Training and validation of a model predicting clusters of long-term cancer-related behavioral symptoms (CRBS) after breast cancer (BC)

12087 Background: Fatigue, cognitive impairment, emotional distress, and sleep disturbance are prevalent and challenging symptoms among patients (pts) with BC. These symptoms often tend to cluster, they are usually responsive to similar behavioral interventions (e.g., exercise, psychosocial), and can be collectively defined as CRBS. We aimed to generate a model to predict clusters of long-term CRBS. Methods: CANcer TOxicity (CANTO; NCT01993498) is a prospective cohort of pts with stage I-III BC collecting longitudinal data at diagnosis (dx) and yearly post-dx. Our outcome was the proportion of pts reporting a cluster of ≥3 severe CRBS (EORTC QLQ-C30/HADS) 4 years post-dx. Clinical, behavioral, treatment (tx)-related predictors, and the baseline Behavioral Symptoms Score (BSS; based on the number of severe CRBS as proxy of symptom burden at dx) were tested in a training cohort (Mar 2012 - Feb 2015) with a multivariable logistic regression model implementing bootstrapped Augmented Backwards Elimination and 10-fold internal cross-validation. Performance was externally validated in pts from a subsequent enrolment period (Mar 2015 - Apr 2018). Results: In the training cohort (N=3555), 92.1% of pts had stage I-II BC, 52.6% received chemo, and 81.9% endocrine tx. At dx, 22.0% pts had severe fatigue (C30≥40/100), 29.7% severe cognitive impairment (C30<75/100), 33.9% and 6.5% clinically suggestive anxiety and depression (HADS≥11/21), respectively, and 35.2% severe insomnia (C30>50/100). 19.0% of pts reported a severe CRBS cluster at dx, then 21.0% did so 4 years post-dx. Younger age, previous psychiatric disorders, and higher baseline BSS were predictors of CRBS clusters (Table;AUC 0.73 [95%CI 0.71-0.75]). Results were confirmed in the validation cohort (N=1533). Conclusions: 1 in 5 pts reports a cluster of severe CRBS 4 years post-dx of BC. Younger pts with previous psychiatric disturbance and a higher symptom burden at dx have greater risk of long-term CRBS. Systematic clinical implementation of predictive tools at dx of BC can aid a more thorough screening and referrals to targeted behavioral interventions. Clinical trial information: NCT01993498 . [Table: see text]