Abstract
This paper critically examines the study by Brizzi et al., which applied explainable machine learning to predict short-term treatment outcomes in patients hospitalized for anorexia nervosa (AN). While the study presents an innovative and promising methodological framework, important conceptual and practical issues warrant further scrutiny. Chief among these is the reliance on body mass index (BMI) change as the sole proxy for treatment efficacy. This unidimensional metric, though pragmatic in acute inpatient settings, fails to capture the broader psychological and behavioral dimensions integral to AN recovery. The paper also interrogates the clinical applicability of machine learning tools, emphasizing both their potential to illuminate complex predictive patterns and the challenges they pose in terms of data sufficiency, interpretability, and real-world integration. Moreover, the identification of body uneasiness, interpersonal difficulties, and personal alienation as key predictive factors aligns with established theoretical models of AN, reinforcing the need for targeted psychotherapeutic interventions. However, further research is needed to explore how such predictors interact with specific treatment modalities and influence long-term outcomes. Overall, this paper underscores the value of integrating psychological variables into predictive modeling while cautioning against reductive interpretations of recovery in complex psychiatric disorders.