Using person‐specific neural networks to characterize heterogeneity in eating disorders: Illustrative links between emotional eating and ovarian hormones

• Published in: The International journal of eating disorders Vol. 51; no. 7; pp. 730 - 740
• Main Authors: Beltz, Adriene M., Moser, Jason S., Zhu, David C., Burt, S. Alexandra, Klump, Kelly L.
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.07.2018; Wiley Subscription Services, Inc
• Subjects: Adult; Brain - physiopathology; Brain Mapping; connectivity; Eating - psychology; Eating disorders; emotional eating; Emotions; Estradiol; estrogen; Feeding and Eating Disorders - physiopathology; Feeding and Eating Disorders - psychology; Feeding Behavior - psychology; Female; heterogeneity; Hormones; Humans; Magnetic Resonance Imaging; Medical imaging; Menstrual Cycle - psychology; Models, Statistical; Neural networks; Neural Networks, Computer; person‐specific; precision healthcare; Progesterone; resting state; Saliva; Surveys and Questionnaires; Twin studies; twin study; Twins; estrogen; resting state; connectivity; twin study; emotional eating; progesterone; person-specific; precision healthcare; heterogeneity
• ISSN: 0276-3478; 1098-108X; 1098-108X
• DOI: 10.1002/eat.22902

Objective Emotional eating has been linked to ovarian hormone functioning, but no studies to‐date have considered the role of brain function. This knowledge gap may stem from methodological challenges: Data are heterogeneous, violating assumptions of homogeneity made by between‐subjects analyses. The primary aim of this paper is to describe an innovative within‐subjects analysis that models heterogeneity and has potential for filling knowledge gaps in eating disorder research. We illustrate its utility in an application to pilot neuroimaging, hormone, and emotional eating data across the menstrual cycle. Method Group iterative multiple model estimation (GIMME) is a person‐specific network approach for estimating sample‐, subgroup‐, and individual‐level connections between brain regions. To illustrate its potential for eating disorder research, we apply it to pilot data from 10 female twins (N = 5 pairs) discordant for emotional eating and/or anxiety, who provided two resting state fMRI scans and hormone assays. We then demonstrate how the multimodal data can be linked in multilevel models. Results GIMME generated person‐specific neural networks that contained connections common across the sample, shared between co‐twins, and unique to individuals. Illustrative analyses revealed positive relations between hormones and default mode connectivity strength for control twins, but no relations for their co‐twins who engage in emotional eating or who had anxiety. Discussion This paper showcases the value of person‐specific neuroimaging network analysis and its multimodal associations in the study of heterogeneous biopsychosocial phenomena, such as eating behavior.