Related Experiment Video
Updated: Jul 12, 2025
Competing-Risk Nomogram for Predicting Cancer-Specific Survival in Multiple Primary Colorectal Cancer Patients after Surgery
Published on: September 27, 2024
Validation of the PREDICT Prognostication Tool in US Patients With Breast Cancer
Nickolas Stabellini1,2,3, Lifen Cao3, Christopher W Towe4,5
11Case Western Reserve University School of Medicine, Cleveland, Ohio.
Background:
PREDICT is an online prognostication tool derived from breast cancer registry information on approximately 6,000 women treated in the United Kingdom that estimates the postsurgical treatment benefit of surgery alone, chemotherapy, trastuzumab, endocrine therapy, and/or adjuvant bisphosphonates in early-stage breast cancer. Our aim was to validate the PREDICT algorithm in predicting 5- and 10-year overall survival (OS) probabilities using real-world outcomes among US patients with breast cancer.
Methods:
A retrospective study was performed including women diagnosed with unilateral breast cancer in 2004 through 2012. Women with primary unilateral invasive breast cancer were included. Patients with bilateral or metastatic breast cancer, no breast surgery, or missing critical clinical information were excluded. Prognostic scores from PREDICT were calculated and external validity was approached by assessing statistical discrimination through area under time-dependent receiver-operator curves (AUC) and comparing the predicted survival to the observed OS in relevant subgroups.
Results:
We included 708,652 women, with a median age of 58 years. Most patients were White (85.4%), non-Hispanic (88.4%), and diagnosed with estrogen receptor-positive breast cancer (79.6%). Approximately 50% of patients received adjuvant chemotherapy, 67% received adjuvant endocrine therapy, 60% underwent a partial mastectomy, and 59% had 1 to 5 axillary sentinel nodes removed. Median follow-up time was 97.7 months. The population's 5- and 10-year OS were 89.7% and 78.7%, respectively. Estimated 5- and 10-year median survival with PREDICT were 88.3% and 73.8%, and an AUC of 0.77 and 0.76, respectively. PREDICT performed most poorly in patients with high Charlson-Deyo comorbidity scores (2-3), where PREDICT overestimated OS. Sensitivity analysis by year of diagnosis and HER2 status showed similar results.
Conclusions:
In this prognostic study utilizing the National Cancer Database, the PREDICT tool accurately predicted 5- and 10-year OS in a contemporary and diverse population of US patients with nonmetastatic breast cancer.
