Data mining identifies meaningful severity specifiers for anorexia nervosa

The fifth edition of the Diagnostic and Statistical Manual of Mental Disorders defines anorexia nervosa (AN) severity based on body mass index (BMI). However, BMI categories do not reliably differentiate the intensity of AN and comorbid symptoms. Shape/weight overvaluation has been proposed as an alternative severity specifier. The present study used structural equation model (SEM) Trees to empirically determine specific levels of BMI and/or shape/weight overvaluation that differentiate AN severity. We also compared whether the SEM Tree-derived severity groups outperformed existing AN severity definitions. Participants were 1,666 adolescents and adults with AN who were receiving eating disorder treatment at one of the three levels of care (outpatient, partial hospital program, or residential). Participants completed self-reported questionnaires assessing eating pathology and comorbid symptoms. SEM Tree analyses first specified an outcome model of AN severity, and then recursively partitioned the outcome model into subgroups based on all values of BMI and shape/weight overvaluation. One-way analyses of variance and t tests determined which severity definition explained the most variance in clinical characteristics. SEM Tree analyses yielded five severity groups, all of which were defined based on increasing intensities of shape/weight overvaluation: < 2.25, 2.25-3.24, 3.25-4.24, 4.25-5.24, and ≥ 5.25. No groups were defined based on BMI. SEM Tree-derived groupings explained more variance in clinical characteristics than existing severity definitions. Taken together, shape/weight overvaluation appears to be a more accurate marker of AN severity than BMI. The empirically determined AN severity scheme accounted for the most variance in clinical characteristics. Future research should assess the predictive value of these empirically defined AN severity indicators. (PsycInfo Database Record (c) 2025 APA, all rights reserved).