3 - (IOP 35) Estimating Longitudinal Eating Disorder Symptom Dimensions Across Time and Within Individuals to Inform Just-in-time Interventions

Restrictive eating disorders (EDs: anorexia nervosa [AN] and atypical AN [AAN]) are severe mental illnesses with high individual heterogeneity, even within diagnostic groups. Despite this heterogeneity, many research and clinical interpretations are based on group-level (e.g., on average) findings. Exploratory graph analysis (EGA), a network science method that identifies symptoms that cluster together in a network (i.e., dimensions), is typically estimated from cross-sectional and group-level data. While this work suggests that ED symptoms form distinct clusters, this research overlooks patterns of symptom evolution over time and within individuals. As a result, clinical conceptualizations of EDs fail to account for how symptoms cluster across time, which can inform just-in-time personalized interventions. Dynamic EGA can model how symptoms co-occur over time (i.e., identify which symptoms have similar rates of change) on the group and individual levels. We applied dynamic EGA to identify clusters of symptoms that change together over time within a sample with restrictive EDs. Ecological momentary assessment (EMA) data from participants diagnosed with AN (N = 12) or AAN (N = 8) who completed >50% EMA four times daily for 30 days were used. We used generalized local linear approximation to derive rate of change over time of 25 ED-related symptoms (e.g., restriction, exercise) and used EGA to identify clusters of co-evolving symptoms on the group (AN, AAN, and combined) and individual levels. In a group-level combined network, we identified 3 dimensions: food and eating related (e.g., eating small portions, food rules), body related (e.g., feeling fat, body checking), and affective (e.g., avoiding emotions, difficulty relaxing), with slight diagnostic differences. The AN network had three dimensions, and the AAN had four dimensions. The number of individual-level network dimensions ranged between 2 and 8. Bootstrapped Ergodicity Information Index (EII) tests showed that the longitudinal network dimensionality was better explained by multiple individual networks rather than a single group-level network for the combined sample (EII = 2.824, mean EII = 1.69, SD = 0.001, p = 0.341) and the AN group (p = 0.85). However, the AAN group network sufficiently represented the individuals in the group (p = 0.04). These findings suggest heterogeneity in the dynamic structural organization of ED symptoms between AN and AAN and between individuals. Consequently, group-level models of ED symptom dimensionality may not adequately capture person-level symptom co-evolution in restrictive EDs. These findings emphasize the need for personalized treatment approaches in EDs, as how symptoms evolve vary between individuals. Future research should explore how individualized symptom trajectories can inform precision medicine approaches.