Risk stratification model for predicting risk of recurrence in human epidermal growth factor receptor 2 (HER2) positive breast cancer patients with pathological complete response after neoadjuvant therapy

Breast cancer (BC), as the most prevalent malignant tumor among women, had approximately 290,560 new BC cases with 43,780 new deaths in 2022 worldwide. In current clinical practice, the human epidermal growth factor receptor 2 (HER2), estrogen-receptor (ER) and progesterone receptor (PR) status divide BC into four subtypes and dominantly impact the therapy regimen. HER2, ER and PR are proteins that can be found on the surface of breast cancer cells. If the proteins are present, the cancer may be classified as HER-positive, ER-positive and/or PR-positive. The presence or absence of these proteins can help determine the types of therapy to which the cancer cells will respond.

HER2-positive BC disease accounts for 15-20% of all BC cases and is more aggressive, with a worse prognosis, than HER2-negative BC disease. Previous studies, which constructed models to predict the prognosis of HER2-positive BC tumors, did not show great predictive performance and were limited by study sample size. Therefore, a high-efficiency model which could perform well in predicting the prognosis of HER2-positive BC tumors is needed.

In our planned study, we aim to identify characteristics of BC and apply them to construct a model to predict the prognosis and/or relapse of HER2-positive BC tumors, specifically, for patients who received pathological complete response (pCR) - the disappearance of all invasive cancer in the breast and nearby lymph nodes ( a network of small structures throughout the body that work as filters for foreign substances, such as cancer cells and contain immune cells to help fight infections) after neoadjuvant therapy. Neoadjuvant therapy is the administration of a treatment, such as chemotherapy drugs which destroy cancer cells, before the main treatment, for example surgery. Based on this model, it is expected the patients with high-risk factors could be accurately identified, and therefore given additional therapies.

Specifically, we will try to pool all the eligible patients from the relevant trials (as requested) together to obtain a large sample size, then divide the overall population into two sets (i.e., training set to construct the model, test set to test the model performance). By selecting significant factors seen where patients have relapsed, we would build a prediction model based on these selected factors using the training set. The proposed predictive model will be applied to the population of the test set to show the model performance using statistical methods. Finally, we will select the patients with high-risk of relapse based on the proposed predictive models.