Predicting Depression, Anxiety, and Their Comorbidity among Patients with Breast Cancer in China Using Machine Learning: A Multisite Cross-Sectional Study

Abstract

Depression and anxiety are highly prevalent among patients with breast cancer. We tested the capacity of personal resources (psychological resilience, social support, and process of recovery) for predicting depression, anxiety, and comorbid depression and anxiety (CDA) among such patients using machine learning (ML). We conducted a cross-sectional survey in Liaoning Province, China, including questions about demographics, COVID-19^′s impact, and personal resources (707 valid responses). In the training set, we used Lasso logistic regression to establish personal resource models. Subsequently, we used six ML methods and a tenfold cross-validation strategy to establish models combining personal resources, demographics, and COVID-19 impacts. Findings indicate that in total, 21.9%, 35.1%, and 14.7% of participants showed depression, anxiety, and CDA, respectively. Loneliness, vitality, mental health, bodily pain, and self-control predicted depression, anxiety, and CDA. Furthermore, general health predicted depression, and physical function predicted anxiety. Demographic and COVID-19 models were far less predictive than personal resource models (0.505–0.629 vs. 0.826–0.869). Among combined models, the support vector machine model achieved the best prediction (AUC: 0.832–0.873), which was slightly better than the personal resource models. Personal resources features with ML and personal resources can help predict depression, anxiety, and CDA in patients with breast cancer. Accordingly, interventions should target loneliness, bodily pain, vitality, mental health, and self-control.