Incorporating progesterone receptor expression into the PREDICT breast prognostic model.

Area of Science:

• Oncology
• Biostatistics
• Clinical Prognostication

Background:

• PREDICT Breast is an established tool for early invasive breast cancer prognostication.
• The current version (2.2) does not incorporate progesterone receptor (PR) status.

Purpose of the Study:

• To integrate the prognostic impact of PR status into a new version of the PREDICT Breast tool.
• To evaluate the performance of the updated PREDICT Breast model against the existing version.

Main Methods:

• Analysis of data from 45,088 European breast cancer patients across 49 studies.
• Utilized Cox proportional hazard models to determine the hazard ratio for PR status.
• External validation performed on 11,365 early invasive breast cancer patients from New Zealand.
• Model performance assessed using calibration and discrimination metrics.

Main Results:

• PR-positive tumors were associated with a 23% (ER-negative) and 28% (ER-positive) reduced risk of breast cancer mortality.
• The addition of PR status improved the area under the ROC curve in the New Zealand cohort.
• Discrimination improved for both ER-negative (0.807 to 0.809) and ER-positive (0.898 to 0.902) tumors.
• Model calibration was modest, with 940 observed deaths versus 1151 predicted.

Conclusions:

• Incorporating PR status enhances the PREDICT Breast model's performance.
• The updated model provides more accurate absolute treatment benefit predictions for individual patients.
• Further research is needed to assess the necessity of recalibrating the baseline hazard function.