Clinical models to predict response and survival in metastatic melanoma (MM) patients (pts) treated with anti-PD-1 alone (PD1) or combined with ipilimumab (IPI+PD1).

9542 Background: Currently there are no robust biomarkers to predict immunotherapy response in MM. Specific clinical and molecular variables have been proposed, but in most cases, these factors have been studied individually. We sought to build a predictive model for response rate (RR), progression-free survival (PFS) and overall survival (OS), by including clinical data available at the point of treatment selection for MM pts treated with PD1 or IPI+PD1. Methods: 786 MM pts were included in 4 cohorts; 447 pts treated with PD1 (discovery, n = 343; validation, n = 104) and 339 pts treated with IPI+PD1 (discovery, n = 229; validation, n = 110). Demographics, disease characteristics and baseline blood parameters were examined. Predictive models were selected using multivariate Cox proportional hazard model, logistic regression and LASSO. ROC curve analyses were performed for each model and validation was measured by discrimination índex (c-statistic). Results: Predictive models for RR and PFS in PD1 pts (AUC = 0.69 and AUC = 0.71, respectively) included mutational status (HR for PFS: BRAF 1; NRAS 0.68; WT 0.57; P = 0.002), primary melanoma site (HR for PFS: occult 1; head and neck 0.67, others 1.04; P = 0.052), elevated LDH (HR for PFS: 1.77, P < 0.0001) and monocyte count > median (HR for PFS: 1.56, P = 0.003). Predictive models for RR and PFS in IPI+PD1 treated pts (AUC = 0.71 and AUC = 0.73, respectively) included AJCC stage M1C/M1D (HR for PFS: 2.12, P = 0.009), elevated LDH (HR for PFS: 2.65, P < 0.0001), liver mets (HR for PFS: 1.63, P = 0.038) and basophil count > median (HR for PFS: 0.50, P = 0.003). ECOG ≥ 1, elevated LDH and brain mets associated with worse OS and were included in predictive models for OS in PD1 (AUC = 0.74) and IPI+PD1 (AUC = 0.85). These models showed consistency with internal and external validation (c-statistic: < 10% difference between the original model and validations for all outcomes). Conclusions: A combination of routinely collected clinical factors are highly predictive of outcome in MM pts treated with PD1 and IPI+PD1. A prognostic index will be presented for each treatment. Such tools may be practical, cheap and valuable for clinical decision making.