Development and Validation of a Prediction Model for Co-Occurring Moderate-to-Severe Anxiety Symptoms in First-Episode and Drug Naïve Patients With Major Depressive Disorder

Abstract

Background: Moderate-to-severe anxiety symptoms are severe and common in patients with major depressive disorder (MDD) and have a significant impact on MDD patients and their families. The main objective of this study was to develop a risk prediction model for moderate-to-severe anxiety in MDD patients to make the detection more accurate and effective.

Methods: We conducted a cross-sectional survey and tested biochemical indicators in 1718 first-episode and drug naïve (FEDN) patients with MDD. Using machine learning, we developed a risk prediction model for moderate-to-severe anxiety in these FEDN patients with MDD.

Results: Four predictors were identified from a total of 21 variables studied by least absolute shrinkage and selection operator (LASSO) regression analysis, namely psychotic symptoms, suicide attempts, thyroid stimulating hormone (TSH), and Hamilton Depression Scale (HAMD) total score. The model built from the four predictors showed good predictive power, with an area under the receiver operating characteristic (ROC) curve of 0.903 for the training set and 0.896 for the validation set. The decision curve analysis (DCA) curve indicated that the nomogram could be applied to clinical practice if the risk thresholds were between 13% and 40%. In the external validation, the risk threshold was between 14% and 40%.

Conclusion: The inclusion of psychotic symptoms, suicide attempts, TSH, and HAMD in the risk nomogram may improve its utility in identifying patients with MDD at risk of moderate-to-severe anxiety. It may be helpful in clinical decision-making or for conferring with patients, especially in risk-based interventions.