External validation of PREDICT v3.1 — a breast cancer prognostic model — in 11,041 Dutch women aged ≥70 years from the Dutch Cancer Registry, assessing calibration and discrimination for 5- and 10-year overall survival (OS).
PREDICT v3.1 overestimated 5-year OS by 7.2% (95% CI 6.3–8.0) and 10-year OS by 12.1% (95% CI 11.1–13.1); O/E ratios were 1.39 at 5 years and 1.30 at 10 years. Discrimination was moderate (AUC 0.75–0.76). Overestimation worsened with older age and greater comorbidity burden.
- Single-country (Dutch) registry data may limit generalizability to other healthcare settings. - Comorbidity data granularity and ascertainment from a cancer registry may be incomplete. - The study cannot account for unmeasured confounders that affect competing non-breast cancer mortality in older women.
Clinicians should not rely on PREDICT v3.1 predictions alone when counseling women aged ≥70, especially those with comorbidities — the model systematically overestimates survival and may lead to overtreatment. A recalibrated or age/comorbidity-adjusted tool is needed for this population.
Explore related topics