Statistical and/or computational models focusing on the well-being and functionality of breast cancer patients are limited. Modelling of psychological resilience requires building theoretically plausible, clinically useful, and computationally sound schemes describing: (i) the predominant mechanisms involved in the process of psychological adaptation to cancer and (ii) the most powerful longitudinal predictors of long-term psychosocial and functional outcomes following treatment for breast cancer. This goal could be effectively addressed using conventional multivariate statistical methods, such as structural equation modelling using latent constructs, logistic regression, and survival statistics, to name the most popular methods. Although these techniques could be adopted for modelling resilience over time as individual trajectories, they could not predict adaptation to illness as a dynamic process through a composite framework wherein the contribution of each trajectory to the end-point outcomes is assessed and evaluated.
BOUNCE aspires to go further and develop a prediction tool that can be used at any point during breast cancer diagnosis and treatment to identify patients at risk for poor psychosocial and functional outcomes—i.e., patients who, at a given point in time, demonstrate poor psychological resilience. In its final form, this tool should have the capacity to identify subgroups of persons defined on individual resilience levels (as a proxy for risk of adverse psychosocial outcome) using a limited number of validated predictors and moderators. The novelty of the BOUNCE computational approach is two-fold:
First, it takes full advantage of longitudinal measurements of potential predictors to test models that include both one-time measurements of each predictor (cross-sectional predictor models) as well as individual trajectories of each predictor.
Secondly, given the inherent complexity of the longitudinal data, BOUNCE will develop and evaluate a Machine Learning (ML) framework to identify subgroups of patients that display distinct psychosocial profiles (at specific time points and over time) in adapting to breast cancer.
Cross-sectional and longitudinal data are exploited by unsupervised and supervised machine learning techniques aiming at identifying patterns of patients’ symptoms and at predicting final and intermediate outcomes at each and across different time points, respectively. A model fusion computational framework is developed to enhance the predictive outcomes of the models. The BOUNCE trajectory predictor will exploit effectively factors measured in the multicenter clinical study. This set of factors consists of: (i) patient-reported outcomes (i.e. mental health, distress level, health- and overall Quality of Life (QoL), and functionality), (ii) illness-related self-regulation variables (i.e. self-rated health etc.), (iii) potentially stressful events taking place during the follow up period, (iv) moderators and facilitators (i.e. self-efficacy, resilience, social support etc.) and (v) lifestyle factors (i.e. health habits etc.).
The BOUNCE computational approach evolves along four main axes each serving crucial clinical scenarios.
Firstly, a cross-sectional clustering methodology aims to determine basic clusters of patients that at a given time point belong to a specific ‘level’ of adaptation to illness. In this approach, resilience is defined according to the observation of affective and functional status.
Secondly, longitudinal data are exploited through a clustering methodology aiming to distinguish patient profiles according to possible transitions from one resilience category to another due to changes in specific factors.
Thirdly, prediction of resilience is performed to determine factors or interactions among them that can more accurately predict final and intermediate outcomes as well as an overall resilience level.
Finally, all clinical predictive outcomes will be entered into a decision-level fusion model to investigate whether the ensemble of the decisions further improves prediction of resilience at a specific time point.
By Prof. P. Simos and E. Karademas, University of Crete and FORTH, Herakleion, Greece
BOUNCE aspires to develop a prediction tool that can be used at any point during the course of breast cancer diagnosis and treatment to identify patients at risk for poor psychosocial and functional outcomes (i.e., poor psychological resilience).
To this end, BOUNCE will develop and evaluate a Machine Learning (ML) framework to identify subgroups of patients that display distinct psychosocial profiles (at specific time points and over time) in adapting to breast cancer.