Development of a model for predicting risk from breast interval cancer in the female population in the Republic of Croatia

Abstract

There are several risk prediction models for screen-detected breast cancer but to the best of our knowledge, none for predicting risk from the interval cancer in breast cancer screening. The challenge for developing such a model was that the risk factors for both cancers appear to be biological aggressiveness. Our model is based on risk factors identified in the female population in the Republic of Croatia. Anonymized data from 472,395 women who participated in the National Program for Early Detection of Breast Cancer during the first three cycles of the program (October 2006-May 2014) were used. Cancer data from the Breast Cancer Screening Registry were linked by the data linkage method with data from the Cancer Registry of the Republic of Croatia. A total of 789 women with interval cancer and 3,530 women with screen-detected cancer were identified.Multivariate logistic regression in R was used to model the difference between participants with screen-detected cancer and those with interval cancer, using the general linear model (glm) function. The variables used for the analysis were selected using the all subset regression analysis method. The criterion of the least complexity parameter, the Cp-Mallows index, was chosen. Three variables were found to be statistically significant in the model: breast tissue density (p = 0.038), hormone replacement therapy (p = 0.034), and a first-degree family history of breast cancer (p < 0.001). The resulting model has a discriminant accuracy of 0.658 (95% CI 0.602 -0.713). Although our model has poorer predictor reliability, its advantage is that it is based on real-world data and that the criteria for interval cancer were strictly followed. It is best suited for use in the Croatian population of women because we have identified the available risk factors for the development of interval cancer in our population, but with knowledge of a specific epidemiological environment, it can be more widely applied. The model can be used to make recommendations for individual screening participants. The variables of breast tissue density and first-degree of family history of breast cancer increase the likelihood of interval cancer and indicate an increased risk of detecting interval cancer between mammograms. Consequently, individualized risk screening should be considered (modification of screening interval or additional screening by magnetic resonance or ultrasound). According to the model, hormone replacement therapy is positively related to screendetected cancer, and participants who use hormone replacement therapy must be medically monitored due to the increased risk of screen-detected cancer. In addition, participants in the screening program who use hormone replacement therapy and have a higher density of breast tissue should be encouraged to have more frequent mammograms.