PRAIM

Nationwide real-world implementation of AI for cancer detection in population-based mammography screening

Paciente / Población Intervención / Exposición Comparación Desenlace

In 463,094 women aged 50–69 years in organised mammography screening at 12 German sites, AI-supported double reading increased the breast cancer detection rate compared with standard double reading without AI in the same programme, with a non-inferior recall rate.

463.094

Diseño

Non-randomised nationwide implementation study, Germany (12 sites)

Desenlace

Breast cancer detection rate (superiority) and recall rate (non-inferiority)

Relevancia

ResultadoCancer detection 6.7 vs 5.7 per 1,000 (+17.6%, 95% CI +5.7% to +30.8%); recall 37.4 vs 38.3 per 1,000 (−2.5%, 95% CI −6.5% to +1.7%; non-inferior). PPV of recall 17.9% vs 14.9%; PPV of biopsy 64.5% vs 59.2%. 260,739 women were screened with AI by 119 radiologists.

Eisemann N, et al. Nat Med. 2025;31(3):917-924. 10.1038/s41591-024-03408-6

Discusión y crítica

The largest real-world AI screening study, showing that AI support added to double reading raises detection without extra recalls across a national programme. Not randomised: radiologists chose voluntarily whether to use AI, so selection bias is built in; detection is a surrogate, and interval cancers and mortality lie beyond this design. MASAI supplies the randomised evidence.