Unique longitudinal relationships between symptoms of psychopathology in youth: A cross‐lagged panel network analysis in the ABCD study

Abstract

Background The network theory suggests that psychopathology may reflect causal relationships between individual symptoms. Several studies have examined cross‐sectional relationships between individual symptoms in youth. However, these studies cannot address the directionality of the temporal relationships hypothesized by the network theory. Therefore, we estimated the longitudinal relationships between individual internalizing, externalizing, and attention symptoms in youth. Methods Data from 4,093 youth participants in the Adolescent Brain Cognitive Development (ABCD) study were used. Symptoms were assessed using the Brief Problem Monitor, which was administered at three time points spaced six months apart. Unique longitudinal relationships between symptoms at T1 and T2 were estimated using cross‐lagged panel network modeling. Network replicability was assessed by comparing this network to an identically estimated replication network of symptoms at T2 predicting symptoms at T3. Results After controlling for all other symptoms and demographic covariates, depressed mood, inattention, and worry at T1 were most predictive of other symptoms at T2. In contrast, threats of violence and destructiveness at T2 were most prospectively predicted by other symptoms at T1. The reciprocal associations between depressed mood and worthlessness were among the strongest bivariate relationships in the network. Comparisons between the original network and the replication network (correlation between edge lists = .61; individual edge replicability = 64%–84%) suggested moderate replicability. Conclusions Although causal inferences are precluded by the observational design and methodological considerations, these findings demonstrate the directionality of relationships between individual symptoms in youth and highlight depressed mood, inattention, and worry as potential influencers of other symptoms.

Commentary: Extending longitudinal network approaches – a reflection on Funkhouser et al. (2020)

“…The application of longitudinal networks to individuals could also facilitate recommendations for therapy. For example, in Funkhouser et al (2020), symptoms of depressed mood , inattention , and worry were predictive of symptoms six months later, with this finding also being moderately replicated at a subsequent wave (one year out). Thus, these symptoms might be worthy of targeted intervention efforts, especially if replicated in other samples.…”

“…Such ‘personalized network modeling’ techniques have just recently been developed (Epskamp, 2020) and have the potential for facilitating theoretically and clinically useful insights into the nature of psychopathology at the individual level of analysis. These techniques, while being similar to those used in Funkhouser et al, (2020), require intensive time‐series data from a single participant (rather than a few time points from several participants). To assess within‐individual patterns of covariation over time, each symptom is regressed onto all symptoms including itself at a previous time point (often the previous measurement occasion).…”

“…These limitations have also generally extended to the identification and conceptualization of highly connected (i.e., central) symptoms within networks, despite this identification being one of the key findings across network studies. That is, differentiating whether a central symptom ‘causes other symptoms or is caused by other symptoms’ has not been possible using cross‐sectional networks (Funkhouser et al, 2020).…”

“…The article by Funkhouser et al (2020) sought to address this limitation by utilizing cross‐lagged panel network (CLPN) modeling techniques, with the resulting networks accommodating an exploration of directional relations among symptoms using longitudinal panel data. Further, examination of symptoms’ centrality within the CLPN framework included a distinction between central symptoms that prospectively predicted several other symptoms (or strongly predicted a few symptoms) and those that were predicted by several other symptoms (or were strongly predicted by a few symptoms).…”

Evaluating the Criterion Validity of Hierarchical Psychopathology Dimensions Across Models: Familial Aggregation and Associations with Research Domain Criteria (RDoC) (Sub)constructs

“…The generalizability of the bifactor criterion validity results to the higher-order model may therefore be weaker for spectra than for the general factor. Additionally, comorbidity may be explained by direct causal relationships between symptoms (Borsboom, 2017), and research examining HiTOP dimensions from this perspective (Funkhouser, Chacko, et al, 2020;McElroy et al, 2018) may be beneficial.…”