Predicting Response to Immunotherapy in Non-Small Cell Lung Cancer - From Bench to Bedside

C. Montoya, B. Spieler, S. M. Welford, A. Dal Pra, T. Diwanji, R. Yechieli, G. Lopes, Ivaylo B Mihaylov

Research output: Contribution to journalArticlepeer-review


PURPOSE/OBJECTIVE(S): Immune checkpoint inhibitor (ICI) therapy has emerged as the standard of care for patients with advanced non-small cell lung cancer (aNSCLC). Immuno-monotherapy (ImT) offers a better side effect profile than combination ICIs, but treatment response rates remain less than 25%. Predicting response to ImT could spare potential responders from adverse events and direct non-responders toward alternative treatments. Radiomics, a datamining technique that extracts patterns from medical imaging, has been incorporated into predictive models for many cancers. We hypothesized that a model combining pre-ImT radiomics, clinical, and laboratory data can predict patient response and guide clinical decision-making. MATERIALS/METHODS: A murine model was designed to search for texture features and blood biomarkers predictive of ImT response. Nineteen C57BL/6 mice had Lewis Lung Carcinoma tumors injected into bilateral flanks. Pre-ImT CTs and bloodwork were obtained on day 7 post-implant. On day 8, 15 mice underwent radiotherapy (RT) to the right flank tumor (24 Gy in 3 daily fractions), followed by daily anti-PDL-1 ImT. Tumors on both flanks were assessed for treatment response. Bilateral response was seen in 4 mice. Pre-ImT CTs were mined for texture features correlating with systemic response. Next, 117 consecutive aNSCLC patients treated with nivolumab monotherapy were identified in an IRB-approved database. Baseline clinical features were assessed, including laboratory data and RT history. For all patients, the primary tumor in the last pre-ImT CT was segmented per RTOG guidelines. These volumes were analyzed for texture features associated with overall survival (OS). Univariate and multivariate cox regression were used to identify pre-ImT imaging, clinical, and laboratory parameters associated with OS. Areas under the receiver operating characteristic curve (AUC) were estimated to assess accuracy to predict greater than median OS. RESULTS: Identical texture features (surface-to-mass ratio, average Gray, and 2D kurtosis) and neutrophil-to-lymphocyte (NLR) correlated with OS in humans (P = 0.041, < 0.001) and systemic response in mice (P < 0.001, = 0.01). A murine model using NLR and radiomics predicted systemic response with AUC of 1. A human model incorporating baseline ECOG, sex, pre-ImT NLR and radiomics had an AUC for predicting greater than median OS (308 days) of 0.808. Half of patients had a history of RT, but location, dose, fractionation, and timing were heterogeneous and RT was not significant. CONCLUSION: For patients with aNSCLC treated with ImT, a combined model of pre-ImT radiomics, NLR, ECOG, and sex is predictive of OS. Texture features and NLR correlated with treatment response in both human and murine cohorts, suggesting generalizability. RT history was too heterogeneous to factor significantly as a predictor of OS. Future directions include identification of genomic biomarkers and model validation on an external patient cohort.

Original languageEnglish (US)
Pages (from-to)e421-e422
JournalInternational Journal of Radiation Oncology, Biology, Physics
Issue number3
StatePublished - Nov 1 2021

ASJC Scopus subject areas

  • Radiation
  • Oncology
  • Radiology Nuclear Medicine and imaging
  • Cancer Research


Dive into the research topics of 'Predicting Response to Immunotherapy in Non-Small Cell Lung Cancer - From Bench to Bedside'. Together they form a unique fingerprint.

Cite this