Correlation of Predictive Accuracy of PREDICT Version 2.2 of Indian Women With Operable Breast Cancer

Adjuvant therapy for breast cancer is based on clinic-pathological prognostic and predictive markers.The most important prognostic marker is still presence of lymph node involvement1,2. Other factors that contribute to planning adjuvant systemic therapy include, tumor size3, grade3, hormone receptor status4, Her2/neu overexpression5-7, proliferation markers8-9, age at presentation, patient preferences, performance status and comorbidities. Accurate survival estimates, and the likely benefit of adjuvant therapy, are important aspects of information oncologists consider when making decisions following surgery for invasive, early breast cancer. Currently these decisions are based on known pathological prognostic factors including tumour size, tumour grade and lymph node status in addition to the relative risk reductions of any adjuvant therapy1-7. The prognostic and predictive strengths of different factors are variable and the same factor can have different predictive or prognostic value according to the molecular subtype of breast cancer. These markers are not completely independent of each other10. Several predictive models are now available to help estimate the survival and treatment benefits for individual patients.Multivariate Prediction Models (MPM) takes into consideration not just each marker but the effect with all possible combinations of these markers10. MPMs are of two types. They can either be multivariate prognostic model or a multigene predictive model. Examples10 of multivariate prognostic models are IHC4 assay, Adjuvant! Online and PREDICT. Multigene predictive models are OncotypeDx, MammaPrint, PAM50, EndoPredict. Web based mathematical models which use algorithms to predict survival with or without systemic therapy after surgery, like 'Adjuvant! Online' and PREDICT V2.0 use patient characteristics to predict the survival with or without treatment. The inputs required are tumour size, number of nodes involved, grade of tumour, hormone receptor status, Her2 overexpression, Ki67 and comorbidities. Based on these inputs using an algorithm these tools calculate the overall survival at end of 5 and/or 10 years. Then they also predict what would be the added benefit of adjuvant systemic therapies singularly or with combinations. However majority of these models that have been evaluated use the datasets of cancer registries in a particular geographical location or singles institute11,12. This makes blind application of these models to untested populations unpredictable. Various studies have tested web based prognostic models in different populations. In 2011 Hajage D, et al published their results regarding external validation of 'Adjuvant! Online',in a French and Dutch population13. The prediction was overall well-calibrated in the French data. But there was discordance in some subgroups of patients having high grade tumours and HER2 overexpression. Addition of HER2 status, Mitotic Index and Ki67 significantly improved the predictions. In the Dutch data set, the overall 10-year survival was overestimated by 'Adjuvant! Online', particularly in patients less than 40 years of age.Bhoopathyet al, in 2012 tested this tool in an Asian population and concluded that although it differentiates between good and bad prognosis, it systematically overestimates the survival and requires adaptation before usage in Asian population14. Predict is an online prognostication and treatment benefit tool developed in the UK, using cancer registration and survival data recorded by the Eastern Cancer Registration and Information Centre (ECRIC) for 5694 women diagnosed in East Anglia from 1999-2003.15The model was validated in a second cohort of 5468 women from the West Midlands Cancer Intelligence Unit and is available online (www.predict.nhs.uk) providing 5-and 10-year survival estimates and treatment benefit predictions. Wong et al, tested the predictive accuracy of PREDICT V1.0 in the southeast Asian population16. There were 67% Chinese patients while 13% were Indians. The median age in their study was 50 yrs. They showed concordance in observed and predicted OS in most subgroups except for women whore less than 40 years of age. After reviews in literature, for a better fit in various groups, PREDICT V1.0 was updated to version v2.0. V2.0 is equivalent to V1.0 but calibration of V2.0 has improved over V1.0 in patients diagnosed under the age of 40.17 Multigene predictive models like OncotypeDx, MammaPrint, PAM50, EndoPredict are restrictive in their use due to high cost, thus many oncologists in India use the freely available Web based mathematical models, like Adjuvant Online! or PREDICT V2.0. However, there is no data suggesting the validity of prediction using these models in Indian patients. Hence we propose a study to validate the tool within a trial setting, before advising its use in clinical practice. With the aim to compare the 5-year survival estimates from Predict with observed 5-year outcome from the TMC dataset of Indian women treated for operable breast cancer.