A hierarchical clustering approach to dissect behavioral symptoms in early-stage breast cancer (BC).

12036 Background: Fatigue, cognitive impairment, insomnia, anxiety, and depression are cancer-related behavioral symptoms that frequently co-occur and share underlying risk factors. We investigated the clustering of these symptoms at different time points, aiming to identify drivers of symptom segregation. Methods: Patients with stage I-III BC from CANTO (NCT01993498) were included. Hierarchical cluster analysis was performed at three time points: baseline (BC diagnosis), year (Y) 1 (3-6 months post-surgery, chemotherapy [CT], and/or radiotherapy), and Y2. Clustering used the Ward method via R 2 and Pseudo T 2 statistics, incorporating dichotomized self-reported symptoms based on clinically meaningful thresholds: fatigue (EORTC QLQ-C30 ≥40/100), cognitive impairment (<75/100), insomnia (>50/100), and anxiety or depression (HADS ≥11/21). Results: Among 6,486 patients, the mean age was 56.2 years; 90% had stage I-II BC, 53% received CT, and 83% endocrine therapy (ET). Analysis of dendrograms identified six distinct clusters (CL) consistently across time points. There was substantial difference in symptomatology between baseline/Y1 and baseline/Y2 (Cramer's V 0.23, 0.22), while there was greater symptom overlap post-treatment between Y1 and Y2 (V 0.32). No specific behavioral symptom drove hierarchical segregation at baseline. However, by Y1, depression and fatigue emerged as primary drivers: CL1 (38%) comprised patients with low symptom scores, similarly to other time points; CL2 (14%) was mostly characterized by cognitive dysfunction and anxiety; CL3 (14%) was defined by patients with clinically meaningful insomnia but without fatigue, while CL5 and CL6 included patients with fatigue but no emotional distress, further segregated by the presence (CL5, 12%) or absence (CL6, 10%) of cognitive dysfunction. Patients reporting depression were consistently grouped in CL4. Notably, CL4 (12%) was heterogeneous and characterized by multi-symptomatology (Table). The use of CT and ET was significantly associated with membership in CL4 and CL5 at Y1 (p < .001). At Y2, the clustering remained consistent with what observed at Y1, however the prevalence of individual symptoms was higher in the multi-symptom clusters, including CL4 where all patients with depression were exclusively segregated. Conclusions: Hierarchical clustering revealed dynamic changes over time, potentially reflecting the impact of the acute treatment phase on interrelationships among symptoms. Depression and fatigue emerged as key drivers of segregation. Accounting for variability in symptom clustering can enable better targeting of therapeutic options. Clinical trial information: NCT01993498 . Proportions of patients with clinically meaningful symptoms by CL at Y1 (may not add up to 100% due to multiple symptoms). Fatigue Cognitive dysfunction Insomnia Anxiety Depression CL1 0 0 0 0 0 CL2 16 75 0 47 0 CL3 0 34 100 20 0 CL4 91 74 85 82 51 CL5 100 100 57 0 0 CL6 100 0 48 0 0