Impact of an Artificial Intelligence Management Model on Psychological and Behavioral Outcomes in Breast Cancer Patients Receiving Home-based CDK4/6i Therapy: A Multicenter Randomized Controlled Trial

Abstract Background: Endocrine therapy combined with CDK4/6 inhibitors (CDK4/6i) is the preferred regimen for first-line and adjuvant treatment of HR+/HER2 − breast cancer. Oral administration has shifted the primary setting of care to home environments. However, traditional care models rely heavily on episodic outpatient visits, failing to provide proactive, continuous, and personalized transitional care. This creates a "service gap," leading to psychological distress, compromised compliance, and delayed toxicity management during long-term out-of-hospital therapy. We aimed to evaluate the effects of an artificial intelligence (AI)-driven management model and a customized active follow-up model on psychological and behavioral outcomes in these patients. Methods: This multicenter randomized controlled trial dynamically evaluated patients using the Self-Rating Anxiety Scale (SAS), Self-Rating Depression Scale (SDS), EORTC QLQ-C30, and the Patient-Reported Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE). Findings: Between 2024 and 2025, 189 patients receiving CDK4/6i across multiple hospitals in Tianjin were randomized. Critically, 161 patients completed the final analysis (AI-driven management: \(\:n=53\); customized active follow-up: \(\:n=54\); standard care: \(\:n=54\)), with balanced baseline characteristics (\(\:P>0.05\)). For the primary outcome, both the AI-driven and customized follow-up groups demonstrated significantly superior efficacy in maintaining anxiety stability and alleviating depression compared with standard care (both \(\:P\le\:0.001\)). In the standard care group, psychological states deteriorated significantly over time, with depression scores peaking at week 16 (T2). For secondary outcomes, both intervention groups showed significantly better cognitive, emotional, and social functioning, as well as global health status, than the standard care group (all \(\:P<0.05\)). Multiple correspondence analysis revealed a robust positive intra-family covariation of negative emotions within patient-caregiver dyads, closely clustering with a fixed caregiver and comorbidities. SAS and SDS scores were positively correlated with fasting blood glucose and liver function indicators, notably alanine aminotransferase (ALT) (all \(\:P<0.05\)). Furthermore, real-time AI monitoring yielded a higher PRO-CTCAE adverse event capture rate, mitigating the risk of high-grade toxicities. Interpretation: Both AI-driven and customized follow-up models significantly improve anxiety, depression, and quality of life in breast cancer patients on home-based CDK4/6i therapy. The AI model offers distinct scaling advantages in cognitive support and real-time toxicity monitoring, providing robust evidence for a holistic "bio-psycho-social" precision oncology home-management paradigm. Clinical Trial Registration NCT07409298||https//www.clinicaltrials.gov/(Registration date February 12, 2026 First submitted January 20, 2026)