ASPO Abstracts
Combined use of competing risk model and whole mammogram image data improves second breast event prediction after initial DCIS diagnosis.
Category: Early Detection & Risk Prediction
Conference Year: 2023
Abstract Body:
Purpose: Classifying risk of second breast events (after ductal carcinoma in situ (DCIS) remains a challenge. Contralateral events are often ignored or censored. To address gaps in risk classification we 1) assess the improvement in accuracy with the inclusion of contralateral breast events in predicting the risk of future breast events; 2) quantify the value of using the whole mammogram image using the FPCs extracted from both breasts to predict the risk of future breast events and; 3) quantify the value of using both the competing risk model and whole mammogram image in combination. Methods: From the St. Louis registry of breast cancer (DCIS and invasive cancer), we identified 74 women with subsequent events at least 6 months after the initial diagnosis of DCIS. We matched women of the same age with DCIS diagnosed in the same year resulting in a total of 185 women for analysis. Breast cancer risk factors and mammograms (Hologic) are available at entry from prospectively collected breast screening records. We evaluate the improvement in accuracy of different approaches to handling competing events when predicting 5-year risk of a second breast event, either pathology confirmed DCIS or invasive breast cancer. We repeat competing risk analysis adding whole breast image. Results: Both censoring and ignoring the competing breast events led to a substantial bias in model estimation. The 5-year prediction accuracies summarized as AUC are significantly improved when competing risk model is adopted (ipsilateral improved from 0.51 to 0.61 and contralateral from 0.52 to 0.65). In the fully adjusted model, the AUC was significantly improved for contralateral events from 0.65 for age and WMI to 0.75 (P <0.01). Conclusion: The competing risk model improves prediction performance and reduces bias for both ipsilateral and contralateral breast events among women with DCIS. The addition of the whole mammogram image can significantly improve the second contralateral event prediction. This approach can better inform precision medicine decision-making and subsequent clinical trial design. This work was supported by the Breast Cancer Research Foundation grant number (BCRF 21-028), and in part by NCI (R37 CA256810).
Keywords: competing risk; whole mammogram image; risk prediction; contralateral breast