10,377 ways for major depression, but 341,737 ways for melancholia

We wrote up the core arguments and calculations of this blog post as an article for Lancet Psychiatry.

Major depression: a highly heterogeneous disorder

Major Depressive Disorder (MDD) is a complex and heterogeneous disorder. The DSM-5 that is commonly used to diagnose patients with MDD requires the presence of at least one of the two core criteria (sad mood and/or anhedonia), and at least 5 of the 9 criteria in total ¹.

A symptom profile that qualifies for depression, according to the DSM, would for instance be XXXXXX000 (X=yes, 0=no, position=symptom number). In this case, the patient endorses the first 6 symptoms, but not the last 3, which is sufficient for a diagnosis. A person with 00000X00X would not qualify, neither would a person with 00XXXXXXX (because the person does not have either of the two core symptoms).

Interestingly, many of these symptoms are compound criteria, i.e. symptoms that cover a broad range of subsymptoms. For instance, you can qualify for criterion 3 if you have either weight loss or weight gain or appetite loss or appetite gain. This makes counting symptom profiles a bit difficult, i.e. the question how many different ways there are to qualify for MDD. But we will get to that later.

Recent empirical papers have shown that patients with MDD differ drastically in their symptom profiles ^{2 3 4}. For instance, we identified over 1,000 unique depression symptom profiles in 3,700 patients diagnosed with MDD. Such heterogeneity can sometimes spell trouble … if you see 20 patients, and they all differ substantially in their problems, it is less likely that a one-size-fits-all treatment will help all of them. This intuition seems to line up well with recent empirical findings showing that different symptoms seem to respond differential to treatment ^{5 6}.

Here, for instance, are the symptoms listed across 7 common rating scales of depression, symptoms that are also fairly commonly endorsed by patients. Since depressed patients have a subset of these (and more) problems, treating all them with the same antidepressant would probably seem like a terrible idea ⁷.

Much has been written about depression heterogeneity — and I wrote my PhD on the topic — so I refer the interested reader to these papers, and references cited within ^{8 9 10}.

Depression subtypes

Over the last decades, a large number of depression subtypes such as psychotic depression, atypical depression, neurotic depression, endogenous depression, exogenous depression, anxious depression, sociotropic depression, autonomous depression, anaclitic depression, introjective depression, melancholic depression, and hopelessness depression have been proposed … and discarded. The main objective of subtyping has been to identify homogenous groups of individuals that may be more responsive to specific forms of treatment. A meta-review of 754 reviews published between the years 2000 and 2011 identified 15 commonly mentioned MDD subtypes ¹¹, and it has been suggested that these subtypes presumably differ from each other in five main aspects: symptom presentation, etiology, time of onset, gender, and treatment resistance. Unfortunately, evidence for these differences is weak. The DSM-5, for instance, notes that subtypes differ in their “presumed etiology”, an expression that reflects the ongoing problems of obtaining reliable etiology-subtype associations. Subtypes are not clearly demarcated, and no agreement has been reached as to their number or validity¹².

In 2015, Arnow et al. ¹³ analyzed data of 1,008 outpatients with Major Depressive Disorder (MDD) assigned to antidepressant treatment with three different drugs. The authors examined the impact of treatment on 3 common depression subtypes: melancholic depression, atypical depression, and anxious depression — and also on all 8 possible combinations of these 3 subtypes. They found that 75% of the patients qualified for at least one subtype, subtypes overlapped substantially, and neither subtypes nor subtype combinations differed in their general response to treatment or to specific antidepressants.

This paper was a good opportunity for my back-then office mate Frederik Coomans and me to look a bit further into the notion of depression subtypes, specifically melancholic depression.

Melancholia

The DSM lists several depression subtypes, or as they are called in the DSM-5: specifiers. The term specifier implies to me that there is something more specific about it than about the more general construct, in our case, MDD. Subtype has the same connotation: It is a more clearly defined part of the larger construct.

So Frederik and I first properly investigated how many unique symptom profiles exist to meet the criteria for MDD. Depending on the definition of criterion, we identified between 227 and 10,377 unique ways to qualify for MDD. You can read up on all the detailed math here, a document Frederik wrote up.

Next up, we looked into depression “with melancholic features”, i.e. the melancholia specifier. In the DSM-5, melancholia is a MDD-specifier characterized by symptoms such as loss of pleasure, lack of reactivity to usually pleasurable stimuli, profound despair, depressive mood especially early in the morning, early-morning awakening, marked psychomotor retardation or agitation, significant anorexia or weight loss, or excessive or inappropriate guilt.

Depression with melancholic features was first defined in the DSM-III, and critically discussed at the time of its implementation; today, the debate about the validity of melancholia is on-going and unresolved ¹⁴. For instance, it has been argued that the DSM criteria are not specific enough to delineate melancholia from other MDD subtypes or MDD itself, and symptoms such as blunted emotional response, pervasive anhedonia, and reduced libido have been proposed. Several concepts exist that are closely related to melancholia (e.g., endogenous, endogenomorphic, autonomous, vital, type A, and typical depression), the descriptive validity of melancholia remains questionable, and the consistency of melancholic features across episodes of depression is weak ^{15 16}. It is further problematic that individuals diagnosed with melancholic depression do not seem to differ in their response to antidepressant treatment from non-melancholic patients, as shown by Arnow et al. ^{17 18}, and it is unclear whether specific antidepressant treatment is more efficacious in patients with melancholic features. While some studies historically reported that specific antidepressants are consistently more effective than others, most studies failed to replicate this notion ¹⁹. In summary, melancholia remains a contentious subtype that is not clearly demarcated from depression or other subtypes in symptomatology and treatment response.

So what did our investigation result in?

We identified between 10,999 and 341,737 unique symptom profiles for melancholia, depending on how compound criteria are treated. This means there are about 50 times more symptom profiles compared MDD when counting conservatively (minimum estimate: 227 vs. 10,999), and over 30 times more symptom profiles (maximum estimate: 10,377 vs 341,737) when taking compound symptoms into account.

Conclusions

The pronounced heterogeneity, along with the substantial overlap of subtypes documented in the literature, casts doubt on the utility of clinical constructs such as melancholia. It is also at odds with the terminology of a “subtype” or “specifier” used in both the clinical and scientific literature on the topic. While our calculation does not show whether patients actually inhabit the full symptom space, this has been shown for other disorders ²⁰, and seems likely.

Looking at these results, and the results of Arnow et al. (2015) showing that depression subtypes are a) common and b) often overlap, I also wonder how much sense it makes to use common depression rating scales to track treatment response. Arnow et al. (2015) used a scale that measures the 9 DSM-5 MDD symptoms to track treatment response, but since most depressed participants qualified for at least one subtype — implying that they exhibited severe problems beyond ‘pure’ MDD — the actual impact of treatment remains unresolved, because patients with e.g. melancholia have many problems that are not part of the MDD criterion symptoms.

Another issue is that many depression symptoms resemble antidepressant side-effects such as weight change and sleep disturbances. Tracking sum-scores of problems over time obfuscates that some symptoms may decrease in response to (specific) treatment, whereas others may increase.

Assessing a large number of individual symptoms in clinical trials — both DSM symptoms and other symptoms prevalent in depressed populations — offers a solution to the aforementioned challenges, and I’m very happy to have found the incredibly talented and tireless Zach Cohen who is currently putting together a mega-analysis of depression studies with me to tackle this question. If you want to collaborate or share your clinical trial data on depression with us, that would be …

I think there is a larger paper to be written about this, trying to tackle the following questions: a) For how many disorders are there subtypes in the DSM? b) What functions do these subtypes officially have, and what functions should they have? c) What criteria are necessary and sufficient to define a subtype? If any clinical editor would be interested in a submission, or someone wanted to collaborate on this, please let me know.

Challenges

An obvious question I did not tackle here is how different unique symptom profiles really are, and that’s something we would need to tackle in a proper paper. Let’s revisit my own work with and put on our skepticism glasses. As I said above, we previously identified 1,000 unique symptom profiles in 3,700 depressed patients, and counted symptom profiles to differ from each other if they differed in a single symptom. And while that is likely the one way to approach the question of unique profiles, it is quite obvious that a patient who endorses the first 8 depression symptoms and not the 9th one, XXXXXXXX0, is very similar to a patient who has all 9 symptoms XXXXXXXXX, or a patient who endorses XXXXXXX0X; whereas 0XXXXX000 differs dramatically from X0000XXXX — they only share a single symptom (#6). There are numerous ways to go about this mathematically, and one way forward is estimating the average distances of symptom profiles in a 9-dimensional space, and quantifying these. Charles Olbert has worked on some methods of quantification ²¹, and I hope to find the time to look into this in more detail in the future.

PS: For this blog, I adapted or copied parts of my dissertation, and of a letter written by Frederik Coomans and me in 2015 that never saw the light of day. Thanks for helping me figure this out Frederik!

1. There are some other requirements such as impairment, but they are not relevant for the purposes of this blog post
2. Fried, E. I., & Nesse, R. M. (2015). Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study. Journal of Affective Disorders, 172, 96–102. http://doi.org/10.1016/j.jad.2014.10.010
3. Zimmerman, M., Ellison, W., Young, D., Chelminski, I., & Dalrymple, K. (2014). How many different ways do patients meet the diagnostic criteria for major depressive disorder? Comprehensive Psychiatry. http://doi.org/10.1016/j.comppsych.2014.09.007
4. Olbert, C. M., Gala, G. J., & Tupler, L. A. (2014). Quantifying Heterogeneity Attributable to Polythetic Diagnostic Criteria : Theoretical Framework and Empirical Application. Journal of Abnormal Psychology, 123(2), 452–462. http://doi.org/10.1037/a0036068
5. Hieronymus, F., Emilsson, J. F., Nilsson, S., & Eriksson, E. (2016). Consistent superiority of selective serotonin reuptake inhibitors over placebo in reducing depressed mood in patients with major depression. Molecular Psychiatry, 21(4), 523–30. http://doi.org/10.1038/mp.2015.53
6. Chekroud, A. M., Gueorguieva, R., Krumholz, H. M., Trivedi, M. H., Krystal, J. H., & McCarthy, G. (2017). Reevaluating the Efficacy and Predictability of Antidepressant Treatments: A Symptom Clustering Approach. JAMA Psychiatry, 06511. http://doi.org/10.1001/JAMAPSYCHIATRY.2017.0025
7. Unless different symptoms are the result of the same disorder of course, which is entirely possible; syphilis is a great example. But I think it is unlikely to be the case for depression; here’s why
8. Fried, E. I. (2017). The 52 symptoms of major depression. Journal of Affective Disorders, 208, 191–197. http://doi.org/10.1016/j.jad.2016.10.019
9. Fried, E. I., & Nesse, R. M. (2015). Depression sum-scores don’t add up: why analyzing specific depression symptoms is essential. BMC Medicine, 13(72), 1–11. http://doi.org/10.1186/s12916-015-0325-4
10. Fried, E. I. (2015). Problematic assumptions have slowed down depression research: why symptoms, not syndromes are the way forward. Frontiers in Psychology, 6(306), 1–11. http://doi.org/10.3389/fpsyg.2015.00309
11. Baumeister, H., & Parker, J. D. (2012). Meta-review of depressive subtyping models. Journal of Affective Disorders, 139(2), 126–40. http://doi.org/10.1016/j.jad.2011.07.015
12. I list about 10 references to reviews of depression subtypes in my dissertation
13. Arnow, B. A., Blasey, C., Williams, L. M., Palmer, D. M., Rekshan, W., Schatzberg, A. F., … Rush, A. J. (2015). Depression Subtypes in Predicting Antidepressant Response: A Report From the iSPOT-D Trial. American Journal of Psychiatry, 172(8), 743–750. http://doi.org/10.1176/appi.ajp.2015.14020181
14. Melartin, T., Leskelä, U., Rytsälä, H., Sokero, P., Lestelä-Mielonen, P., & Isometsä, E. (2004). Co-morbidity and stability of melancholic features in DSM-IV major depressive disorder. Psychological Medicine, 34(08), 1443. http://doi.org/10.1017/S0033291704002806
15. Melartin, T., Leskelä, U., Rytsälä, H., Sokero, P., Lestelä-Mielonen, P., & Isometsä, E. (2004). Co-morbidity and stability of melancholic features in DSM-IV major depressive disorder. Psychological Medicine, 34(08), 1443. http://doi.org/10.1017/S0033291704002806
16. Baumeister, H., & Parker, J. D. (2012). Meta-review of depressive subtyping models. Journal of Affective Disorders, 139(2), 126–40. http://doi.org/10.1016/j.jad.2011.07.015
17. See also: Brown, W. A. (2007). Treatment response in melancholia. Acta Psychiatrica Scandinavica, (433), 125–9. http://doi.org/10.1111/j.1600-0447.2007.00970.x
18. McGrath, P. J., Khan, A. Y., Trivedi, M. H., Stewart, J. W., Morris, D. W., Wisniewski, S. R., … Rush, A. J. (2008). Response to a selective serotonin reuptake inhibitor (citalopram) in major depressive disorder with melancholic features: a STAR*D report. The Journal of Clinical Psychiatry, 69(12), 1847–55. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/19026268
19. Uher, R., Dernovsek, M. Z., Mors, O., Hauser, J., Souery, D., Zobel, A., … Farmer, A. (2011). Melancholic, atypical and anxious depression subtypes and outcome of treatment with escitalopram and nortriptyline. Journal of Affective Disorders, 132(1–2), 112–20. http://doi.org/10.1016/j.jad.2011.02.014
20. Zimmerman, M., Ellison, W., Young, D., Chelminski, I., & Dalrymple, K. (2014). How many different ways do patients meet the diagnostic criteria for major depressive disorder? Comprehensive Psychiatry. http://doi.org/10.1016/j.comppsych.2014.09.007
21. Olbert, C. M., Gala, G. J., & Tupler, L. A. (2014). Quantifying Heterogeneity Attributable to Polythetic Diagnostic Criteria : Theoretical Framework and Empirical Application. Journal of Abnormal Psychology, 123(2), 452–462. http://doi.org/10.1037/a0036068