Good afternoon, my name is Javier Louro, I’m from the Hospital del Mar Medical Research Institute which is placed in Barcelona. This work is a collaboration between my centre and the BreastScreen Norway, which is the national screening programme in Norway. It’s part of my PhD that I did last year but due to COVID we had to delay the collaboration and finally this work is going to be ended later than my PhD.
So, briefly, this work, to put you all a little bit in context, is in order to improve the risk-benefit balance of breast cancer screening that has been a little bit controversial in the last years. Some studies in the last decade have proposed personalised screening strategies based on each woman’s individual breast cancer risk. For this reason, in 2019 there was a group of experts in the European Conference on Personalised Early Detection and Prevention, which was called ENVISION, in Austria, which stated that there was a need in Europe to develop breast cancer risk prediction models based on data from large-scale cohorts and including risk factors and where there is easily obtainable screening participation.
So in this context the main aim of this collaboration was to develop a breast cancer risk prediction model with information from women attending this BreastScreen Norway, the screening programme for Norway, in order to have a tool to identify which are at the higher or lower risk and to personalise the breast cancer screening. We used the information of 57,000 women that were screened in four counties of BreastScreen Norway, the four counties that had systematically collected information on breast density between 2007 and 2019. We followed up then to early 2022 to have three years of follow-up.
This screening programme, the BreastScreen Norway, is a well-established programme; it has 76% participation which is more than that considered high by the European guidelines. We had a total of 180,000 mammograms of these women. So we used a statistical model that is called a partly conditional Cox regression model to obtain the hazard ratios of the different risk factors. We finally identified age, body mass index, mammographic density, family history of breast cancer, benign breast disease, alcohol intake and hormone replacement therapy as factors that confer a higher risk of breast cancer and age at menarche, exercise and pregnancy as protective factors.
With all these factors we gave this mathematical tool that more or less calculated the four-year risk of developing a breast cancer of a woman based on her individual characteristics. Then we used it in our cohort to see more or less how was the distribution of this risk. We found that the risk varied between 0.2%, which is pretty low, to 7.4%, which is really high, that showed us that there are real differences between the risk of all these women. The quartiles were 0.8, 1.1 and 1.4.
Then we used a validation methodology that is called [?] methodology with 1,003 samples to validate the statistics to the model which are the Expected-to-Observed ratio and the area under the ROC curve. The Expected-to-Observed ratio was 1.1 so it showed that the model slightly overestimated the risk but it was a really good confirmation of the model. Then the area under the ROC curve was 63% which is similar for the models that are being used in breast cancer because in breast cancer the area under the ROC curve is always low because it’s hard to predict which women… We have to remember that these are asymptomatic women, it’s hard to predict which ones are going to develop a breast cancer in four years or which won’t. Then we plotted the effect of these variables to see which ones had a higher effect on the model. We found that breast density was the one that had the highest effect but all of them had some effect. More or less that was all.
Now we have the tool so more or less the conclusion of this work is that we designed, we created and we validated this risk prediction model that can be used to estimate the breast cancer risk in every woman participating in BreastScreen Norway. If we want to use it in another context we should have to calibrate it with the incidence of the other countries but we designed it just for Norway.
This model can be key in the future for designing risk-based personalised strategies in order to improve this risk-benefit balance of breast cancer screening.
Anything you’d like to add?
This study, for just itself, doesn’t change what is the paradigm right now of breast cancer screening but it’s important to take it into account in the future. Right now to move from the actual breast cancer screening to the personalised screening strategies is important but we have to do it carefully because it has an impact, an ethical issue, because you are going to offer the women with low risk less screening than the women with high risk and this can have consequences. So we have to be sure that these women are effectively at a low risk and that it is effective, that it improves the effectiveness of the screening. But there are some clinical trials right now, at least within the United States and Europe which we are working with that are going to, in the following years, prove that screening strategies improve the effectiveness of breast cancer screening.
So in that moment that we are sure that we can implement these personalised screening strategies, I think this model is the first one that is developed just with the Norwegian cohort and it has a robust methodology and good information. It can be the one that is used in Norway to do so, to estimate the risk of these women.