Purpose: Stratification of patients with triple-negative breast cancer (TNBC) for anti-PD-L1 therapy is based on PD-L1 expression in tumor biopsies. This study sought to evaluate the risk of PD-L1 misclassification.
Methods: We conducted a high-resolution analysis on ten surgical specimens of TNBC. First, we determined PD-L1 expression pattern distribution via manual segmentation and measurement of 6666 microscopic clusters of positive PD-L1 immunohistochemical staining. Then, based on these results, we generated a computer model to calculate the effect of the positive PD-L1 fraction, aggregate size, and distribution of PD-L1 positive cells on the diagnostic accuracy.
Results: Our computer-based model showed that larger aggregates of PD-L1 positive cells and smaller biopsy size were associated with higher fraction of false results (P < 0.001, P < 0.001, respectively). Additionally, our model showed a significant increase in error rate when the fraction of PD-L1 expression was close to the cut-off (error rate of 12.1%, 0.84%, and 0.65% for PD-L1 positivity of 0.5-1.5%, ≤ 0.5% ,and ≥ 1.5%, respectively, P < 0.0001). Interestingly, false positive results were significantly higher than false negative results (0.51-22.62%, with an average of 6.31% versus 0.11-11.36% with an average of 1.58% for false positive and false negative results, respectively, P < 0.05). Furthermore, heterogeneous tumors with different aggregate sizes in the same tumor, were associated with increased rate of false results in comparison to homogenous tumors (P < 0.001).
Conclusion: Our model can be used to estimate the risk of PD-L1 misclassification in biopsies, with potential implications for treatment decisions.
Keywords: Atezolizumab; Biopsy; Misclassification; PD-L1; Triple-negative breast cancer.
© 2022. The Author(s).