Fig. 1

Identification of prognostic markers. A Overview of the study design and data analysis. H&E, hematoxylin and eosin; HRD, homologous recombination deficiency. B Consort diagram showing patient enrollment and biopsy collection. QC, quality control. NGS profiling statuses for four patients are not shown: OT-only (n = 1), OT/PD (n = 1), and PD-only (n = 2). Forest plots of clinical variables (C) and molecular features (D) significantly associated with PFS. HR based on PFS calculated using univariate Cox regression analysis and variables with log-rank P-value < 0.05 considered significantly different in PFS. Continuous variables divided into high and low groups based on the median. C AI: letrozole/letrozole + GnRH/exemestane + GnRH. PR, progesterone receptor; M1, palliative treatment; ILC, invasive lobular carcinoma; IDC, invasive ductal carcinoma. D Signature: COSMIC Mutational Signature (version 2). TMB, tumor mutation burden; PAM50, intrinsic breast cancer subtype; non-luminal, HER2-enriched, basal, or normal-like subtype. E Correlogram (center) among clinical and molecular features significantly associated with PFS shows clusters of highly correlated covariates. Averages of variable importance for increased risk of progression (vertical right) based on 500 random resamples of survival elastic net models (75% training sets) show TP53 status, HRD (S3), and nuclear grade as the most important predictors of progression. In survival elastic net multivariate models, variable importance value is absolute standardized beta estimates