Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

ePoster Display

1352P - A highly predictive blood-radiomics classifier in advanced NSCLC treated with immunotherapy


16 Sep 2021


ePoster Display


Giulia Mazzaschi


Annals of Oncology (2021) 32 (suppl_5): S949-S1039. 10.1016/annonc/annonc729


G. Mazzaschi1, R. Scandino2, G. Milanese3, C. Pavone3, B. Maurizio3, M. Silva3, R. Ledda3, R. Minari1, F. Trentini1, S. Buti1, P. Bordi1, A. Leonetti1, F. Quaini4, N. Sverzellati3, A. Romanel5, M. Tiseo1

Author affiliations

  • 1 Medical Oncology Unit, University Hospital of Parma, 43126 - Parma/IT
  • 2 Department Of Cellular, Computational And Integrative Biology (cibio), University of Trento, 38123 - Trento/IT
  • 3 Radiological Science, University Hospital of Parma, 43126 - Parma/IT
  • 4 Medicine And Surgery Department, University Hospital of Parma, 43126 - Parma/IT
  • 5 Department Of Cellular, Computational And Integrative Biology, University of Trento, 38123 - Trento/IT

Abstract 1352P


Untangling inter- and intra- tumor heterogeneity functional to clinical decision-making represents an unmet need of the actual Immune checkpoint inhibitors (ICIs)-driven treatment landscape. Multidimensional interrogation of circulating parameters may complement radiomics to non-invasively provide clinically suitable biomarkers. Thus, we aimed at integrating blood hallmarks of systemic inflammation (SI) with high-throughput CT imaging features to develop a predictive model in ICI treated advanced NSCLC.


A cohort of NSCLC patients undergoing ICIs was investigated (n=117). Baseline CT scans and peripheral blood (PB) samples served as a source of, respectively, radiomic features (RFs) and SI indices including derived Neutrophil-to-Lymphocyte ratio (dNLR) and lactate dehydrogenase (LDH). Primary endpoint was the response to ICIs per RECIST. To this end, a Monte Carlo cross-validation exploiting feature selection and logistic regression was used to uncover informative RFs. These RFs together with SI descriptors and a subset of clinical features were mined through an ad hoc genetic algorithm to maximize the classifier accuracy. The discriminative ability of the resulting top 20 selected features was assessed by Kaplan Meier and log-rank tests.


We enrolled 117 advanced NSCLC treated with ICIs as Ist (11%) or IInd or more line (89%). Median OS was 7.8 (95%CI, 5.4-12.8) and PFS 2.5 months (95% CI, 1.1-3.9). According to RECIST criteria, 35% (n:41) were responders (R) while 76 patients belonged to non responders (NR). Using a leave-one-out cross-validation approach we assessed a baseline efficiency of PB-SI (AUC = 0.71). Strikingly, our classifier, including PB-SI and 18 RFs, was significantly associated with clinical response, reaching high performances especially in terms of AUC (AUC = 0.91). Importantly, NSCLC patients classified as NR according to our blood-radiomics predictor displayed a significantly worse OS (median OS 4.1 months, 95% CI, 3.6-6.4) compared to responders (median OS not reached).


Blood-derived pro-inflammatory markers and CT-based RFs may enclose a highly predictive classifier of ICI response in NSCLC affording the non-invasive interception of tumor-immune interactions.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.


Has not received any funding.


All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings