Clinical Models to Define Response and Survival With Anti-PD-1 Antibodies Alone or Combined With Ipilimumab in Metastatic Melanoma.

Abstract

PURPOSE

Currently, there are no robust biomarkers that predict immunotherapy outcomes in metastatic melanoma. We sought to build multivariable predictive models for response and survival to anti-programmed cell death protein 1 (anti-PD-1) monotherapy or in combination with anticytotoxic T-cell lymphocyte-4 (ipilimumab [IPI]; anti-PD-1 ± IPI) by including routine clinical data available at the point of treatment initiation.

RESULTS

The final model for predicting ORR (area under the curve [AUC] = 0.71) in immunotherapy-treated patients included the following clinical parameters: Eastern Cooperative Oncology Group Performance Status, presence/absence of liver and lung metastases, serum lactate dehydrogenase, blood neutrophil-lymphocyte ratio, therapy (monotherapy/combination), and line of treatment. The final predictive models for PFS (AUC = 0.68) and OS (AUC = 0.77) included the same variables as those in the ORR model (except for presence/absence of lung metastases), and included presence/absence of brain metastases and blood hemoglobin. Nomogram calculators were developed from the clinical models to predict outcomes for patients with metastatic melanoma treated with anti-PD-1 ± IPI.

CONCLUSION

Newly developed combinations of routinely collected baseline clinical factors predict the response and survival outcomes of patients with metastatic melanoma treated with immunotherapy and may serve as valuable tools for clinical decision making.

METHODS

One thousand six hundred forty-four patients with metastatic melanoma treated with anti-PD-1 ± IPI at 16 centers from Australia, the United States, and Europe were included. Demographics, disease characteristics, and baseline blood parameters were analyzed. The end points of this study were objective response rate (ORR), progression-free survival (PFS), and overall survival (OS). The final predictive models for ORR, PFS, and OS were determined through penalized regression methodology (least absolute shrinkage and selection operator method) to select the most significant predictors for all three outcomes (discovery cohort, N = 633). Each model was validated internally and externally in two independent cohorts (validation-1 [N = 419] and validation-2 [N = 592]) and nomograms were created.