Small Biopsies Misclassify up to 35% of PD-L1 Assessments in Advanced Lung Non–Small Cell Lung Carcinomas. Issue 10 (November 2018)

• Record Type: Journal Article
• Authors: Bigras, Gilbert; Mairs, Simon; Swanson, Paul E.; Morel, Didier; Lai, Raymond; Izevbaye, Iyare
• Is Part Of: Applied immunohistochemistry & molecular morphology. Volume 26:Issue 10(2018)
• Publication Date: 2018-11
• Subjects: PD-L1 -- NSCLC -- sampling computer simulation; Diagnostic immunohistochemistry -- Periodicals; Immunohistochemistry -- Periodicals; Cells -- Morphology -- Periodicals; Molecular diagnosis -- Periodicals; 616.079
• DOI: 10.1097/PAI.0000000000000698
• Languages: English
• ISSNs: 1541-2016

Abstract

Abstract : Pembrolizumab is an FDA-approved immune-checkpoint (IC) inhibitor that targets programmed cell death protein PD-1, and recent phase III trials have demonstrated its superiority over chemotherapy in the treatment of patients with advanced non–small cell lung cancer (NSCLC). Eligibility for treatment with Pembrolizumab is based on demonstration of PD-L1 expression on tumoral cells using the approved companion test 22C3 PharmDx (Dako). Access to the drug depends on a tumor proportion score (TPS) expressing the PD-L1 protein above predetermined cutoffs. The scoring interpretation guide requires a minimum of 100 viable cells to be considered adequate for evaluation. Recent studies have questioned the adequacy of the sampling process when small biopsies are utilized. To further explore this concern, the viable tumor area of 426 consecutive NSCLC biopsies and surgical excisions submitted for PD-L1 assessment was measured and recorded with corresponding PD-L1 expression. About 14.6% of all biopsies measured <2 mm 2 creating 2 groups (<2 mm 2 and ≥2 mm 2 ) whose PD-L1 categories distribution [negative (<1%), low expressor (≥1% and <50%), and positive (≥50%)] were compared. Results were significantly different between both groups (χ 2 test; P =0.0012). To help understand this difference, 1, 407, 000 in silico simulated biopsies of various sizes were performed on 201 numerical tumors created from digitalized full sections and analyzed. Not only the same results shown in actual biopsies were reproduced, but the model calculated that up to 35% of very small biopsies were misclassified including a mixture of false negative and false positive results. The percentage decreased to 10% with a threshold of 5 mm 2 . In era of precision medicine, appropriate sampling is more than ever critical to achieve accurate assessment of the NSCLC PD-L1. Ignored in most clinical trials, recording of biopsy size would permit refining data analysis and increase predictive accuracy of current and future biomarkers.