Comprehensive Summary
In this study, Sandoval-Araujo et al. investigate the use of machine learning (ML) to differentiate between anorexia nervosa (AN) and atypical anorexia nervosa (atypical AN). Both disorders involve restrictive eating and body image concerns, but differ in that AN is associated with a BMI under 18.5 kg/m2, while atypical AN is not. The study included 448 participants diagnosed with either AN or atypical AN and utilized clinical questionnaires to assess symptomatology. Various ML algorithms, including logistic regression, decision trees, and random forests, were employed. Models were tested with and without BMI data. Results showed that models including BMI had higher accuracy (mean accuracy of 0.74) compared to those without (mean accuracy of 0.59). Decision trees performed best with a mean accuracy and AUC of 0.81. The findings suggest that BMI is crucial for differentiating between the two conditions, supporting the hypothesis that AN and atypical AN may represent a single condition. This challenges the current classification system and highlights the need for inclusive treatment approaches.
Outcomes and Implications
The research suggests that the distinction between AN and atypical AN may not be clinically relevant, as ML models could not reliably differentiate between them without BMI. This has significant implications for the field of eating disorders, emphasizing that individuals with atypical AN require the same level of care as those with AN. The study advocates for a re-evaluation of the classification system to recognize the severity of restrictive eating at any weight, potentially reducing stigma and improving care for those not considered underweight. Clinically, this research supports the idea that treatment approaches should be similar for both conditions, and highlights the potential of ML in refining psychiatric diagnostic categories.