Radiomic and clinical data integration using machine learning predict the efficacy of anti-PD-1 antibodies-based combinational treatment in advanced breast cancer: a multicentered study.

BACKGROUND

Immune checkpoint inhibitors (ICIs)-based therapy, is regarded as one of the major breakthroughs in cancer treatment. However, it is challenging to accurately identify patients who may benefit from ICIs. Current biomarkers for predicting the efficacy of ICIs require pathological slides, and their accuracy is limited. Here we aim to develop a radiomics model that could accurately predict response of ICIs for patients with advanced breast cancer (ABC).

METHODS

Pretreatment contrast-enhanced CT (CECT) image and clinicopathological features of 240 patients with ABC who underwent ICIs-based treatment in three academic hospitals from February 2018 to January 2022 were assigned into a training cohort and an independent validation cohort. For radiomic features extraction, CECT images of patients 1 month prior to ICIs-based therapies were first delineated with regions of interest. Data dimension reduction, feature selection and radiomics model construction were carried out with multilayer perceptron. Combined the radiomics signatures with independent clinicopathological characteristics, the model was integrated by multivariable logistic regression analysis.

RESULTS

Among the 240 patients, 171 from Sun Yat-sen Memorial Hospital and Sun Yat-sen University Cancer Center were evaluated as a training cohort, while other 69 from Sun Yat-sen University Cancer Center and the First Affiliated Hospital of Sun Yat-sen University were the validation cohort. The area under the curve (AUC) of radiomics model was 0.994 (95% CI: 0.988 to 1.000) in the training and 0.920 (95% CI: 0.824 to 1.000) in the validation set, respectively, which were significantly better than the performance of clinical model (0.672 for training and 0.634 for validation set). The integrated clinical-radiomics model showed increased but not statistical different predictive ability in both the training (AUC=0.997, 95% CI: 0.993 to 1.000) and validation set (AUC=0.961, 95% CI: 0.885 to 1.000) compared with the radiomics model. Furthermore, the radiomics model could divide patients under ICIs-therapies into high-risk and low-risk group with significantly different progression-free survival both in training (HR=2.705, 95% CI: 1.888 to 3.876, p<0.001) and validation set (HR=2.625, 95% CI: 1.506 to 4.574, p=0.001), respectively. Subgroup analyses showed that the radiomics model was not influenced by programmed death-ligand 1 status, tumor metastatic burden or molecular subtype.

CONCLUSIONS

This radiomics model provided an innovative and accurate way that could stratify patients with ABC who may benefit more from ICIs-based therapies.