bipolar disorder. In schizophrenia, adolescent-onset illness is associated with approximately twice the risk of developing TRS compared to adult-onset illness. In OCD, childhood onset (before age 10) is associated with higher rates of tic-related OCD, greater familial loading, and poorer SRI response. Longer overall illness duration, independently of episode characteristics, also predicts treatment resistance — likely reflecting progressive neurobiological changes including synaptic loss, white matter changes, and allostatic load effects of chronic stress system activation.

The integration of multiple predictors into prognostic models represents the next step beyond individual biomarker or clinical predictor research. Several frameworks have been proposed.

Clinical Staging Models

Adapted from oncology, clinical staging models in psychiatry propose that mental disorders progress through identifiable stages — from at-risk states to first episodes to recurrent illness to treatment resistance. McGorry and colleagues developed a staging model for psychosis and mood disorders that has been influential in conceptualizing treatment resistance as a late-stage phenomenon characterized by accumulated neurobiological damage, comorbidity, and psychosocial impairment. The Maudsley Staging Method (MSM) for TRD, developed by Fekadu et al. (2009), integrates treatment failure history, symptom severity, and episode duration into a composite score that predicts subsequent treatment outcomes. Higher MSM scores predict poorer response to ECT and pharmacological interventions, with one study showing that patients in the highest MSM staging category had a remission rate of only approximately 9% with standard pharmacotherapy compared to approximately 50% in the lowest staging category.

Machine Learning Approaches

Computational approaches have increasingly been applied to treatment outcome prediction. Machine learning models — including random forests, gradient-boosted models, support vector machines, and deep learning architectures — have been trained on clinical, neuroimaging, genetic, and electronic health record data to predict treatment resistance. The STAR*D dataset has been extensively used for retrospective model development, with several studies achieving area under the receiver operating characteristic curve (AUC) values of approximately 0.65–0.75 for predicting antidepressant non-response using clinical variables alone, and modestly higher values (AUC ≈ 0.70–0.80) when neuroimaging or genetic data are incorporated.

Chekroud et al. (2016) developed a machine learning model using data from the STAR*D trial and a validation dataset from the CO-MED (Combining Medications to Enhance Depression Outcomes) trial that predicted antidepressant non-response with an AUC of approximately 0.70 using 25 clinical variables. Important features included baseline functional impairment, insomnia severity, comorbid anxiety, cognitive symptoms, and loss of interest. While promising, these models have not yet been validated prospectively in routine clinical settings at the scale needed for clinical implementation.

The PRISM (Predicting Response to Depression Treatment) initiative and similar large-scale collaborative efforts aim to develop clinically deployable prediction tools by pooling data across multiple trials and clinical sites. The key challenge remains the generalizability gap — models trained on research cohorts may perform differently in real-world clinical populations characterized by greater comorbidity, treatment heterogeneity, and demographic diversity.

Multimodal Prediction Frameworks

The most promising prognostic models combine multiple data types. A framework proposed by Trivedi and colleagues integrates clinical features, pharmacogenomics, inflammatory biomarkers, and neuroimaging into a composite prediction model. The Texas Medication Algorithm Project (TMAP) and its successors have implemented measurement-based care protocols that, while not predictive models per se, systematically track treatment response and guide sequential treatment decisions in a manner that implicitly accounts for treatment resistance risk factors. Evidence from TMAP implementation studies suggests that measurement-based care improves remission rates by approximately 10–15 percentage points compared to treatment as usual.

Understanding treatment resistance requires moving beyond single-neurotransmitter models to circuit-level and systems-level neuroscience frameworks.

Monoaminergic System Limitations

The monoamine hypothesis — that depression results from deficient serotonergic, noradrenergic, or dopaminergic neurotransmission — has been the dominant pharmacological framework for decades. However, its limitations are evident in treatment resistance. Standard monoaminergic antidepressants (SSRIs, SNRIs, TCAs, MAOIs) achieve remission in approximately 30–40% of patients with a first adequate trial. The fact that monoamine depletion does not reliably produce depression in healthy individuals, and that synaptic monoamine levels increase within hours of antidepressant administration while clinical effects take weeks, suggests that monoaminergic modulation is a necessary but insufficient mechanism. Treatment resistance may arise from downstream signaling abnormalities — including deficits in BDNF-TrkB signaling, impaired neuroplasticity, glucocorticoid receptor resistance, and mitochondrial dysfunction — that persist despite monoaminergic augmentation.

Glutamatergic System

The glutamate system has emerged as a central focus of treatment resistance research. The rapid antidepressant effects of ketamine, an NMDA receptor antagonist, in TRD — with approximately 50–70% response rates within 24 hours in controlled trials, compared to approximately 10–20% for placebo — represent a paradigm shift. Ketamine's mechanism involves blockade of NMDA receptors on GABAergic interneurons, leading to disinhibition of glutamatergic pyramidal neurons, increased AMPA receptor activation, and rapid enhancement of BDNF release and synaptic plasticity via the mTOR (mechanistic target of rapamycin) pathway. Esketamine (the S-enantiomer), administered intranasally, was FDA-approved for TRD in 2019 based on the TRANSFORM and SUSTAIN trial programs, showing remission rates of approximately 36% versus 31% for active comparator (the clinical significance of this margin has been debated). Glutamatergic dysfunction in TRS — evidenced by elevated anterior cingulate glutamate in MRS studies — provides a parallel mechanistic explanation for why dopamine-blocking antipsychotics fail in this subgroup.

GABAergic Deficits

GABAergic interneuron dysfunction has been implicated in treatment resistance across depression, schizophrenia, and OCD. Postmortem and neuroimaging studies demonstrate reduced GABA levels in the cortex of patients with TRD. Parvalbumin-positive (PV+) fast-spiking interneurons are particularly affected in schizophrenia, and their loss or dysfunction disrupts gamma oscillation generation — a mechanism thought to underlie cognitive deficits and potentially treatment resistance. In OCD, reduced GABA levels in the medial prefrontal cortex have been reported in treatment-resistant cases.

Neuroplasticity and Neurotrophic Mechanisms

Impaired neuroplasticity is increasingly recognized as a transdiagnostic mechanism underlying treatment resistance. Chronic stress, elevated cortisol, and inflammatory signaling converge to impair long-term potentiation (LTP), reduce dendritic spine density, and suppress adult hippocampal neurogenesis. Effective treatments — including antidepressants, ketamine, ECT, and psychotherapy — share the common downstream effect of enhancing neuroplasticity, often via BDNF-TrkB signaling. Treatment resistance may reflect a failure of neuroplastic restoration, potentially due to epigenetic modifications (e.g., BDNF promoter methylation), persistent inflammatory signaling, or structural damage that exceeds the brain's regenerative capacity.

Psychiatric comorbidity is arguably the most practically important predictor of treatment resistance, yet it is frequently underappreciated in clinical practice and inadequately addressed in treatment guidelines that tend to focus on single disorders.

Depression and anxiety comorbidity: Approximately 50–60% of patients with MDD have comorbid anxiety disorders. In the STAR*D trial, comorbid anxiety was the strongest clinical predictor of non-remission. Anxious depression (MDD with prominent anxiety symptoms or comorbid anxiety disorders) shows antidepressant response rates approximately 10–20% lower than non-anxious depression, and patients with anxious depression show preferential response to SSRIs/SNRIs over bupropion, and may benefit from augmentation with buspirone or atypical antipsychotics.

Substance use disorder (SUD) comorbidity: SUD co-occurs with virtually all major psychiatric disorders and is a potent driver of treatment resistance. In schizophrenia, comorbid SUD (present in approximately 40–50% of patients) is associated with medication nonadherence, higher relapse rates, increased hospitalization, and treatment resistance. In PTSD, comorbid SUD (approximately 30–50%) complicates both psychotherapeutic and pharmacological treatment. Alcohol use directly interferes with antidepressant pharmacodynamics, while stimulant and cannabis use can exacerbate psychotic symptoms.

Personality disorder comorbidity: Comorbid personality disorders, present in approximately 40–60% of patients with treatment-resistant depression and 25–35% of patients in general psychiatric outpatient samples, predict poorer outcomes across treatment modalities. Borderline personality disorder (BPD) comorbidity in MDD is associated with lower antidepressant response rates, higher dropout rates from psychotherapy, and greater chronicity. However, specialized psychotherapies for BPD (dialectical behavior therapy, mentalization-based treatment) can improve outcomes when personality pathology is directly addressed alongside the comorbid Axis I disorder.

Chronic medical comorbidity: Approximately 25–30% of patients with chronic medical conditions (cardiovascular disease, diabetes, autoimmune disorders, chronic pain) have comorbid depression, and this subgroup shows higher rates of treatment resistance. The bidirectional relationship between inflammation, chronic disease, and psychiatric treatment resistance suggests that untreated medical comorbidity may be a modifiable driver of apparent treatment resistance. Systematic assessment and management of medical comorbidity is therefore an essential component of treatment resistance evaluation.

Before concluding that a patient is treatment-resistant, clinicians must rigorously evaluate for pseudo-resistance — apparent treatment failure attributable to modifiable factors rather than true biological non-responsiveness. Estimates suggest that 25–50% of cases labeled as treatment-resistant may involve pseudo-resistance.

Diagnostic Misclassification

Diagnostic error is a common and often underappreciated contributor to apparent treatment resistance. The most consequential misdiagnosis in clinical practice is the failure to identify bipolar disorder in patients presenting with depression. Approximately 10–20% of patients initially diagnosed with unipolar MDD are subsequently rediagnosed with bipolar disorder, and antidepressant monotherapy in bipolar depression is associated with poor response, mixed features, and cycling. Other frequently missed diagnoses include ADHD (present in approximately 10–15% of adults with TRD), hypothyroidism, obstructive sleep apnea, and chronic traumatic encephalopathy. A careful longitudinal history, collateral information, and systematic screening for hypomanic/manic symptoms (e.g., Mood Disorder Questionnaire) are essential in the treatment resistance evaluation.

Inadequate Treatment Trials

Many apparent treatment failures reflect inadequate dose, duration, or adherence. Studies of TRD referrals to specialty clinics consistently find that 30–50% of patients have not received truly adequate treatment trials. Common problems include sub-therapeutic dosing (e.g., fluoxetine 20 mg maintained without titration despite non-response), insufficient trial duration (less than 6–8 weeks at therapeutic dose), and poor adherence (medication adherence rates for psychiatric medications are approximately 50–65% across disorders). Therapeutic drug monitoring (TDM), particularly for clozapine, tricyclic antidepressants, and lithium, can identify pharmacokinetic non-response — patients who are rapid metabolizers or non-adherent may show subtherapeutic serum levels despite adequate prescribed doses.

Perpetuating Psychosocial Factors

Ongoing psychosocial stressors — including intimate partner violence, financial insecurity, housing instability, social isolation, and chronic interpersonal conflict — can maintain psychiatric symptoms despite otherwise adequate biological treatment. Similarly, untreated comorbid substance use, personality pathology, or cognitive impairment may undermine treatment effectiveness. A comprehensive biopsychosocial assessment is essential before labeling a patient as treatment-resistant.

The recognition that treatment resistance involves distinct pathophysiology has driven the development of novel interventions targeting non-monoaminergic mechanisms.

Neuromodulation

Electroconvulsive therapy (ECT) remains the most effective acute treatment for TRD, with remission rates of approximately 50–65%, significantly higher than any pharmacological intervention. However, cognitive side effects — particularly retrograde amnesia — and the need for repeated anesthesia limit its acceptability. Repetitive transcranial magnetic stimulation (rTMS) targeting the left dorsolateral prefrontal cortex (DLPFC) shows response rates of approximately 30–35% and remission rates of approximately 20% in TRD, with an NNT of approximately 6–8 versus sham. The Stanford Neuromodulation Therapy (SNT) protocol — an accelerated, high-dose theta-burst stimulation protocol with fMRI-guided targeting — demonstrated a remission rate of approximately 79% in a small initial RCT by Cole et al. (2022), though replication in larger trials is needed. Vagus nerve stimulation (VNS) has FDA approval for chronic TRD but shows modest acute effects (response rates approximately 15–20% above sham at 3 months), with potential for cumulative benefit over 1–2 years.

Psychedelic-Assisted Therapies

Psilocybin-assisted therapy has shown remarkable preliminary results in TRD. A randomized controlled trial by Carhart-Harris et al. (2021) comparing psilocybin to escitalopram demonstrated comparable efficacy, with greater rapid response in the psilocybin group, though the trial was not powered for superiority testing. Phase 2 data from COMPASS Pathways' trial in TRD showed a significant dose-response relationship, with the 25 mg dose producing a response rate of approximately 37% at 3 weeks versus 18% for 1 mg control. MDMA-assisted therapy for treatment-resistant PTSD showed dramatic results in Phase 3 trials (MAPS studies), with approximately 67% of participants no longer meeting PTSD criteria after three sessions (versus approximately 32% for therapy plus placebo), though the FDA's advisory committee raised methodological concerns and the initial application was not approved in 2024.

Targeted Pharmacological Strategies

Beyond ketamine/esketamine, several novel pharmacological targets are in development. Neurosteroid GABAergic modulators, including brexanolone (approved for postpartum depression) and zuranolone (approved for postpartum depression and major depression), act on GABA-A receptors and represent a mechanistically distinct approach. Anti-inflammatory strategies, including targeted anti-cytokine therapies (anti-TNF-α agents, anti-IL-6 agents) are being investigated in biomarker-stratified populations with elevated inflammation. The muscarinic M1/M4 agonist xanomeline-trospium (KarXT) received FDA approval for schizophrenia in 2024, representing the first non-D2-blocking antipsychotic mechanism and a potential option for patients with TRS, though its efficacy specifically in TRS populations has not yet been established.

The accumulating evidence on predictors of treatment resistance has several important clinical implications and points toward critical future directions.

First, clinicians should adopt a proactive rather than reactive approach to treatment resistance. Rather than waiting for multiple treatment failures, early identification of high-risk patients — those with early onset, childhood adversity, comorbid anxiety or personality pathology, elevated inflammatory markers, or family history of treatment resistance — should prompt more aggressive initial treatment strategies, including combination therapy from the outset, early psychotherapy integration, and closer monitoring.

Second, the systematic evaluation of apparent treatment resistance must include rigorous exclusion of pseudo-resistance. A structured treatment resistance evaluation should include: verification of prior treatment adequacy (dose, duration, adherence via TDM when available), diagnostic reassessment (screening for bipolarity, ADHD, medical causes), assessment for comorbidity (substance use, personality disorders, medical conditions), and evaluation of perpetuating psychosocial factors.

Third, measurement-based care — the systematic, quantitative tracking of symptoms and functional outcomes at each clinical visit — is essential for detecting early non-response and guiding timely treatment modifications. The Early Medication Change (EMC) study by Szegedi et al. demonstrated that lack of ≥ 20% improvement in depression severity by Week 2 of antidepressant treatment is a strong negative predictor (negative predictive value ≈ 80–90%) of eventual response, supporting early intervention strategies.

Fourth, the future of treatment resistance prediction lies in multimodal, personalized approaches that integrate clinical features, biomarkers, and computational modeling. The Research Domain Criteria (RDoC) framework, emphasizing dimensional constructs (negative valence, positive valence, cognitive systems, arousal/regulatory systems, social processes) rather than categorical diagnoses, may better capture the transdiagnostic dimensions that underlie treatment resistance. Large-scale collaborative efforts — including the ENIGMA consortium for neuroimaging, the PGC for genetics, and emerging data-sharing initiatives for clinical trial data — are essential for building sufficiently powered prediction models.

Fifth, translational research must address the critical gap between biomarker discovery and clinical implementation. Biomarkers that are reliable, cost-effective, minimally invasive, and prospectively validated in diverse clinical populations are needed. Current candidates — including CRP, cortisol measures, EEG biomarkers, and pharmacogenomic profiles — are most likely to achieve clinical utility as components of multivariate prediction models rather than as standalone tests.

Ultimately, understanding treatment resistance as a biologically and clinically heterogeneous phenomenon — rather than a monolithic entity — is the key paradigm shift required. Treatment resistance in depression is not a single condition; it likely encompasses inflammatory subtype depression, glutamatergic-deficit depression, HPA-axis-driven depression, and other pathophysiologically distinct entities that require different treatment approaches. The same logic applies to TRS, refractory OCD, and treatment-resistant PTSD. Precision psychiatry — matching patients to treatments based on their individual biological and clinical profiles — remains the aspirational goal, with treatment resistance prediction as one of its most impactful potential applications.