The accuracy of early weight gain in predicting treatment outcome in a large outpatient sample of patients with anorexia nervosa.

Abstract

Previously, we evaluated early weight gain as a predictor of weight restoration for patients with anorexia nervosa using receiver operating characteristic (ROC) analysis. Models had low performance, and high rates of misclassification. Regression models including percent target weight at admission in addition to early weight gain performed better. This study evaluated the performance of early weight gain as a predictor of remission for patients with AN. We also explore the limitations of ROC analysis and show that the analogous logistic regression models outperform their ROC counterparts. Participants (N = 233) were patients with AN who received virtual outpatient FBT. ROC analyses used early weight gain to predict remission in week 20. Weight gain at week 8 performed best (AUC = 0.65 [0.58-0.72]). The optimal cutpoint was 8.9 pounds; 36% of the patients were misclassified. A regression model, which included percent target weight at admission in addition to early weight gain as a predictor variable, outperformed the ROC and returned the probability that a patient will remit. These data suggest that using early weight gain alone to set cutpoints misclassifies many patients with AN. Accounting for starting weight at admission improves model predictions.