The MASAI trial, a landmark study in AI-supported mammography screening, reported non-inferiority for interval cancer rate alongside higher sensitivity and unchanged specificity. That finding has sparked a flurry of correspondence in The Lancet, where researchers have scrutinized the trial's design and conclusions.
Several correspondents argue that interval cancer rates and observed sensitivity are shaped by tumor biology and the duration of the preclinical-detectable phase, not just test accuracy. Others worry that the trial's recommendations for implementation rest on intermediate performance endpoints despite epidemiological conclusions that raise concern for potential overdiagnosis.
The MASAI trial itself concluded that AI-supported screening "can efficiently improve screening performance" and "may be considered for implementation in clinical practice." But critics contend these recommendations are premature without longer-term outcome data.
Overdiagnosis—the detection of cancers that would never cause symptoms—remains a unresolved risk in any screening program. If AI amplifies detection without corresponding mortality benefit, it could subject women to unnecessary biopsies and treatment.
One corresponding author noted that further research should measure benefit in terms of breast cancer mortality, not just surrogate endpoints. Without such data, wide adoption risks outpacing the evidence.