Abstract P4-05-14: Validation of PREDICT version 2.2 on a retrospective cohort of Indian women with operable breast cancer

Abstract Introduction: Several online prediction models that use known prognostic factors in breast cancer (BC) are routinely used to assist in decisions regarding overall survival (OS) benefit of adjuvant systemic therapy. PREDICT ver 2.2 (P2.2) is one such freely available online tool, which uses T size, LN positive, grade, age, hormone receptor (HR), HER2neu status and Ki67.We performed an external validation of P2.2 for overall survival in a retrospective cohort of patients with non-metastatic invasive breast cancer treated at a single tertiary care centre in India. Methodology: Women treated for operable BC (OBC) between 2008 and 2016 with non-metastatic, T1-T2 invasive, HER2neu receptor negative BC and with available 5-year OS data were selected for this study. Median predicted 5- year OS rates were used to calculate predicted events for the whole validation cohort and subgroups. Chi-squared test was used to evaluate the goodness- of-fit of the tool. Results: A total of 2780 eligible patients were included in the analysis with median age of 50(25-82) years, median pT size of 2.5 (0.1-5) cm, 2006 (72.2%) having grade 3 tumour, 1190(42.8 %) having node-positive disease and 883 (31.8%) having triple negative breast cancer(TNBC). The observed and predicted 5-year OS in the whole cohort were 93.74% (95% CI92.77-94.58) and 89.00% (95% CI 89-90), respectively, with an absolute difference of 4.74%(95% CI 3.27-6.22, p <0.001). The observed and predicted 5-year OS were significantly different among patients with node positive disease (91.12%, 95%CI 89.34-92.6, versus 85%,95%CI 85-86, p=0.0003), grade 3 tumours (92.92%, 95%CI 91.71-93.97, versus 86%, 95%CI 86-87, p <0.001), pT>2 cm (92.82%, 95%CI 91.69-93.88, versus 87%, 95%CI 86-88, p <0.001) and TNBC (90.93%, 95%CI 88.80-92.67, versus 82%, 95%CI 82-83, p<0.001). CONCLUSION: PREDICT version 2.2 failed to estimate the 5- year overall survival rates accurately in this retrospective cohort on Indian patients with a significant underestimation in survival across many subgroups. The reasons for this discrepancy could be different biological characteristics in the tumors and the possibility of selection bias in our cohort. We recommend a population-based validation of the tool and caution in its use until a validation is done for our patients in India. Citation Format: Bhavika Kothari, Nita Nair, Sadhana Kannan, Rohini Hawaldar, Vani Parmar, Shabina Siddique, Vaibhav Vanmali, Ashutosh Tondare, Garvit Chitkara, Purvi Thakkar, Tanuja Shet, Shalaka Joshi, Rajendra Badwe. Validation of PREDICT version 2.2 on a retrospective cohort of Indian women with operable breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr P4-05-14.