Advertisement

Machine Learning–Based Scoring of TILs and Outcomes With Immunotherapy in Patients With NSCLC


Advertisement
Get Permission

In a retrospective study reported in JAMA Oncology, Rakaee et al found that tumor-infiltrating lymphocyte (TIL) levels ascertained via machine learning–based scoring on standard histologic images were associated with response in patients receiving immune checkpoint inhibitor therapy for non–small cell lung cancer (NSCLC).  

Study Details

The multicenter study included 685 patients who received immune checkpoint inhibitor treatment between February 2014 and September 2021, including 446 in a discovery cohort and 239 in a validation cohort. A machine-learning automated method was developed to count tumor, stroma, and TIL cells in whole-slide hematoxylin-eosin–stained images of NSCLC tumors. Tumor mutational burden (TMB) and PD-L1 expression were separately assessed. Clinical responses to immune checkpoint inhibitor therapy were identified from medical records.

Key Findings

Median follow-up was 38.1 months in the discovery cohort and 43.3 months in the validation cohort.

In multivariate analysis, high TIL level (≥ 250 cells/mm2) vs lower level was independently associated with improved progression-free survival (HR = 0.71, P = .006) and overall survival (HR = 0.74, P = .03). Similar findings were made in the validation cohort, with hazard ratios of 0.80 (P = .01) for progression-free survival and 0.75 (P = .001) for overall survival.

In the combined cohorts, high vs low TIL level was significantly associated with improved progression-free survival (HR = 0.52, P < .001) and overall survival (HR = 0.59, P < .001) among patients receiving first-line immune checkpoint inhibitor monotherapy and improved progression-free survival (HR = 0.76, P = .008) and overall survival (HR = 0.77, P = .01) among those receiving immune checkpoint inhibitor monotherapy in second- or later-line treatment. 

In analysis of objective response in the discovery cohort, the receiver operating characteristic area under the curve (AUC) for differentiating responders from nonresponders was highest with the combined model of high PD-L1/high TMB (AUC = 0.70). Both the PD-L1/TMB model and a TIL/PD-L1 model (AUC = 0.56) had higher specificity vs PD-L1 alone (AUC = 0.52).

Among PD-L1–negative patients, TIL levels had greater classification accuracy for immune checkpoint inhibitor response v nonresponse (AUC = 0.77) compared with TMB (AUC = 0.65).

The investigators concluded, “In these cohorts, TIL levels were robustly and independently associated with response to immune checkpoint inhibitor treatment. Patient TIL assessment is relatively easily incorporated into the workflow of pathology laboratories at minimal additional cost, and may enhance precision therapy.”

David J. Kwiatkowski, MD, PhD, of the Cancer Genetics Laboratory, Brigham and Women’s Hospital, is the corresponding author for the JAMA Oncology article.

Disclosure: The study was supported by the Norwegian Cancer Society, National Institute for Health Research, Associazione Italiana per la Ricerca sul Cancro, and others. For full disclosures of the study authors, visit jamanetwork.com.

The content in this post has not been reviewed by the American Society of Clinical Oncology, Inc. (ASCO®) and does not necessarily reflect the ideas and opinions of ASCO®.
Advertisement

Advertisement



Advertisement