Classification systems like the Diagnostic and Statistical Manual of Mental Disorders (DSM) identify psychiatric disorders when individuals manifest a certain number of co-occurring symptoms from predefined lists. This approach results in “polythetic” diagnoses: individuals with the same diagnosis may share some symptoms but not all, leading to different symptom profiles. The “polythetic” nature of psychiatric diagnoses is sometimes used to question the validity of the classification systems for mental disorders and of mental disorders themselves.

A previous study (Fried & Nesse 2015) concluded that “depression is not a consistent syndrome” because individuals with major depressive disorder (MDD) showed more than 1,000 different combinations of symptoms assessed with a questionnaire. Such strong conclusion seems unwarranted.

First, it is important not to confuse the instrument (a questionnaire) used to measure a phenomenon (MDD) with the phenomenon itself – a conceptual error known as “reification”(Hyman 2010). To over-simplify: the scale used to measure weight is not the body’s mass. Moreover, it has been argued (Nunes et al. 2020) that simply counting potential symptom combinations doesn’t adequately capture the heterogeneity of psychiatric disorders; it’s important to evaluate how frequently the different combinations actually occur. This evaluation could reveal the most common forms of a disorder – despite the multitude of potential theoretical combinations of symptoms – that we are likely to encounter in reality, and toward which we should direct our research and clinical attention. This was the focus of the recent study by Spiller et al. (2024), which examined patterns of symptom combinations across different mental disorders.

Methods

This study leveraged various types of data.

First, Spiller et al. performed a computer simulation of a fictitious mental disorder in 500 individuals. The computer generated scores that represented answers to an hypothetical clinical instrument that examined 5 symptoms. Two out of the 5 symptoms were needed for a diagnosis, with a total of 32 possible combinations. The investigators performed 100 computer simulations, each time generating a positive diagnosis of the fictitious disorder in ~50% of the 500 subjects. Every simulation cycle reproduced conditions similar to those happening in reality, including different scores for each symptom and different interrelations between the symptoms.

Second, the authors used existing data from four large-scale datasets from the USA (3 from the Department of Veteran Affairs and 1 from the National Institute of Mental Health Data Archive) including electronic medical records with self-report instruments used to derive four DSM diagnoses:

1. PTSD Checklist for DSM-5 (PCL-5; 20 items) for Post-Traumatic Stress Disorder (PTSD) in 41,543 individuals.
2. Patient Health Questionnaire (PHQ-9; 9 items) for Major Depressive Disorder (MDD) in 46,259.
3. Generalized Anxiety Disorder questionnaire (GAD-7; 7 items) for Generalized Anxiety Disorder (GAD) in 63,742.
4. Positive and Negative Syndrome Scale (PANSS; 7 items) for Probable schizophrenia in 3,959.

In both the simulated and real-world data, the investigators calculated the frequencies of occurrence for every symptom combination.

Results

In the first part of the study that was based on computer simulations, the authors found that not all symptom combinations had the same probability of being expressed, rather the results showed:

highly skewed distribution of the combinations’ probabilities with few highly probable combinations and a majority with much lower probabilities.

This means that just a select few symptom combinations were likely to occur very frequently, while most other combinations were rare and unlikely to manifest. The same pattern of results emerged from the analyses of real-world data in the second part of the study.

In all datasets, most symptom combinations occurred rarely. For instance, in the depression dataset 90.5% of symptom combinations, that is 201 of the 222 combinations found, were reported by less than 1% of the sample. The proportion of symptom combinations endorsed by less than 1% of the subjects was also very high in the other datasets: 99.8% for PTSD, 50% for GAD and 41.7% for probable schizophrenia. In sum, these results suggest that in all disorders, the majority of potential theoretical combinations of symptoms were extremely rare.

Considering individuals, those endorsing one of the 10 most common combination of symptoms were the overwhelming majority in all datasets: 70.4% of the subjects with PTSD, 55.4% of those with MDD, 91.3% of those with probable schizophrenia and 84.9% of those with GAD. This suggests that the majority of individuals presented with a select few of the most common combinations of symptoms.

Conclusions

Overall, the study results demonstrated that assessing mental disorders using the current DSM classification system produced, as expected, a variety of symptom profiles. However, this heterogeneity in clinical presentations followed a consistent pattern. As the authors noted,

a few combinations of symptoms have an exceedingly high probability to occur, while this probability is markedly lower for most other possible symptom combinations.

In other words, although diagnoses of major mental disorders may show many different faces, some of these faces are far more common than others.

Strengths and limitations

I would like to highlight two major strengths of the study. First, the authors tackled the main research question using various methods, including computer simulations and analyses of real-world data. This triangulation approach — combining evidence from different methods — adds strength and consistency to the study’s findings. Study designs are each susceptible to different forms of bias. If we obtain similar results from multiple different study designs that aim to answer the same research question this gives us more confidence in the overall results and vice versa. Additionally, the analysis of real-world data was based on large samples that included detailed measurements of individual symptoms. This was essential for deriving overall diagnoses and counting the number of symptom combinations.

The authors also pointed out three main limitations of their study.

• First, the method of counting different symptom combinations did not take into account the fact that some combinations might share a significant number of symptoms. This is important because simply counting symptom combinations could give a misleading impression of a wide heterogeneity of clinical manifestations, which might not accurately reflect the clinical reality. This is especially true if many key symptoms are shared across different combinations.
• Second, the clear-cut distinction between the presence and absence of specific symptoms may oversimplify their varying degrees of expression.
• Lastly, the diagnoses of mental disorders were not based on structured psychiatric interviews; instead, they were derived by mimicking DSM criteria using self-report questionnaires from the datasets.

I would like to add one final limitation: data from three out of the four existing datasets (the larger ones) were drawn from electronic health records of the U.S. Department of Veterans Affairs. This means that the data may predominantly include individuals with more severe disorders, as those are the patients who are more likely to seek and receive care in specialized clinical institutions. Consequently, less severe disorders and their symptom manifestations that may be present in the general population may have been missed.

Implications for practice

Findings from this study by Spiller et al. stimulate us to acknowledge the heterogeneity inherent to polythetic mental disorder diagnoses. This is something that clinicians encounter daily in their practices: patients with the same diagnoses manifest very different, if not opposing, symptoms. For example, during a major depressive episode, some patients may experience a significant decrease in appetite and sleep, while others may experience an increase in both.

Furthermore, we need to learn how to harness this heterogeneity. We could hypothesise that different clinical manifestations may reflect partially divergent underlying pathophysiology’s requiring different treatments. This is the essence of a personalised medicine approach, in which patients are selected based on specific bio-clinical profiles and matched to treatments targeting particular disease mechanisms. For instance, in the field of depression accumulating evidence points to the involvement of inflammation as a disease mechanism in a portion of patients (read more about the role of inflammation in depression in these past Mental Elf blogs by Fairweather 2024 and Wessa 2024). Consistently, ongoing clinical studies (Khandaker et al. 2018; Otte et al. 2020; Zwiep at al. 2022; Wessa et al. 2024) in various European countries are trying to capture this subset of patients with depression based on combinations of biological parameters (blood concentrations inflammatory markers or body mass index levels) and clinical features (including symptoms like anhedonia, fatigue, appetite and sleep disturbances); the aim is to test the efficacy of anti-inflammatory add-on treatment for these patients. Such a personalised approach is still far away from being delivered to every-day psychiatry practice, as more research is needed to fully characterise the pathophysiology of different clinical manifestations, from environmental exposures to molecular mechanisms.

The results of Spiller et al. are reassuring: we can begin our exploration of heterogeneity not on an infinitely overwhelming range of clinical manifestations, but rather by focusing on the few prototypical symptom profiles that occur more frequently.

The findings of Spiller et al. further stimulate us to maintain a more pragmatic approach toward current diagnostic systems. First, it is important to acknowledge their value in the history of psychiatry, having enabled clinicians and researchers to communicate in a standardised way about mental disorders. At the same time, it is important to acknowledge their limitations. It is important to avoid the “reification” error and consider these systems for what they are, somehow simple tools through which we try to measure the very complex entities of mental disorders. These tools are far from perfect and are not designed to be definitive and set in stone, but they will keep evolving together with our understanding of the intrinsic mechanisms of mental disorders.

Statement of interests

Yuri is involved in a research line focusing on the exploration of depression heterogeneity, but he was not involved with the study presented here or its peer-review evaluation.

Links

Primary paper

Spiller TR, Duek O, Helmer M, et al. (2024) Unveiling the Structure in Mental Disorder Presentations. JAMA Psychiatry. 2024;81(11):1101–1107. doi:10.1001/jamapsychiatry.2024.2047

Other references

Fairweather, S. Inflammation and depression: new insights into sex differences in adolescents. The Mental Elf, 06 June 2024.

Fried EI, Nesse RM. Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study. J Affect Disord. 2015 Feb 1;172:96-102. doi: 10.1016/j.jad.2014.10.010. Epub 2014 Oct 14. PMID: 25451401; PMCID: PMC4397113.

Hyman SE. The diagnosis of mental disorders: the problem of reification. Annu Rev Clin Psychol. 2010;6:155-79. doi: 10.1146/annurev.clinpsy.3.022806.091532. PMID: 17716032.

Nunes A, Trappenberg T, Alda M. The definition and measurement of heterogeneity. Transl Psychiatry. 2020 Aug 24;10(1):299. doi: 10.1038/s41398-020-00986-0. PMID: 32839448; PMCID: PMC7445182.

Khandaker GM, Oltean BP, Kaser M, Dibben CRM, Ramana R, Jadon DR, Dantzer R, Coles AJ, Lewis G, Jones PB. Protocol for the insight study: a randomised controlled trial of single-dose tocilizumab in patients with depression and low-grade inflammation. BMJ Open. 2018 Sep 21;8(9):e025333. doi: 10.1136/bmjopen-2018-025333. PMID: 30244217; PMCID: PMC6157523.

Otte C, Chae WR, Nowacki J, Kaczmarczyk M, Piber D, Roepke S, Märschenz S, Lischewski S, Schmidt S, Ettrich B, Grabe HJ, Hegerl U, Hinkelmann K, Hofmann T, Janowitz D, Junghanns K, Kahl KG, Klein JP, Krueger THC, Leicht G, Prvulovic D, Reif A, Schoettle D, Strauss M, Westermair A, Friede T, Gold SM. Simvastatin add-on to escitalopram in patients with comorbid obesity and major depression (SIMCODE): study protocol of a multicentre, randomised, double-blind, placebo-controlled trial. BMJ Open. 2020 Dec 1;10(12):e040119. doi: 10.1136/bmjopen-2020-040119. PMID: 33262189; PMCID: PMC7709515.

Zwiep JC, Bet PM, Rhebergen D, Nurmohamed MT, Vinkers CH, Penninx BWJH, Milaneschi Y, Lamers F. Efficacy of celecoxib add-on treatment for immuno-metabolic depression: Protocol of the INFLAMED double-blind placebo-controlled randomized controlled trial. Brain Behav Immun Health. 2022 Dec 30;27:100585. doi: 10.1016/j.bbih.2022.100585. PMID: 36655056; PMCID: PMC9841030.

Wessa, C. Anti-inflammatory treatments for youth depression: promising but not yet implementable. The Mental Elf, 22 March 2024.

Wessa C, Janssens J, Coppens V, El Abdellati K, Vergaelen E, van den Ameele S, Baeken C, Zeeuws D, Milaneschi Y, Lamers F, Penninx B, Claes S, Morrens M, De Picker L. Efficacy of inflammation-based stratification for add-on celecoxib or minocycline in major depressive disorder: Protocol of the INSTA-MD double-blind placebo-controlled randomised clinical trial. Brain Behav Immun Health. 2024 Sep 19;41:100871. doi: 10.1016/j.bbih.2024.100871. PMID: 39350954; PMCID: PMC11440344.

Photo credits

• Photo by Rostyslav Savchyn on Unsplash
• Photo by Markus Spiske on Unsplash
• Photo by Nathan Dumlao on Unsplash
• Photo by Carlos Torres on Unsplash
• Photo by Filip Filkovic Philatz on Unsplash
• Photo by Headway on Unsplash