Towards personalized medicine: subtyping using functional profiles

Rates of Substance Use Disorders (SUDs) are at an all-time high, especially in the United States. Notably, a low percentage of individuals with SUDs seek available treatments, due at least in part to low confidence in treatment effectiveness and prior negative experiences. Psychiatric illnesses, including SUDs, are vastly complex [1], with many underlying neurobehavioral mechanisms that each may contribute to a different degree in different individuals with the same psychiatric illness, which has been termed the ‘heterogeneity problem’ [2]. This makes it unlikely that a single treatment will effectively treat all individuals with the same psychiatric illness. Therefore, understanding how profiles of behavioral and neural function vary from individual to individual within a diagnostic group is critical for advancing our ability to develop effective treatments.

Research on SUDs provides empirical evidence for the relevance of at least three domains of function: (1) Approach Behavior, (2) Executive Function, and (3) Negative Emotionality [1, 3, 4]. While these functional domains have often been described as dependent, decades of psychological research as well as insights from neuroscience inform us that they are, in fact, partially independent or separable, as we recently demonstrated empirically [3] (Fig. 1A). Importantly, if functional impairments on one domain are not correlated with functional impairments on another domain, what follows mathematically is the emergence of subtypes (Fig. 1B). We empirically demonstrated this for SUDs [3]. After establishing the relevance of multiple functional domains, we used a subtyping approach to define subtypes based on their functional profiles across these multiple domains. We found three subtypes, with each being uniquely impaired on only one of the three addiction-relevant domains [3]. Importantly, each subtype also presented a unique profile of brain function that was linked to their behavioral profile [3].

Fig. 1. The emergence of subtypes.

A We empirically demonstrated three separable domains of function (green: Approach Behavior, blue: Executive Function, red: Negative Emotionality) based on factor correlations from an exploratory factor analysis (N = 593) [3]. In this simplified representation, factors are depicted as nodes and factor correlations as edges connecting these nodes. Nodes that are more strongly correlated are plotted closer together [adapted from Drossel and colleagues [3]]. B (i) Scatter plot of a simulation that represents a community sample drawn from the entire population (N = 1000). Each dot in this simulation represents an individual’s measures of function on the three domains: Approach Behavior, Executive Function, Negative Emotionality. In this simulation, two assumptions are made: the measures of function in the population follow a normal distribution (distributions: mean = 0, standard deviation = 1) and function is not correlated across the three domains. (ii) Individuals who scored high (>1 standard deviation above mean) on one domain have been color-coded based on the domain in which they scored highly (green: Approach Behavior, blue: Executive Function, red: Negative Emotionality). Individuals with high scores in two domains are color-coded purple, dark green and dark brown. Individuals with high scores on all three domains are colored light brown. (iii) Only the individuals who scored high on at least one domain are depicted. (iv) Only individuals with high scores on a single domain are shown. Under the assumption that psychiatric patients will have a high score for at least one domain and that function is uncorrelated across domains, most psychiatric patients would have one but not multiple high scores, leading to the emergence of subtypes. Specifically, a majority (the ‘healthy people’) would have low/average scores on all three domains (black dots), while three subtypes of psychiatric patients with high scores on only one domain would emerge (red, blue, and green dots) in addition to smaller subsets of psychiatric patients who have high scores on multiple domains (other colored dots).

We refer to our approach as ‘mechanism-based’ subtyping, in contrast to ‘symptom-based’ subtyping approaches that have often been employed in the past. Among the over 120 addiction subtyping studies published since 1981, the vast majority identified subtypes using data assessing clinical symptom severity rather than underlying functional mechanisms. Unfortunately, symptom-based subtyping approaches have proven ineffective for defining functional targets that are necessary for the development of tailored addiction treatments. In contrast, ‘mechanism-based’ subtyping can provide such treatment targets, bringing a personalized medicine approach within grasp. For example, in a post-randomized controlled trial analysis, individuals with Alcohol Use Disorder were grouped into people who drank to relieve negative emotions versus to experience pleasure. Not surprisingly, only the individuals that drank to experience pleasure were successfully treated with the ‘pleasure-reducing’ medication naltrexone [5].

In conclusion, we propose that ‘mechanism-based’ subtyping has the potential to solve the ‘heterogeneity problem’ and move the field toward a personalized medicine approach. Another form of ‘mechanism-based’ subtyping is ‘brain-based’ subtyping (for a systematic review, see [6]), which may be particularly important for deriving functional targets for pharmacological or neuromodulation treatments. We are optimistic about a future in which a psychiatric patient can enter a clinic, complete a short subtyping assessment, and receive effective treatment based on their individual needs.

Author contributions

GD and AZ wrote the manuscript and prepared the figure.

Funding

GD is supported by a grant from the National Institute on Drug Abuse (5T32DA007234-29). AZ is supported by a grant from the National Institute on Alcohol Abuse and Alcoholism (RO1AA029406). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Competing interests

The authors declare no competing interests.