4 - (SYM 28) Using Control Theory to Examine Individual Trajectories of Eating Disorder Symptom Remittance

Eating disorders (EDs) are serious psychiatric illnesses that are notoriously difficult to treat. Data-driven approaches to personalizing ED treatment may be one way to improve treatment outcomes. Preliminary research using network modeling suggests that targeting individuals’ ‘most central symptoms’ (i.e., symptoms with high strength centrality) reduced individual ED symptoms (Levinson et al., 2023). While these results are encouraging for future personalized treatments, it is possible other centrality indices, such as controllability centrality and out-strength, may also offer clinical utility (Henry et al., 2022). In this study, we use three network-informed symptoms (i.e., highest strength-centrality, average controllability, and out-strength) to simulate individual intervention effects among a sample with EDs. 

Method: One hundred individuals with EDs completed ecological momentary assessment surveys 4 times/day for 25 days (100 total time points/person). We estimated a multilevel integrated vector autoregressive model to obtain a network of participant symptom dynamics and then calculated the average controllability centrality for each symptom. Using control theory, we simulated a 30-session intervention that targeted the symptom with the highest strength centrality (fear of weight gain), average controllability (thinking one had over-eaten), and out-strength (thinking about dieting). The treatment simulations were applied to each participant’s network. We calculated the number of days it took reduce the target symptom to a near-zero value for each participant.

Results: Most individuals’ symptoms remitted around session 15. Thinking one had overeaten was reduced to near-zero between sessions 14-17 (M=15.18, SD=0.85), thinking about dieting between sessions 11-21 (M=14.98, SD=1.39), and fear of weight gain between sessions 13-29 (M=15.05, SD=1.18).

Conclusions: Results from the current study show that participants displayed similar shapes in treatment response across symptoms targeted. Further, while there was some variability in precisely when symptoms remitted for a given participant, most symptoms remitted for most participants about halfway through the total length of the theoretical intervention. Results should be interpreted cautiously, as these models reflected theoretical idiographic trajectories rather than actual trajectories from treatment delivered. Future work is needed to compare these results to empirical findings to optimize network-informed personalized treatment for EDs.