Abstract P2-07-05: A clinical calculator to predict disease outcomes in women with hormone receptor-positive advanced stage breast cancer treated with first-line endocrine therapy

Purpose: Endocrine based therapy is an effective strategy to manage hormone receptor-positive, human epidermal growth factor receptor 2-negative (HR+/HER2-) advanced breast cancer (ABC). However, nearly all patients exhibit/develop either de novo or acquired resistance. While prognostic biomarkers of endocrine responsiveness are well established for the adjuvant treatment in ER+ breast cancer, less is known regarding prognostic and predictive biomarkers of response in the first line ABC setting. We sought to develop a clinical calculator based on clinical criteria for predicting progression-free survival (PFS) and overall survival (OS) of women with HR+/HER2- ABC who will be receiving endocrine monotherapy as first-line treatment for ABC.

Methods: The development of the clinical calculator will be based on data from modern clinical trials in women with HR+/HER2- ABC. The studies to be included in the final analyses are given in Table 1. The control arm data from trials1-6 will form the training dataset (N = 1,223) and be used to construct the clinical prediction models. Variables considered include age, race, ECOG status, disease measurability, body mass index, disease-free interval, number of metastatic sites, locations of metastatic sites, prior endocrine therapy, and prior chemotherapy. Missing values will be imputed using single imputation with all variables included in the imputation model. For continuous variables, restricted cubic splines will be used to determine if non-linear effects may be more appropriate. The Lasso regression will be used as a variable selection technique to reduce the dimensionality of covariates; initially all pairwise interactions will be included in the model. Following Lasso regression, the multivariable Cox proportional hazards models will be constructed for PFS and OS including only variables retained in Lasso. The final model will be internally validated for discrimination and calibration using 10-fold cross-validation. External validation will be performed using control arm data from EGF 30008 (N = 536).

Results: To date, control arm data from four trials (trials 1-4) have been received. The preliminary results presented here are based on pooled data from C40503 and LEA, for which data elements have been harmonized. Models for predicting PFS and OS have good calibration and are associated with bias-corrected C-indices of 0.61 and 0.65, respectively. These models will be updated using pooled data from trials 1-6.

Conclusions: Our preliminary data demonstrate that clinical calculators based on baseline clinical factors can provide accurate prediction of PFS and OS in patients with HR+/HER2- ABC treated with first-line ET. If validated, these tools may be used for risk stratification in future clinical trials and to identify patients who may require more or less aggressive therapy.

Table 1:Studies to be includedTrial NumberTrial NameTrial PISample Size in Control Arm1C40503Maura Dickler152 (letrozole)2LEAMiguel Martin179 (letrozole)3FACTJonas Bergh188 (anastrozole)4FALCONJohn Robertson194 (anastrozole)5S0226Rita Mehta345 (anastrozole)6MONARCH 3Matthew Goetz165 (nonsteroidal AI)7EGF 30008Stephen Johnston536 (letrozole)

Citation Format: Polley M-YC, Dickler MN, Johnston S, Goetz MP, de la Haba J, Loibl S, Mehta RS, Bergh J, Roberston J, Barlow W, Liu H, Tenner K, Martin M. A clinical calculator to predict disease outcomes in women with hormone receptor-positive advanced stage breast cancer treated with first-line endocrine therapy [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P2-07-05.