Skip to content

Diagnostic criteria and heterogeneity

TL;DR — Major depressive disorder (MDD) is a clinical syndrome, not a laboratory-defined disease. DSM-5 requires at least five of nine symptoms during the same two-week period, including depressed mood or loss of interest, plus distress or impairment and exclusions; ICD-11 uses a similar symptom-domain approach but differs in counting and severity rules. This permissive architecture generates hundreds of qualifying symptom profiles, and STAR*D found 1,030 distinct profiles among 3,703 outpatients; nearly half occurred in only one person (Fried 2015, PMID 25451401). Specifiers identify clinically important dimensions, but do not solve biological heterogeneity. The practical diagnostic priority is to establish episode, impairment, course, safety, medical/substance contributors, and the possibility of bipolarity before assigning a unipolar label.

Operational criteria

Element DSM-5/DSM-5-TR operationalization Research consequence
Core syndrome At least five of nine symptoms in the same 2 weeks Threshold converts a dimensional symptom distribution into a category
Mandatory feature Depressed mood or anhedonia People dominated by fatigue, sleep, or cognition may fall below threshold
Change Symptoms represent change from prior functioning Requires longitudinal history, not score alone
Significance Clinically significant distress or impairment Context and clinician judgment remain essential
Exclusions Substance/medical causation; psychotic disorders; no manic or hypomanic episode Bipolar and secondary-depression errors alter treatment logic
Bereavement No automatic bereavement exclusion Grief context remains part of differential diagnosis

DSM-5 changes from DSM-IV included removal of the bereavement exclusion, revised chronic-depression categories, and new anxious-distress and mixed-features specifiers; these changes affect both case counts and trial comparability (Uher 2014, PMID 24272961). A diagnostic review emphasizes that classification must be paired with treatment-relevant dimensions rather than treated as a sufficient explanation of illness (Kennedy 2018, PMID 29278937).

The combinatorial problem

Nine binary symptoms with a five-symptom threshold produce many legal combinations even before symptom direction is considered. Sleep, appetite, and psychomotor symptoms can each move in opposite directions; severity and timing vary; anxiety, irritability, pain, and cognition are incompletely represented.

In STAR*D, 3,703 participants expressed 1,030 unique symptom profiles; 864 profiles occurred in five or fewer participants and 48.6% occurred in one individual (Fried 2015, PMID 25451401). The result does not prove 1,030 diseases. It demonstrates that the diagnosis is a broad equivalence class and that group-average biology can wash out subgroup signals.

Heterogeneity axis Examples Why it matters
Symptom content insomnia vs hypersomnia; loss vs gain of appetite Opposite physiology can share one total score
Course single episode, recurrent, chronic Recurrence and chronicity affect prognosis and maintenance decisions
Severity mild through psychotic Baseline severity changes absolute treatment benefit and urgency
Age/context adolescent, perinatal, late-life, medical illness Differential diagnosis and risk-benefit balance shift
Comorbidity anxiety, substance use, personality, pain Drives impairment and apparent non-response
Mechanism inflammatory, circadian, stress-related, synaptic Proposed subgroups are not yet validated clinical taxa

Specifiers and presentations

Specifier/presentation Defining signal Evidence status
Anxious distress tension, restlessness, fear of losing control Common: 74.6% in NESARC-III DSM-5 MDD; associated with worse course (Hasin 2018, PMID 29450462)
Mixed features subthreshold manic symptoms during depression 15.5% in NESARC-III; should prompt repeated bipolar assessment (Hasin 2018, PMID 29450462)
Melancholic anhedonia/non-reactivity plus psychomotor, diurnal, appetite, guilt features Clinically recognizable, but inter-rater and biomarker boundaries remain imperfect
Atypical mood reactivity plus hypersomnia, hyperphagia, leaden paralysis, rejection sensitivity Descriptive phenotype; not synonymous with mild illness
Psychotic delusions or hallucinations during episode Changes acute treatment; pharmacologic evidence supports combined strategies over antidepressant alone (Oliva 2024, PMID 38360024)
Seasonal pattern recurrent temporal relationship to season Requires longitudinal pattern, not winter worsening once
Peripartum onset pregnancy or early postpartum timing Separate hormonal, safety, and service considerations

Boundary with bipolar depression

A current depressive episode can look identical in bipolar and unipolar illness. The decisive evidence is lifetime mania or hypomania, which is frequently missed when history focuses only on current distress. Earlier onset, recurrent brief episodes, family history of bipolar disorder, antidepressant-associated activation, mixed features, and episodic decreased need for sleep raise suspicion but are not individually diagnostic. Differential-diagnosis reviews therefore center longitudinal course and collateral history (Hirschfeld 2014, PMID 25533909).

The cost of a false-unipolar classification is asymmetric: it may expose a bipolar patient to antidepressant monotherapy and delay mood-stabilizing treatment. The opposite error can expose an MDD patient to unnecessary long-term mood stabilizers or antipsychotics. Screening questionnaires can structure history but cannot replace a diagnostic interview.

Measurement is not diagnosis

Instrument Typical role Limitation
PHQ-9 screening and serial symptom measurement Positive screen is not MDD; item 9 requires direct safety assessment
QIDS symptom severity and STAR*D-style measurement Total score hides symptom direction and content
HAM-D clinician-rated trial endpoint Item weighting and version differences complicate comparison
MADRS clinician-rated change-sensitive endpoint Does not establish exclusions, course, or bipolarity

Measurement-based care improves decisions when scales trigger systematic action, but scales are adjuncts: a randomized trial found rating-guided care superior to standard care, not that a rating scale can diagnose MDD by itself (Guo 2015, PMID 26315978).

Differential diagnosis and secondary depression

The assessment must separate MDD from bipolar depression, persistent depressive disorder, adjustment disorder, grief disorders, substance-induced states, medication effects, sleep disorders, endocrine disease, neurological illness, delirium, and primary psychotic illness. The evidentiary threshold for testing depends on history and examination; indiscriminate panels create false positives, while no medical review risks misclassification.

Red flags for an alternative or additional explanation include abrupt onset with cognitive fluctuation, focal neurological signs, new late-life first episode, temporal linkage to a drug or substance, episodic elevated energy/decreased need for sleep, psychosis outside mood episodes, and symptoms confined to a specific physiological window. These signals are developed in red flags and safety concerns.

What a better classification would need

A useful next-generation system would preserve reliable clinical communication while adding dimensions that predict course or treatment. Candidate layers include symptom networks, recurrence, cognitive profile, inflammatory/metabolic state, sleep/circadian phenotype, trauma exposure, and circuit-level measures. Prospective biomarker synthesis has not produced a marker with clinical diagnostic performance (Kennis 2020, PMID 31745238), and current genetic effects are far too small for individual diagnosis (Howard 2019, PMID 30718901).

Screening is triage, not case ascertainment

Instrument/threshold Reference comparison Accuracy estimate Consequence
PHQ-9 ≥10 Semistructured interview; 58 studies, 17,357 participants sensitivity 0.88 (95% CI 0.83–0.92); specificity 0.85 (0.82–0.88) Predictive value still depends on prevalence (Levis 2019, PMID 30967483)
PHQ-9 ≥10, updated 100 studies, 44,503 participants sensitivity 0.85 (0.79–0.89); specificity 0.85 (0.82–0.87) The update narrowed uncertainty but did not make the scale diagnostic (Negeri 2021, PMID 34610915)
PHQ-2 ≥2 Semistructured interview sensitivity 0.91 (0.88–0.94); specificity 0.67 (0.64–0.71) Efficient first-stage triage sacrifices specificity (Levis 2020, PMID 32515813)
PHQ-2 ≥2 then PHQ-9 ≥10 Two-stage strategy sensitivity 0.82 (0.76–0.86); specificity 0.87 (0.84–0.89); 57% fewer full PHQ-9 administrations Workflow efficiency is not a DSM/ICD diagnosis (Levis 2020, PMID 32515813)

Reference-standard choice changes apparent caseness. At the same symptom burden, the MINI classified perinatal depression more often than the CIDI (adjusted OR 3.72, 95% CI 1.21–11.43) (Levis 2019, PMID 31568624). In a broader comparison, MINI classification odds were 2.10-fold those of CIDI (95% CI 1.15–3.87), while fully structured versus semistructured interviews behaved differently across PHQ-9 ranges (Levis 2018, PMID 29717691). European prevalence modeling corrected PHQ-8 estimates for imperfect diagnostic accuracy, illustrating why screening-derived country rankings require calibration (Fischer 2023, PMID 37024144). A PHQ-2 diagnostic meta-analysis supports its use as a brief route to assessment, not as a stand-alone label (Manea 2016, PMID 27371907).

Longitudinal diagnostic instability

Cohort Follow-up result Strong predictors Boundary implication
Swedish population registers 13-year bipolar conversion 5.84% (95% CI 5.72–5.96) bipolar family-risk score HR 2.73 (2.43–3.08), inpatient setting HR 2.64 (2.44–2.84), psychotic depression HR 2.58 (2.14–3.11) A unipolar diagnosis is time-indexed, especially after severe/psychotic episodes (Rhee 2023, PMID 37427550)
Finnish first-hospitalization cohort, n=43,495 15-year conversion 11.1%: bipolar 7.4%, schizophrenia 2.5%, schizoaffective 1.3% psychotic depression predicted all three; risk peaked in year 1 Follow-up must reassess psychosis, mania/hypomania, and course (Baryshnikov 2020, PMID 32385906)
Danish hospital cohort, n=91,587 bipolar conversion 8.7% in women and 7.7% in men parental bipolar disorder aHR 2.60 (2.20–3.07); inpatient care 1.76 (1.63–1.91); psychotic depression 1.73 (1.48–2.02) Risk markers shift probability but do not diagnose bipolar disorder (Musliner 2018, PMID 29498031)

Additional cohorts converge on family bipolarity, early onset, recurrence, psychotic features, hospitalization, and selected substance exposures as risk markers, but have not established transportable decision thresholds (Oliveira 2021, PMID 33493732; de Azevedo Cardoso 2020, PMID 32306467; Zhu 2025, PMID 40225731). Mixed-symptom scales improve description, yet validation of a scale does not establish a discrete mixed-depression disease entity (Sani 2018, PMID 29459190).

Do subtypes carve nature at its joints?

A systematic review of 20 studies and 34 latent-variable analyses found two to five classes, but most between-class symptom differences tracked overall severity; no stable symptom clusters emerged across analytic choices (van Loo 2012, PMID 23210727). This argues against treating one cross-sectional clustering solution as a biological subtype. It does not rule out longitudinal, mechanistic, or treatment-responsive subtypes.

Position Supporting observation Counterweight
Melancholic/atypical/psychotic categories are treatment-relevant Psychotic depression changes acute strategy; atypical-depression trials sometimes yield differential rankings (Fornaro 2025, PMID 40412292) Definitions, comparator sets, and diagnostic reliability vary
Data-driven classes will solve heterogeneity Latent methods can discover structure without DSM assumptions Published solutions largely reproduce severity and depend on preprocessing/model choice (van Loo 2012, PMID 23210727)
A screen can stand in for diagnosis at scale PHQ tools are efficient and reasonably accurate Interview method and base rate materially change who becomes a “case” (Negeri 2021, PMID 34610915)
Unipolar/bipolar separation is cross-sectionally observable Some history and symptom features shift likelihood Conversion cohorts show that decisive information may emerge only longitudinally (Rhee 2023, PMID 37427550)

Open questions

  • Which symptom dimensions are stable enough across episodes to define reproducible subgroups rather than temporary states (Fried 2015, PMID 25451401)?
  • Can a classification combining course, symptoms, and biology outperform DSM/ICD categories in prospective treatment selection (Kennis 2020, PMID 31745238)?
  • How often does the mixed-features specifier identify future bipolar disorder rather than severe unipolar MDD (Hasin 2018, PMID 29450462)?
  • Which minimal longitudinal assessment most reduces bipolar misclassification without excessive false positives (Hirschfeld 2014, PMID 25533909)?

References

  1. Uher R, et al. Major depressive disorder in DSM-5: implications for clinical practice and research of changes from DSM-IV. Depression and Anxiety. 2014. PMID 24272961
  2. Kennedy SH. Unpacking Major Depressive Disorder: From Classification to Treatment Selection. Canadian Journal of Psychiatry. 2018. PMID 29278937
  3. Fried EI, et al. Depression is not a consistent syndrome: an investigation of unique symptom patterns in the STAR*D study. Journal of Affective Disorders. 2015. PMID 25451401
  4. Hasin DS, et al. Epidemiology of Adult DSM-5 Major Depressive Disorder and Its Specifiers in the United States. JAMA Psychiatry. 2018. PMID 29450462
  5. Hirschfeld RM. Differential diagnosis of bipolar disorder and major depressive disorder. Journal of Affective Disorders. 2014. PMID 25533909
  6. Oliva V, et al. Pharmacological treatments for psychotic depression: a systematic review and network meta-analysis. Lancet Psychiatry. 2024. PMID 38360024
  7. Guo T, et al. Measurement-Based Care Versus Standard Care for Major Depression: A Randomized Controlled Trial With Blind Raters. American Journal of Psychiatry. 2015. PMID 26315978
  8. Kennis M, et al. Prospective biomarkers of major depressive disorder: a systematic review and meta-analysis. Molecular Psychiatry. 2020. PMID 31745238
  9. Howard DM, et al. Genome-wide meta-analysis of depression identifies 102 independent variants and highlights the importance of the prefrontal brain regions. Nature Neuroscience. 2019. PMID 30718901
  10. van Loo HM, et al. Data-driven subtypes of major depressive disorder: a systematic review. BMC Medicine. 2012;10:156. PMID 23210727
  11. Rhee SJ, et al. Predictors of diagnostic conversion from major depression to bipolar disorder. Psychological Medicine. 2023;53:7805-7816. PMID 37427550
  12. Baryshnikov I, et al. Diagnostic conversion from unipolar depression to bipolar disorder, schizophrenia, or schizoaffective disorder. Bipolar Disorders. 2020;22:582-592. PMID 32385906
  13. Musliner KL, et al. Patterns and predictors of conversion to bipolar disorder in 91,587 individuals. Acta Psychiatrica Scandinavica. 2018;137:422-432. PMID 29498031
  14. Levis B, et al. Accuracy of PHQ-9 for screening to detect major depression. BMJ. 2019;365:l1476. PMID 30967483
  15. Negeri ZF, et al. Accuracy of PHQ-9: updated systematic review and IPD meta-analysis. BMJ. 2021;375:n2183. PMID 34610915
  16. Levis B, et al. Accuracy of PHQ-2 alone and combined with PHQ-9. JAMA. 2020;323:2290-2300. PMID 32515813
  17. Levis B, et al. Classification probability using SCID, CIDI, and MINI in pregnancy/postpartum. International Journal of Methods in Psychiatric Research. 2019;28:e1803. PMID 31568624
  18. Levis B, et al. Classification using semi-structured versus fully structured interviews. British Journal of Psychiatry. 2018;212:377-385. PMID 29717691
  19. Fischer F, et al. MDD prevalence in 27 European countries accounting for imperfect PHQ-8 accuracy. BMJ Mental Health. 2023;26:e300675. PMID 37024144
  20. Fornaro M, et al. Pharmacological treatments for atypical depression: network meta-analysis. European Neuropsychopharmacology. 2025;96:46-57. PMID 40412292
  21. Oliveira JP, et al. Predictors of conversion from MDD to bipolar disorder. Psychiatry Research. 2021;297:113740. PMID 33493732
  22. de Azevedo Cardoso T, et al. Lifetime cocaine use and conversion from MDD to bipolar disorder. Psychiatry and Clinical Neurosciences. 2020;74:418-423. PMID 32306467
  23. Zhu T, et al. Predicting conversion from unipolar depression to bipolar disorder and schizophrenia. Depression and Anxiety. 2025;2025:4048082. PMID 40225731
  24. Sani G, et al. Koukopoulos Mixed Depression Rating Scale validation study. Journal of Affective Disorders. 2018;232:9-16. PMID 29459190
  25. Manea L, et al. Identifying depression with the PHQ-2: diagnostic meta-analysis. Journal of Affective Disorders. 2016;203:382-395. PMID 27371907