Tumor Cell Expression of PD-L1 is Key in Predicting Efficacy of Pembrolizumab for Patients With mTNBC

Data from the KEYNOTE-119 study (NCT02555657) produced estimated trends that suggest tumor cell expression is a significant component for PD-L1, acting as a predictive biomarker for pembrolizumab (Keytruda) efficacy in patients with metastatic triple-negative breast cancer (mTNBC), according to a poster presented at the 38th Annual Miami Breast Cancer Conference, Physicians’ Education Resource®, LLC®.

More, the research suggests a significant number of responders were missed and overall survival (OS) benefits saw higher hazard ratio estimates when only immune cells were used to measure PD-L1 expression.

“Pembrolizumab monotherapy did not significantly improve overall survival (OS) compared with chemotherapy as second- or third-line treatment for patients with metastatic triple-negative breast cancer in the randomized, open-label, phase 3 KEYNOTE-119 study,” wrote the investigators. “However, the benefit of pembrolizumab compared with chemotherapy appeared to be greater with increasing PD-L1 expression, as quantified by combined positive score (CPS).”

Of the 622 patients enrolled in study, 601 had tumor samples that were available for the exploratory analysis. The patient population was randomized 1:1 to either the pembrolizumab arm (n = 309) or to the chemotherapy arm (n = 292).

In the pembrolizumab arm, the overall response rate was recorded at 9.7% (30 of 309 patients) versus 11.3% (33 of 292 patients) in the chemotherapy arm.

When comparing quantitative immune cell density (QID), tumor proportion score (TPS), and CPS, the featured scoring methods, the receiver operating characteristics (ROC) analysis found that CPS (0.69; 95% CI, 0.58-0.80) is better at enriching for responders better than both QID (0.66; 95% CI, 0.55-0.77) and TPS (0.55; 95% CI, 0.46-0.64) across all practical cutoffs.

Specifically, at each cutoff point, the research team observed that QID had both a lower estimated sensitivity and Youden Index than CPS, suggesting that immune cells measured with QID missed responders, not immune cells measured with CPS.

“Using a lower cutoff with QID to capture missed responders is not a viable solution because there will always be a CPS cutoff that detects at least as many responders as a QID cutoff but that misclassifies fewer nonresponders,” wrote the investigators.

In terms of the ORR advantage with CPS, there appears to be a translation with an OS benefit. Specifically, the hazard ratio for OS was slightly lower for CPS than it was for QID.

The data suggest that in tumor cells and immune cells, CPS should be utilized to quantify PD-L1 status for patients with mTNBC.

In regard to the scoring methods, TPS was defined by the research team as “the percentage of PD-L1‒expressing tumor cells (partial or complete membrane staining) relative to the total number of tumor cells.” More, CPS was “the number of PD-L1‒staining cells (tumor cells, lymphocytes, macrophages) divided by the total number of viable tumor cells, multiplied by 100” and QID was “CPS minus TPS.”

Patient eligibility required centrally confirmed mTNBC and those who underwent 1 or 2 systemic treatments for metastatic disease. Once deemed eligible, patients were then randomized 1:1 to receive either 200 mg of pembrolizumab every 3 weeks for up to 35 cycles or a single-agent chemotherapy including capecitabine, eribulin, gemcitabine, or vinorelbine.

“Emerging evidence has demonstrated that PD-L1 expression on tumor cells and immune cells is associated with better antitumor activity in response to immune checkpoint inhibitors in mTNBC,” wrote the investigators.

These results were previously reported at the Virtual 2020 San Antonio Breast Cancer Symposium, held December 8 through 12, 2020.

Reference:

Emancipator K, Winer EP, Lipatov O, et al. Analysis From KEYNOTE-119 in Triple-Negative Breast Cancer (TNBC) Exploring the Contribution of Tumor and Immune Cells to PD-L1 as a Predictive Biomarker. Presented at: 38th Annual Miami Breast Cancer Conference® Virtual Conference; March 4-7, 2021. Poster 29.