Systems‐based approaches to mental disorders are the only game in town

Stein et al's paper ¹ provides an excellent overview of current directions in psychiatric diagnosis. The paper makes clear that, although there has been considerable work investigating novel approaches to psychiatric nosology, psychiatric diagnosis has in practice changed relatively little in recent decades. Mental disorders are defined and diagnosed today in pretty much the same way they have been for many years: as sets of symptoms that tend to cluster in somewhat reliable ways. Hallucinations are often accompanied by delusions; sad mood by self‐reproach; anxiety by avoidance. Thresholds based on such symptom clusters are typically used to operationally define mental disorders, and the presentation of symptoms in a person is phenomenologically matched to these definitions to arrive at a diagnosis that guides treatment.

In recent years, much research operated under the assumption that, under the hood, psychiatric disorders are brain disorders ² , and that advances in neuroscience and genetics would reveal “what mental disorders really are”. It is evident that no such breakthrough has materialized. It seems that most mental disorders simply lack central pathogenic pathways. Instead, they have turned out to be massively multifactorial in causes and constitution, involving a highly complicated and barely understood interplay between genes, neural processes, behavior, environment and culture. Symptoms of different disorders often overlap; many disorders exhibit exceptional levels of comorbidity; and transdiagnostic factors and processes are the norm rather­ than the exception. For these reasons, the separation of mental disorders into distinct disease entities often appears artificial, and the diagnostic categories used in the DSM and ICD can be a Procrustean bed when applied to individual cases.

Given this massively multifactorial back­ground, the biopsychosocial model has the best cards as a framework for understanding mental disorders. After all, the scientific evidence shows that: a) factors at societal, psychological and biological levels are involved in mental disorders; b) these factors interact across different time scales and levels of analysis; and c) interactions between factors feature nonlinearities (i.e., factors do not combine in a simple additive fashion). However, as Stein et al note, unless one answers the question of how psychological, biological and social factors interact to cause and maintain mental disorders, the net theoretical content of this model is close to zero. How then shall we address this question in the next century of research on psychiatric diagnosis and treatment? We suggest that, in this respect, a systems‐based approach is the only game in town.

A systems‐based approach, as practiced in other domains of science, allows us to explicitly model the interactions among a set of components across time scales and levels of analysis. These models, in turn, allow us to investigate those systems and evaluate how they give rise to the phenomena of interest. In the domain of mental health, a systems‐based approach allows us to take the compelling but vague biopsychosocial framework and make it concrete, positing the precise system that gives rise to the etiology, maintenance and recovery from a mental disorder. The past decades have seen massive advances in methodology and modeling strategies suited to study complex systems ³ . If humanity can build climate models to project the effect of political interventions on global temperatures, it should also be possible to build models that can project the effect of therapeutic interventions on mental disorders.

Central to a systems‐based approach are models that express our theories about how components of a system interact in the language of mathematics or computational programming. Such mathematical or computational models are generative, which means that they allow us to simulate the etiology and maintenance of mental disorders. For instance, our group has used very simple network models to show how interactions between symptoms could lead people to get “stuck” in an episode of depression ⁴ . Generative models also allow us to make changes to the system and thereby simulate treatment interventions. For instance, in a network model, one can simulate shocks to network elements or the effect of breaking links between them5, 6. This procedure has already been used to mimic existing interventions ⁷ , and could be used to discover new ones.

It is this ability to precisely deduce what our theories predict about etiology and treatment that make mathematical or computational models so crucial to the future of psychiatric diagnosis and treatment. It is all but impossible to intuit the behavior of complex systems through mental reasoning alone. Indeed, the complex systems literature is replete with examples of even relatively simple systems behaving in chaotic and unpredictable ways (e.g., the simple Lorenz equations giving rise to the famous butterfly‐shaped strange attractors). Given the heterogeneous and multifactorial nature of mental disorders, it will be all but hopeless to advance our understanding of these disorders without the assistance of formal models.

Importantly, generative models are different from the data‐analytic models in which mental health researchers are primarily trained. Data‐analytic models can be estimated from a single dataset and rep­resent patterns in the data. In contrast, generative models are developed by integrating empirical findings from many studies with different data on different levels of analysis (e.g., neuroscientific and behavior­al) and creating a model that represents the real‐world system that gave rise to those empirical findings.

How to best use empirical research to inform a generative model is an open question and an important area of research, though potential approaches already exist in the mental health literature8, 9. For our purposes here, the key is that generative models provide a tool that is distinct from the data‐analytic models that dominate much of psychiatric research. Critically, this means that future generations of modelers should focus on building generative models alongside data‐analytic ones. Theory building skills will be as important to the future of psychiatric research as empirical research skills, and the curriculum we offer students should reflect that.

The models that have been developed in early efforts to adopt a systems‐based approach in psychiatric research are relatively simple and, in most cases, best seen as toy models. However, the fact that it has been possible to construct these models gives rise to a modest hope. It is important to emphasize the word “modest” here. Examples spanning from pandemics to financial crashes and from climate change to polarization have taught us that the behavior of complex systems is extremely difficult to predict and control, even with the assistance of formal models. We should not expect magic bullets or free lunches. Similar to Stein et al, we believe that understanding mental disorders will require an integrative and iterative process of systematic clinical observation, painstaking research, and creative thinking. The value of a systems‐based approach is that it provides a framework for organizing and tools for promoting the accumulation of knowledge through this iterative process and equips us to better leverage that knowledge to improve psychiatric care.