Abstract P3-08-09: Validation of the online prediction model CancerMath in the Dutch breast cancer population

Abstract

Abstract Background: In breast cancer treatment decision-making, PREDICT is one of the most frequently used tools in clinical practice. The tool estimates the expected benefit of adjuvant systemic therapy on 5- and 10-year overall survival (OS). The model is based on an UK population and was previously validated on several other populations. OS is accurately predicted in these populations. However, PREDICT under- or overestimated OS in several predefined subgroups, which may lead to inadequate treatment decisions. A possible solution for this may be to use CancerMath. This model is based on a UK population and estimates not only OS, but also long-term breast cancer specific survival (BCSS). In this study, we aimed to validate the CancerMath model to determine its accuracy in several predefined subgroups based on patient-, tumor- and treatment-related characteristics. Methods: All women diagnosed with stage I-III primary invasive breast cancer in 2005 who received surgery were identified from the Netherlands Cancer Registry. The accuracy of CancerMath was evaluated by its calibration and discrimination overall, and in several predefined subgroups based on patient-, tumor- and treatment-related characteristics. For BCSS, we used distant metastasis followed by death as a proxy for breast cancer mortality. Calibration was assessed by comparing 5- and 10-year predicted and observed OS and BCSS using chi-squared tests. A difference >3% was considered to be clinically relevant, as the Dutch guidelines advise adjuvant systemic therapy in case of an at least 3-5% expected benefit on survival. Discrimination was assessed by area under the receiver operating characteristic (AUC) curves. Results: In total, 8,032 women were included. CancerMath underestimated 5- and 10-year OS by 2.2% and 1.9%, respectively. AUCs of 5- and 10-year OS were both 0.77. Divergence between predicted and observed OS was most pronounced in grade II tumors, patients without positive nodes, tumors 1.01-2.00 cm, hormonal receptor positive disease, and patients 60-69 years, in which OS was significantly underestimated. CancerMath underestimated 5- and 10-year BCSS by 0.5% and 0.6%, respectively. AUCs were 0.78 and 0.73, respectively. No significant difference was found in any subgroup. Conclusion: CancerMath predicts OS accurately for most patients with early breast cancer although outcomes should be interpreted with care in some subgroups. BCSS is predicted accurately in all subgroups. Importantly, a 3% difference may not be clinically relevant for someone with a predicted benefit of 12% and an observed benefit of 15%, as adjuvant systemic therapy will be indicated in both cases. However, in case these percentages are 1% and 4%, the indication for adjuvant systemic therapy will change. It is therefore crucial to interpret the over- or underestimations for individual patients with consideration of the initial predicted survival without adjuvant systemic therapy. In conclusion, CancerMath can reliably be used in (Dutch) clinical practice to estimate 5- and 10-year OS and BCSS, with a careful interpretation of OS in some subgroups. We encourage health care professionals and researchers to validate clinical prediction models in the specific target population before using them in that population. Citation Format: Marissa Corine van Maaren, Liza Hoveling, Tom Hueting, Luc JA Strobbe, Mathijs P Hendriks, Gabe S Sonke, Sabine Siesling. Validation of the online prediction model CancerMath in the Dutch breast cancer population [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P3-08-09.