In WP1 the value chain for the project was defined, different stakeholders identified and interviewed, the user requirements and scenarios were defined.
In WP2 the definition and assessment of 1. resilience in women with breast cancer, 2. resilience as a dynamic process critically involved in effective illness adaptation and recovery, 3. multi-level factors potentially affecting resilience, 4. multi-scale factors related to the evolution of resilience, in relation to both constant and time-varying patient characteristics, 5. changes in social and other life circumstances was determined.
WP3 delivered the Data Cleaner (front-end and back-end), introducing several optimizations based on statistical methods, which are now applied on all the collected prospective data. All data collected were successfully ingested and provided as input to WP4/WP5, after being processed with the corresponding tools for data cleaning, recoding, identification of missing data, assessment of missing data patterns, integration and homogenization.
In WP4 both unsupervised and supervised learning analysis pipelines have been formulated and implemented in order to model resilience as a process and resilience as outcome over different intervals of the first 18 months post-diagnosis. In order to address resilience as a process, unsupervised clustering schemes and latent-class mixed-effects regression analysis to identify subgroups of patients who display distinct profiles of change in mental health symptoms, HADS depression, HADS anxiety and global QoL, over the first 12 and 18 months post diagnosis have been developed and tested. With respect to resilience as outcome, generalizable models for optimal prediction of 12- and 18-month patient outcomes (in terms of mental health symptoms, HADS depression, HADS anxiety and global QoL) by aggregating all available patient information from the early phase of illness (i.e. M0 and M3 measurement waves) have been developed and tested. Both pooled data across sites and generalizability analyses have been conducted to establish a robust framework that could be exploited to other clinically relevant problems. The present work extends the explainability domain by adopting model-agnostic analysis in the models developed. Accuracy evaluation, and quality assurance of the BOUNCE In Silico Resilience Trajectory Predictor have been performed.
In WP5, the final release of the integrated BOUNCE software platform has been delivered. It was redesigned, focusing on its frontend part, based on the suggestions from clinical partners in order to increase its usability. The final version of all components was successfully integrated, while enhancements and optimizations were added. It also incorporates predictive models developed in WP4.
In WP6 prospective data on clinical, psychological, functional and quality of life factors were collected at baseline in Finland, Italy, Israel, Portugal for 746 breast cancer patients. Follow-up data collection was performed on months 3, 6, 9, 15, 18. Collected data have been provided to the technical partners for modelling. Following the analyses on prospective data, two pilot studies have been designed in HUS and IEO to test the efficacy of the preliminary resilience prediction model.
In WP7 additional validation has been conducted via an experiment. A cost-benefit model has been constructed were the cost and the benefit parameters have been defined and the data sources outlined, and the preliminary cost-benefit estimated.
In WP8 the work focused on dissemination and exploitation. Stakeholders have been engaged via conference presentations, workshops, individual interviews, and focus group interviews. The consortium has discussed and identified ownership interests, identified and evaluated different business models’ scenarios for the BOUNCE assets, and a preliminary marked analysis has been conducted.
In WP9 issues related to ethics, risk management and administration have been handled.
