Background
It is encountering the dilemma of lacking precise biomarkers to predict the response to neoadjuvant chemoimmunotherapy (NACI) and determine whether patients should use immune checkpoint inhibitors (ICIs) in early breast cancer (BC). We aimed to develop a gene signature to predict NACI response for BC patients and identify individuals suitable for adding ICIs.
Patients and methods
Two I-SPY2 cohorts and one West China Hospital cohort of patients treated with NACI were included. Machine learning algorithms were used to identify key genes. Principal component analysis was used to calculate the ImPredict (IP) score. The interaction effects between biomarkers and treatment regimens were examined based on the logistic regression analysis. The relationship between the IP score and immune microenvironment was investigated through immunohistochemistry (IHC) and multiplex IHC.
Results
The area under the curves of the IP score were 0.935, 0.865, and 0.841 in the discovery cohort, validation cohort 1, and in-house cohort. Marker-treatment interaction tests indicated that the benefits from immunotherapy significantly varied between patients with high and low IP scores (p for interaction <0.001), and patients with high IP scores were more suitable for immunotherapy addition.
Conclusions
Our IP model shows favorable performance in predicting NACI response and is an effective tool for identifying BC patients who will benefit from ICIs. It may help clinicians optimize treatment strategies and guide clinical decision-making.