Phenoscreening: A Developmental Approach to Rdoc-Motivated Sampling

Background: The Research Domain Criteria (RDoC) initiative has formalized research strategies for parsing the heterogeneity/variability inherent to the etiology, phenotypic presentation, and treatment response of major psychiatric disorders, but has not effectively integrated developmental considerations. Identifying multi-dimensionally-determined profiles of risk via data driven computational approaches represents one potential avenue to 1) improve early identification of at-risk phenotypes and 2) select unbiased samples for various research questions.

Objectives: The primary objective of this research is to derive multiple “high-risk” phenotypes, while remaining agnostic to traditional DSM categories, by leveraging multiple dimensional constructs and a data-driven computational strategy during a developmentally sensitive period. A secondary aim is to verify the prognostic utility of the risk profiling approach.

Methods: Parents of 17-25 month-old toddlers (n = 1570), drawn from a community-based sample, completed the MCDI, the Video-Referenced Rating Scale for Reciprocal Social Behavior (vrRSB; Marrus et al., 2015), and the Repetitive Behavioral Scales for Early Childhood (RBS-EC; Wolff, Boyd, & Elison, 2016). To identify discrete developmental phenotypes, we used factor mixture modeling (FMM), a statistical method for parsing population heterogeneity that identifies groups of empirically testable entities (i.e., latent classes). Statistically, FMM mixes factor analysis, used to estimate unobserved continuous variables, with latent class analysis, used to estimate latent categorical groups. Conceptually, FMM is a person-centered statistical approach that focuses on similarities and differences among profiles to identify homogenous subgroups of individuals, with each subgroup possessing a unique set of characteristics that differentiates it from other subgroups. To validate the predictive utility of these risk profiles, a subsample of toddlers (n = 107) was assessed on a distal, independent outcome (the Infant Toddler Social Emotional Assessment; ITSEA) at an average of 10 months after the initial assessment.

Results: FMM results, based in part on 2 factors derived from seven dimensional manifest variables, identified five asymmetrically sized subgroups (ranging from n = 4 to n = 1230). One percent of the sample (17/1570), captured by two subgroups, was characterized by increased repetitive behaviors and social communicative impairments. An additional 5% of the sample fell into a nominally moderate risk group (79/1570), whereas the remaining toddlers fell into 2 nominally low-risk clusters. Follow-up analyses on a subsample of 107 infants confirmed the predictive validity of the risk profiles, showing significant differences between high-, moderate-, and low-risk groups on internalizing, externalizing, and dysfunctional behavior as assessed by the ITSEA. Comparison of high- and low-risk groups revealed large effect sizes for internalizing (d = 1.39), externalizing (d = 0.83), and dysregulation (d = 1.87).

Conclusions: A data-driven computational approach yielded 5 homogenous subgroups of n = 1570 community-ascertained toddlers, the clinical utility of which was corroborated by outcomes measured longitudinally. Data-driven approaches, leveraging multiple developmentally appropriate dimensional/quantitative constructs holds promise for future efforts aimed toward early identification of at-risk-phenotypes. Further, we expect this method to inform a more personalized approach to clinical recommendation/intervention, as compared to the binary approach of traditional screening/sampling schemes.