Improved tools for breast cancer treatment decisions

Prediction models are tools that provide clinicians and patients with estimates of for example the risk of developing a breast cancer recurrence with and without chemotherapy. These tools combine large amounts of scientific evidence to calculate these estimates. Using these tools can help clinicians and patients make better informed decisions and to tailor care better to individual patients’ risk profile. However, for prediction models to be of added value to clinical practice in the long-term, models need to be 1) updated regularly to keep up with scientific advances, 2) evaluated thoroughly to make sure that their estimates are reliable enough for them to be used in clinical practice, and 3) embedded in daily clinical practice. To achieve these goals, three key issues urgently need to be addressed.

First, current prediction models only incorporate clinical prognostic factors (e.g., the tumor size and presence of metastases), whilst in clinical practice, genomic information (e.g., Mammaprint) is also increasingly being considered. However, how can new factors best be incorporated into an existing prediction model? Second, rigorous evaluation is needed prior to the implementation of prediction models in clinical practice. Yet, when is a model “good enough” to support medical decisions? When are additional expensive and/or invasive additional tests of added value? Third, prediction model development and evaluation are time-consuming and require large datasets with long-term follow-up. Novel IT solutions are needed to allow available data sources (e.g., hospitals’ electronic patient record systems) to be utilised safely to support long-term integration of valid prediction models in clinical practice.

In this project, solutions for these three key issues will be developed, using the extensively validated PREDICT v2.1 prediction model for early-stage breast cancer as a ‘proof of concept’. The results of this work will lead to better embedding of good prediction models in the clinical workflow.