Evaluation of the impact of a third mammographic reading in the French population-based breast cancer screening program.

BACKGROUND

The population-based breast cancer screening program (PBCSP) in France relies on a systematic second reading (Read-2) of mammograms initially interpreted as normal or benign at the first reading (Read-1). To reduce false-positive results and avoid unnecessary diagnostic procedures, a third reading (Read-3) was implemented in two French departments. This study evaluates its impact in terms of the incidence of interval cancers.

METHODS

The retrospective cohort study includes 1328,597 index mammograms (MT0) classified as negative in Read-1 and reviewed in Read-2. These MT0 were performed between 2001 and 2022 on 426,603 women aged 50-74 years, residing in Seine-Saint-Denis and Val-d'Oise, France. Two groups were compared: i)- Index mammograms classified as negative in Read-2 (Control-group, 1312,341 MT0); ii)- Index mammograms initially positive in Read-2 but reclassified as negative in Read-3 (Third-reading-group, 16,256 MT0). Two hierarchical models explored the determinants of positivity in Read-2 and Read-3. The incidence of cancers, particularly interval cancer (≤23 months after Index mammograms), were compared using an adjusted Cox proportional Hazard-ratio model.

RESULTS

Confirmation of positivity by Read-3 was more likely when the radiological abnormality (Architectural-distortion, Focal-asymmetry, Microcalcifications, Round-mass, Spiculated-mass) was seen in Read-3, regardless of the assessment made in Read-2. The risk of interval cancer was higher in the Third-reading-group (Hazard-ratio=2.1 [1.2-3.8], p = 0.009). However, no increased risk was observed beyond 24 months (Hazard-ratio=0.9 [0.7-1.2], p = 0.52). Read-3 avoided more than 6000 ultrasounds, 700 biopsies, and nearly 7600 additional mammograms.

CONCLUSION

While it reduces the overuse of exams, Read-3 was associated with a higher interval cancer risk. These results underline the interest of a national reflection on the place of a targeted Read-3 with a security reference framework, or even its replacement by AI-based decision-making tools.