Description
DEEP.PISTE, a project that sheds new light on the performance of organized breast cancer screening (DOCS)
Context
Breast cancer is the deadliest among women in France: nearly 12 women die from it each year. Organized screening was generalized throughout the country in 000. It aims to carry out a screening mammogram every two years in women aged 2004 to 50, which should make it possible to reduce mortality by almost 74%. . However, even today, for example, false positives (people diagnosed even though they do not have cancer) following a mammogram examination are around 21%; and approximately one in six cancers is not detected during this examination (it is a false negative). There is therefore still room for improvement, and it is necessary to measure the effectiveness of the screening program.
Expression of need
The DEEP.PISTE project presented its latest results during the 18th French-speaking CLINical EPIdemiology conference (EPICLIN) and 31st Days of Statisticians of Cancer Centers (JSCLCC) organized in Dijon, from May 15 to 17, 2024.
The DEEP.PISTE project proposes a new methodology to measure the performance of the organized breast cancer screening program by collecting data systematically. The DEEP.PISTE project cross-references data from mammograms carried out as part of breast cancer screening with Health Insurance data. Thanks to this matching, it becomes possible to precisely identify breast cancers occurring after each screening. Performance between 2008 and 2018 was measured in terms of sensitivity (75.15%), positive predictive value (20.15%), interval cancer rate (27.84% of all identified cases) and rate of identified cancers ( 12.5 cases per 1000 screenings). Although the calculation of these metrics is recommended by European directives, they are not calculated routinely, in particular due to the lack of homogeneous data from breast cancer registries in France. The proposed method has the advantage of being able to be easily applied to the entire territory. Performance metrics can also be broken down based on different data, such as age groups or territorial level.
All processing was carried out within the Health Data Hub technological platform. In a secure workspace, data scientists from Epiconcept in coordination with doctors from CRCDC-OC were able to access pseudonymized data and data analysis tools based on open technologies (open source). This work made it possible to produce software building blocks capable of simplifying and accelerating the analysis of SNDS data. These algorithms will be registered in the Open Library of Health Algorithms (BOAS), available on the Health Data Hub website, so that any researcher can have access to them. The project is therefore part of an open source approach, with the aim of improving the care pathway for purposes of public interest.
In addition to the reconstruction of the screening process and breast cancer, each positive digital mammogram was annotated by expert radiologists. These data will be used to adapt and optimize state-of-the-art open source algorithms for breast cancer detection. The objective is therefore to evaluate the real impact that the inclusion of artificial intelligence in the screening system could have, with the aim of improving women's health and advancing the fight against cancer.
The DEEP.PISTE project is the result of collaboration between the public, private and academic sectors. Supported by the Occitanie Regional Screening Center (CRCDC-OC), Epiconcept (Developer of screening solutions), the HEKA laboratory (INRIA, Inserm, APHP) and the Rouget de l'Isle Radiology practice. It also counts on the support of the Health Data Hub and the CIFRE system. The project was the winner of the first call for projects from the Health Data Hub launched in 2019. The DEEP.PISTE project team was also able to benefit from the support of Health Insurance for the provision of data from the database main body of the SNDS.
