Skip to content

Open questions — major depressive disorder

Last curated: 2026-08-30

Stable IDs persist if wording changes. Tier 1 questions could change practice and are designable now; Tier 2 questions require enabling methods, longer horizons, or stronger preliminary evidence.

Tier 1 — practice-changing and designable now

OQ-1 — What is the clinically relevant antidepressant benefit within severity strata?

Large network evidence establishes average superiority to placebo, while FDA and patient-level analyses find benefit concentrated at higher severity (Cipriani 2018, PMID 29477251; Kirsch 2008, PMID 18303940; Fournier 2010, PMID 20051569). An individual-participant meta-analysis using absolute outcomes, functioning, and prespecified severity interactions could resolve the practical magnitude.

OQ-2 — Which STAR*D cumulative remission estimate answers patient decisions?

The original theoretical 67% cumulative remission conflicts with a protocol-faithful reanalysis using different denominators and criteria (Rush 2006, PMID 17074942; Pigott 2023, PMID 37491091). A transparent multi-estimand reanalysis should publish every denominator, attrition state, and sustained-remission outcome.

OQ-3 — What is the best second-step treatment after a first adequate failure?

Switching, combination, lithium or antipsychotic augmentation, psychotherapy, TMS, and rapid-acting treatments lack a definitive pragmatic head-to-head sequence (Nuñez 2022, PMID 34986373; Scott 2023, PMID 35861202). A platform trial could randomize within preference/safety strata.

OQ-4 — Who should receive initial psychotherapy–medication combination?

Combination improves response over either treatment alone (RR about 1.25–1.27), but added burden may not be justified universally (Cuijpers 2020, PMID 31922679). Prespecified effect-modifier analysis should target chronicity, severity, preference, and comorbidity.

OQ-5 — How should long-term antidepressant tapering be individualized?

Withdrawal incidence and duration estimates are heterogeneous. A 2025 RCT-only synthesis found a small average week-1 symptom increment (SMD 0.31, 95% CI 0.23–0.39) but large symptom-specific odds for dizziness and vertigo, while the earlier mixed-design review reported more frequent and prolonged syndromes (Kalfas 2025, PMID 40632531; Davies 2019, PMID 30292574). Trials should compare hyperbolic/proportional taper schedules, stratify by exposure duration and drug half-life, and prospectively adjudicate withdrawal, relapse, and nocebo.

OQ-6 — Which continuation strategy best prevents post-ECT relapse?

Relapse remains common after successful ECT; lithium may help but evidence is limited (Jelovac 2013, PMID 23774532; Lambrichts 2021, PMID 33506961). A factorial comparison of pharmacologic and continuation-ECT strategies is feasible.

OQ-7 — Does personalized accelerated TMS outperform standard targeting and scheduling?

SNT produced a large effect in a small controlled trial, but individualized targeting, acceleration, and high dose were bundled (Cole 2022, PMID 34711062). Multisite factorial or dismantling trials should isolate components and measure one-year durability.

OQ-8 — What is the optimal ketamine/esketamine maintenance schedule and stopping rule?

Randomized withdrawal supports maintenance among esketamine responders, while repeated IV ketamine evidence remains smaller (Daly 2019, PMID 31166571; Phillips 2019, PMID 30922101). Head-to-head interval and taper trials should include cognition, urinary outcomes, misuse, and function.

OQ-9 — Does psilocybin retain benefit after expectancy and functional unblinding are measured?

Trials show symptom benefit, but the escitalopram comparison missed its primary endpoint and psychedelic allocation is conspicuous (Carhart-Harris 2021, PMID 33852780; Goodwin 2022, PMID 36322843). Trials need repeated treatment guesses, independent raters, credible active controls, and long-term adverse-event capture.

OQ-10 — Can an inflammatory marker prospectively select an effective treatment?

Inflammation is associated with depression in subsets, but marker results conflict: drug-naïve synthesis found a large TNF-α difference while CRP and interferon-γ confidence intervals included no effect, and cytokine-response meta-analysis has not validated a selector (Islam 2023, PMID 37625799; Liu 2020, PMID 31427752). Biomarker-enriched randomized trials must prespecify assay, threshold, and a treatment-by-marker interaction—not within-arm change.

OQ-11 — Which interventions reduce suicide attempts or deaths after acute care transitions?

Cohort predictors have limited individual precision (Li 2022, PMID 35101521). Cluster-randomized health-system trials of rapid follow-up, safety planning, means intervention, and outreach can measure attempts and mortality.

OQ-12 — Can measurement-based care scale without widening disparities?

Rating-guided care improved outcomes in a randomized trial (Guo 2015, PMID 26315978), while global minimally adequate treatment remains uncommon (Thornicroft 2017, PMID 27908899). Implementation trials should measure clinician action, patient burden, language access, and equity.

OQ-21 — Which screening pathway improves outcomes rather than only detection?

PHQ-9 ≥10 has sensitivity and specificity near 0.85 against semistructured interviews, and a PHQ-2→PHQ-9 sequence can reduce full-form administration by 57%, but neither establishes that population screening improves remission, function, or mortality (Negeri 2021, PMID 34610915; Levis 2020, PMID 32515813). Cluster trials should randomize complete pathways—screen, diagnostic assessment, treatment capacity, and follow-up—and measure false-positive burden and equity.

OQ-22 — Can bipolar conversion risk be predicted well enough to change initial MDD treatment?

Register cohorts place long-term bipolar conversion near 6–9%, with family bipolarity, hospitalization, and psychotic depression producing hazard ratios around 1.7–2.7, but no validated threshold establishes a different first-line decision (Rhee 2023, PMID 37427550; Musliner 2018, PMID 29498031). A prospective multi-country model must report calibration, decision curves, and harms from false bipolar labeling.

OQ-23 — Can antidepressant choice optimize a patient-weighted harm profile?

Target-trial emulation found small average six-month weight differences across common agents—escitalopram +0.41 kg and bupropion −0.22 kg relative to sertraline—while sexual, sleep, withdrawal, and overdose outcomes remain measured in separate evidence systems (Petimar 2024, PMID 38950403; Zhou 2023, PMID 37422714). A pragmatic trial should use a preference-weighted composite without allowing one symptom scale to erase harm.

OQ-24 — What supervision intensity preserves task-shared psychotherapy effects at scale?

LMIC individual-participant meta-analysis found symptom benefit (Hedges g −0.32, 95% CI −0.38 to −0.26), response OR 2.11, and remission OR 1.87, but trial supervision may exceed routine capacity (Karyotaki 2022, PMID 35319740). A noninferiority implementation trial can randomize supervision dose while measuring fidelity, worker burden, deterioration, and reach.

OQ-25 — Which part of accelerated TMS produces the rapid effect?

Theta-burst synthesis supports left-DLPFC iTBS and bilateral cTBS+iTBS protocols, while personalized accelerated SNT bundles functional targeting, high pulse dose, multiple daily sessions, and expectancy (Kishi 2024, PMID 38844532; Cole 2022, PMID 34711062). A factorial dismantling platform should isolate target, spacing, total dose, and acceleration with a credible sham and long follow-up.

Tier 2 — important, but enabling work is needed

OQ-13 — Can MDD be partitioned into stable, treatment-relevant phenotypes?

STAR*D contained 1,030 symptom profiles among 3,703 people (Fried 2015, PMID 25451401). Stable longitudinal phenotyping must precede claims that cross-sectional clusters are disease subtypes.

OQ-14 — Which human mechanism is necessary for ketamine response?

Rapid controlled effects established a glutamatergic/plasticity paradigm (Berman 2000, PMID 10686270; Zarate 2006, PMID 16894061), but dissociation, opioid involvement, AMPA signaling, and metabolites remain unresolved.

OQ-15 — Which GWAS loci converge on perturbable cell types?

The 102-variant GWAS implicates synaptic/prefrontal biology but not individual treatment targets (Howard 2019, PMID 30718901). Fine-mapping, ancestry diversity, and functional validation are prerequisites.

OQ-16 — Can biomarkers demonstrate clinical utility rather than association?

Prospective biomarker synthesis found no clinically ready marker (Kennis 2020, PMID 31745238). Locked, external multi-site validation must precede randomized biomarker-guided care.

OQ-17 — How much MDD mortality is mediated by treatable medical pathways?

Mental disorders show roughly doubled all-cause mortality while GBD attributes almost no direct YLLs to them. A 2025 synthesis estimated depression-associated all-cause RR 2.10 (95% CI 1.87–2.35), suicide RR 9.89 (7.59–12.88), and natural-cause RR 1.63 (1.51–1.75), but heterogeneity approached 100% (Walker 2015, PMID 25671328; GBD 2019, PMID 35026139; Chan 2025, PMID 40948054). Linked causal cohorts and target-trial emulations are needed to partition disease, treatment, substance, socioeconomic, and care-access pathways.

OQ-18 — How portable are depression polygenic scores across ancestries and health systems?

Polygenic architecture is established, but discovery samples and prediction accuracy are ancestry-skewed (Howard 2019, PMID 30718901; Flint 2023, PMID 36702864). Multi-ancestry calibration and utility thresholds are needed before clinical use.

OQ-19 — What outcomes define recovery after TRD?

Lived-experience research emphasizes function, relationships, identity, and hope beyond symptom scores (Kerr 2023, PMID 37734244). A co-produced core outcome set should integrate treatment burden and durable participation.

OQ-20 — Which scalable psychotherapy components preserve efficacy?

Behavioral activation and CBT components show comparable network effects, while internet delivery raises engagement and support questions (Ciharova 2021, PMID 34264703; Furukawa 2021, PMID 33957075). Component trials require therapist clustering and deterioration reporting.

Dots not yet connected

These junctions were re-searched in PubMed on 2026-08-30. Most remain positively stated evidence gaps; where a direct preliminary study now exists, the row identifies the narrower replication or utility question that remains.

# Dot A Dot B The missing junction Powers
D-1 1,030 STAR*D symptom profiles antidepressant network averages symptom-profile-by-drug interaction using individual trial data OQ-1, OQ-13
D-2 inflammation-defined subgroup iTBS response external validation and a prespecified treatment-by-marker interaction after an exploratory 54-person clustering study (Pedraz-Petrozzi 2026, PMID 42364721) OQ-7, OQ-10
D-3 ketamine/esketamine rapid response psychotherapy learning a well-powered timing/dismantling trial after small randomized CBT studies produced mixed measure-level results (Wilkinson 2021, PMID 34186531; Wilkinson 2026, PMID 42095692) OQ-8, OQ-14
D-4 psilocybin experiential intensity functional unblinding mediation separating pharmacology, expectancy, and support OQ-9
D-5 pharmacogenomic interaction avoidance antidepressant withdrawal burden whether genotype-informed exposure predicts taper difficulty OQ-5, OQ-16
D-6 polygenic cross-disorder liability mixed-features specifier prospective bipolar conversion within DSM-MDD mixed features OQ-13, OQ-18
D-7 post-ECT relapse digital sleep/activity monitoring whether within-person digital change gives actionable early warning OQ-6, OQ-16
D-8 patient-defined recovery regulatory symptom endpoints a co-primary endpoint accepted in pivotal trials OQ-19
D-9 treatment-gap geography measurement-based care low-resource implementation with equity as a primary outcome OQ-12
D-10 suicide-attempt polygenic liability dynamic intent/access-to-means assessment incremental prospective utility over clinical formulation OQ-11, OQ-18
D-11 insulin resistance exercise network efficacy metabolic enrichment of exercise response OQ-10
D-12 antidepressant sexual dysfunction combination-treatment superiority net-benefit models incorporating preference-weighted sexual outcomes OQ-4, OQ-19
D-13 PHQ screening accuracy global treatment-gap data whether screening adds benefit when treatment capacity is constrained OQ-12, OQ-21
D-14 register-based bipolar conversion antidepressant activation/withdrawal phenotypes whether dynamic medication response improves calibrated conversion prediction OQ-5, OQ-22
D-15 task-shared psychotherapy efficacy patient-defined treatment burden supervision and delivery models optimized for both fidelity and acceptability OQ-19, OQ-24
D-16 cause-specific mortality excess intervention-specific observational mortality reductions a target-trial emulation separating treatment selection from causal survival benefit OQ-17