A Comparison of the Performance of EndoPredict Clinical and NHS PREDICT in 120 Patients Treated for ER-positive Breast Cancer

Computational algorithms, such as NHS PREDICT, have been developed using cancer registry data to guide decisions regarding adjuvant chemotherapy. They are limited by biases of the underlying data. Recent breakthroughs in molecular biology have aided the development of genomic assays which provide superior clinical information. In this study, we compared the performance in risk stratification of EndoPredict Clinical (EPClin, a composite of clinical data and EndoPredict) and PREDICT in a cohort of patients with breast cancer considered potential candidates for chemotherapy by the clinicians. One hundred and twenty patients with biopsy-proven oestrogen receptor positive (ER )/human epidermal growth factor receptor 2-negative (HER2 ) breast cancer who underwent surgery were included. EPClin and PREDICT were determined for every tumour, and the results were compared. Using EPClin scores performed on 120 tumours, the cohort was stratified into low- (n=60) and high-risk (n=60) groups leading to 50% reduction in total chemotherapy prescriptions. PREDICT differentiated the patients into low- (n=45), intermediate- (n=33), and high-risk groups (n=42). Discordance between scores was demonstrated for 50 (41.66%) tumours. Nine (20%) out of 45 patients with low PREDICT scores had high EPClin scores and would otherwise not have received chemotherapy if the NHS PREDICT tool had been used alone. Eight (19%) out of 42 patients at high risk by PREDICT were reclassified as being at low risk by EPClin and avoided adjuvant chemotherapy. The sensitivity, specificity, positive predictive value and negative predictive value for NHS PREDICT to predict the potential need for chemotherapy as determined by EPClin were 85%, 51%, 68% and 80%, respectively. To our knowledge, this is the first clinical study to compare EPClin and PREDICT. The data indicate that computational algorithms such as NHS PREDICT may not accurately predict the need for chemotherapy leading to overtreatment, undertreatment or uncertainty and anxiety in a significant proportion of patients. This underscores the importance of more personalized prognostic tools.