Validation of PREDICT tool v2.2 in patients with early-stage breast cancer enrolled in the GIM and MIG trials

Background: PREDICT v2.2 is a prognostic tool used to estimate 5- and 10-year overall survival (OS) rates in early-stage breast cancer (BC). While it has been validated in several populations, limited data exist on its performance in cohorts derived from randomized clinical trials. Materials and methods: This study analyzed individual data from patients with early-stage BC enrolled in five phase III randomized clinical trials conducted by the Gruppo Italiano Mammella (GIM) and Mammella InterGruppo (MIG) between 1992 and 2012. Patients with stage IV disease, prior neoadjuvant therapy, or missing data were excluded. PREDICT v2.2 was used to calculate 5- and 10-year OS predictions. Observed survival was derived using Kaplan–Meier estimates, and the model's calibration and discriminatory accuracy were evaluated. Results: Among the 6205 patents included, the median predicted 5-year OS was 87.1%, while the observed 5-year OS was 93.0%, resulting in an underestimation of 5.6% [95% confidence interval (CI) 5.0% to 6.3%]. Similarly, the predicted 10-year OS was 73.8%, compared with an observed OS of 82.1%, with an underestimation of 8.3% (95% CI 7.2% to 9.4%). Discrimination was moderate, with area under the curve values of 66.2% (95% CI 63.5% to 68.9%) at 5 years and 63.9% (95% CI 61.7% to 66.1%) at 10 years. Subgroup analyses revealed that underestimations were more pronounced in patients with human epidermal growth factor 2-positive tumors, larger tumor sizes, or older age. Conclusions: PREDICT v2.2 underestimates 5- and 10-year OS in a clinical trial-derived population of early-stage BC patients. Despite reasonable accuracy, its performance is influenced by specific patient and tumor characteristics. Regular updates are necessary to ensure its applicability in contemporary clinical practice.