Unsupervised data-driven stratification of mentalizing heterogeneity in autism

Abstract

Individuals affected by autism spectrum conditions (ASC) are considerably heterogeneous. Novel approaches are needed to parse this heterogeneity to enhance precision in clinical and translational research. Applying a clustering approach taken from genomics and systems biology on two large independent cognitive datasets of adults with and without ASC (n = 694; n = 249), we find replicable evidence for 5 discrete ASC subgroups that are highly differentiated in item-level performance on an explicit mentalizing task tapping ability to read complex emotion and mental states from the eye region of the face (Reading the Mind in the Eyes Test; RMET). Three subgroups comprising 45-62% of ASC adults show evidence for large impairments (Cohen's d = −1.03 to −11.21), while other subgroups are effectively unimpaired. These findings delineate robust natural subdivisions within the ASC population that may allow for more individualized inferences and accelerate research towards precision medicine goals.Autism spectrum conditions (ASC) are currently defined by consensus behavioral criteria of difficulties in social-communication and restricted repetitive behaviors. Although the population is subsumed under a single unitary diagnostic label, variability between affected individuals is considerable 1 . This diversity or 'heterogeneity' inherent within ASC can be seen at multiple levels, from a myriad of different etiological mechanisms 2 , developmental trajectories 3,4 , sex/gender 5 , clinical comorbidities 6 , cognitive/behavioral features (e.g., language development 7 ), and the list could go on 8 . Accordingly, many in the field now subscribe to the idea that there is not just one 'autism' , but rather multiple 'autisms' 9,10 . Given these rich theoretical ideas about the complexity manifesting behind unitary clinical diagnostic labels like 'autism' , it seems like a natural extension of logic that research would follow along such ideas. However, the opposite occurs a majority of the time, whereby clinical and translational research is done utilizing case-control comparison methodology that treats autism as one omnibus group and makes comparison to a matched 'control' or 'comparison' group. While the standard case-control

Big data approaches to decomposing heterogeneity across the autism spectrum

“…From a case-control perspective, adults with autism perform on-average lower on the 'Reading the Mind in the Eyes' Test (RMET) compared to matched typically-developing controls 68 . However, taking a stratified approach, we find that the autistic adult population can be reliably split into subtypes that either are completely unimpaired on the RMET, versus those who are highly impaired 24 (Fig 2). Thus, in this example, while replicable on-average case-control effects appear, a stratified approach that takes into account heterogeneity can isolate higher impact and more precise considerations about mentalizing as measured by the RMET in the adult autistic population.…”

“…Fig 2: Case-control vs stratified model example with adult autism and mentalizing ability. This figure reports data from Lombardo et al, (2016) 24…”

“…Research has begun along these new directions but is highly fractionated because heterogeneity is discussed across multiple levels of analysis, from genetics 11 , neural systems [12][13][14] , cognition 15 , behavior and development [16][17][18] , and clinical topics (e.g., response to treatment, outcome 19,20 ). Approaches differ in how heterogeneity is decomposed, from utilizing theoretical a priori known stratifiers 21,22 or dimensions, to completely data-driven approaches 12,[23][24][25] . Models for understanding heterogeneity also differ considerably, with some conceptualizing distinctions as categorical/qualitative, continuous/dimensional, and/or where distinctions or similarities may cut across diagnostic boundaries [25][26][27] .…”

Reconciling Dimensional and Categorical Models of Autism Heterogeneity: a Brain Connectomics & Behavioral Study

“…A major challenge in developing biomarkers for Autism Spectrum Disorder (ASD) is the high heterogeneity among ASD individuals. This encompasses core ASD symptoms (1), cognitive skills (2), comorbid psychiatric/medical conditions (3), brain atypicalities (4,5), and genetics (6). Consequently, there have been significant efforts in defining ASD subtypes.…”

Commentary: Critical considerations for studying low‐functioning autism

“…In an era of big data, data-driven techniques have a certain allure. Multivariate techniques, such as clustering, show promise when dealing with the sizeable heterogeneity seen in ASD (Lombardo et al, 2016;Romano et al, 2014). These studies usually need considerably larger samples than the ones currently used in most neuroimaging studies of ASD.…”