Screening works better when it is matched to risk, but the societies disagree about how to establish risk, which tool to use and when to apply it. An AJR expert panel review sets out the models in common clinical use, what each society recommends, and a practical route to putting risk assessment into a working breast imaging service.

The problems it names are the ones that stall implementation: variation between society recommendations, uncertainty about the optimal timing of assessment, and - the hardest - translating a calculated percentage into an actual screening decision. A number that changes nothing is a number nobody collects for long. The review also covers deep learning-based prediction from the mammogram itself, which it treats as a research direction rather than a current tool.

For a department, the usable version is to decide three things in advance and write them down: which model, at what point in the pathway, and what each risk band triggers - supplemental MRI, earlier start, different interval, or a referral for genetic assessment. In Indian practice the models matter with a caveat worth stating: most were derived in Western cohorts and their calibration in Indian women is not established, so a risk percentage should inform a conversation rather than be quoted as a fact.

Choose one risk model for the service rather than leaving it to individual preference
Fix the point in the pathway where risk is assessed, so it happens by default
Define what each risk band actually triggers before you start calculating risk
Route high-risk findings to genetic assessment explicitly, not by letter alone
State the calibration caveat when applying Western-derived models to Indian women

Why it matters

Risk-based screening fails at implementation rather than at prediction, and the panel is explicit that translating a percentage into a decision is the unsolved part.

Don't overread it

This is a narrative expert-panel review, not evidence that risk-stratified screening improves outcomes, and the deep learning models discussed are not ready for clinical use.