JMIR Publications

ABSTRACT

Background:

Health professionals are often faced with the need to identify women at risk of manifesting poor psychological resilience following diagnosis and treatment for breast cancer. A crucial first step toward designing personalized psychological interventions to promote psychological well-being in women with breast cancer is identifying risk factors for poor resilience for a given patient.

Objective:

To present the rationale, implementation and cross-validation of Machine Learning (ML) models designed to identify the most important predictors of overall mental health and global Quality of Life in breast cancer survivors.

Methods:

A set of 12 models were implemented based on longitudinal data from a prospective multi-centre clinical pilot at five major oncology centres in four countries (Italy, Finland, Israel, and Portugal) enrolling 706 patients with highly treatable breast cancer, who were followed to 18 months post-diagnosis (BOUNCE project). An extensive set of demographic, life-style, clinical, psychological, and biological variables measured within 3 months after diagnosis served as predictors. Rigorous feature selection isolated key psychological resilience outcomes that could be incorporated into future clinical practice.

Results:

Balanced Random Forest classifiers were successful at predicting wellbeing outcomes with accuracies ranging between 78-82% (for 12-month post-diagnosis endpoints) and between 74-83% (for 18-month post-diagnosis endpoints). Explainability and interpretability analyses, built on the best performing models, were used to identify potentially modifiable psychological and lifestyle characteristics which, if addressed systematically in the context of personalized psychological interventions, would be most likely to promote resilience for a given patient. Further, we describe the implementation of a Decision Support Tool (DST) that incorporates these models to guide health professionals to (i) identify women who are at risk for adverse wellbeing outcomes following breast cancer treatment and (ii) plan customized psychological interventions for women at risk based on an extensive set of personalized clinical recommendations.

Conclusions:

Results highlight the clinical utility of the BOUNCE supervised modeling approach by focusing on resilience predictors that can be readily available to practicing clinicians at major oncology centers. The BOUNCE DST, paves the way for personalized risk assessment methods toward identifying patients at high risk for adverse well-being outcomes and directing valuable resources for those most in need for specialized psychological interventions