New multivariable model predicts anorexia nervosa remission better than BMI-based severity ratings

A new study published in the Journal of Eating Disorders reports that a multivariable prognostic model predicts full remission after treatment in anorexia nervosa substantially more accurately than the DSM-5 severity classification, which relies primarily on body mass index thresholds. The work, based on 398 patients treated in a day patient program, offers a clinically interpretable algorithm for prognostic stratification built from variables routinely collected in practice.

The DSM-5 severity specifier for anorexia nervosa, which grades illness largely by BMI, has shown limited prognostic utility for predicting treatment outcomes. The research team set out to develop a severity model based instead on the risk of achieving full remission following treatment, drawing on anthropometric, clinical and treatment-history information.

The clinical sample comprised 366 women and 32 men aged between 16 and 68 years. Candidate predictors included BMI-related indicators, illness duration, eating disorder symptom severity, previous treatment history, non-suicidal self-injury and lifetime substance use. The researchers fitted an elastic-net logistic regression model and internally validated it using nested cross-validation, then compared its performance with a multilayer perceptron neural network.

The elastic-net model identified current pre-treatment BMI as the strongest predictor of remission, followed by treatment history and non-suicidal self-injury. This finding is notable because it does not discard weight as a prognostic signal; rather, it places BMI within a broader set of clinical variables that together carry far more predictive information than weight alone.

In terms of discrimination, the DSM-5 BMI-based severity classification achieved an area under the curve of 0.56, barely better than chance. The multilayer perceptron largely replicated these results, yielding comparable discrimination and only marginally improved probabilistic accuracy.

The practical difference between the two approaches was most visible in how sharply they separated patients into risk groups. The model-derived severity groups showed a markedly steeper gradient in remission rates, ranging from 43.4 percent in the lowest-severity group to 12.0 percent in the highest, compared with a flatter spread of 34.2 percent to 17.4 percent across the DSM-5 severity categories. This steeper gradient translated into significantly better prognostic stratification.

The model’s performance was not uniform across all patient groups. It showed moderate predictive performance across anorexia nervosa subtypes and age groups, with slightly higher discrimination among patients aged 20 to 24 years and among those with binge-purge symptomatology. The authors report these as modest differences rather than dramatic subgroup effects.

The study’s authors conclude that a multivariable prognostic severity model integrating anthropometric, clinical and treatment-history indicators substantially outperformed the DSM-5 BMI-based severity criterion in predicting treatment outcomes. They argue that the findings support the adoption of multidimensional severity frameworks and provide a clinically interpretable algorithm for prognostic stratification and personalized treatment planning in anorexia nervosa.

Several caveats frame these results. The model was developed and internally validated in a single clinical sample from a day patient program, and its performance figures reflect internal validation rather than independent external testing in other treatment settings. The model estimates the probability of full remission after treatment; it is a prognostic tool, not a causal explanation of recovery, and it does not establish that changing any single predictor would change outcomes.

Within those limits, the study offers a concrete demonstration that severity in anorexia nervosa can be measured in a way that tracks recovery more closely than weight-based categories. By combining pre-treatment BMI with treatment history, self-injury, symptom severity and substance use in a transparent regression model, the approach matches the performance of a more complex neural network while remaining interpretable, which the authors suggest makes it suitable for clinical use in assessment and treatment planning.

Subject of Research: A multivariable prognostic model for predicting remission in anorexia nervosa beyond BMI-based severity classification

Article Title: Redefining severity in anorexia nervosa: a multivariable prognostic model with incremental predictive validity beyond DSM-5 BMI-based classification

Article References: Granero, R., Jiménez-Murcia, S., Camacho-Barcia, L., Torao-Angosto, M., Gorwood, P., Ehrlich, S., Adan, R., Dickson, S. L., Di Lodovico, L., Sánchez, I., Tovar, S., Dieguez, C., Tolle, V., & Fernández-Aranda, F. (2026). Redefining severity in anorexia nervosa: a multivariable prognostic model with incremental predictive validity beyond DSM-5 BMI-based classification. Journal of Eating Disorders. https://doi.org/10.1186/s40337-026-01807-3

Image Credits: AI Generated

DOI: 10.1186/s40337-026-01807-3

Keywords: anorexia nervosa, DSM-5 severity, prognostic model, elastic-net regression, body mass index, remission prediction, machine learning, non-suicidal self-injury, eating disorders, day patient program, treatment outcomes, Redefining