Page 119 - SRPSKO DRUŠTVO ISTRAŽIVAČA RAKA

SDIRSACR Oncology Insights

SESSION 4

IMMUNOONCOLOGY

P22

Integrated Clinical, Molecular, Microbiome, and Radiomic Profiling to Improve Immunotherapy Response
Prediction in Inoperable NSCLC

Fedor Moiseenko , Marko Radulovic , Nadezhda Tsvetkova , Vera Chernobrivceva , Albina Gabina , Any
1
3
1
1,2
1,2
Oganesian , Maria Makarkina , Ekaterina Elsakova , Ilya Agranov , Maria Krasavina1, Daria Barsova1, Elizaveta
1,2
1
1
1
Artemeva1, Valeria Khenshtein1, Natalia Levchenko1, Viacheslav Chubenko1, Vitaliy Egorenkov1, Nikita Volkov1, Alexei
Bogdanov1, Vladimir Moiseyenko1
1N.P Napalkov Saint Petersburg Clinical Research and Practical Centre for Specialized Types of Medical Care (Oncological), Saint
Petersburg, Russia
2N.N. Petrov National Medical Research Center of Oncology Ministry of Public Health of the Russian Federation, Saint Petersburg,
Russia
3Department of Experimental Oncology, Institute of Oncology and Radiology of Serbia, Belgrade, Serbia
Keywords: immune checkpoint inhibitors, NSCLC, PD-L1, biomarkers, gene expression, microbiome, radiomics,
prognosis, machine learning

Background: Checkpoint inhibitors represent one of the most innovative and prominent types of anticancer therapeutics.
Besides their exceptional breadth of usage in solid tumors and remarkable responses in select patients, the majority
of individuals fail to derive meaningful benefit. Unfortunately, no ideal biomarker currently exists to predict which
patients will respond. In non-small cell lung cancer (NSCLC), treatment decisions are typically guided by the level of
PD-L1 expression, yet its predictive value remains limited. Therefore, development of more accurate and sensitive
predictive systems for immunotherapy response remains a priority.
Materials (Patients) and Methods: We conducted a set of complementary studies in Saint Petersburg institutions
aimed at improving immunotherapy patient selection in NSCLC. A retrospective clinical analysis of 415 patients
with inoperable NSCLC was performed to identify prognostic factors using Cox regression modeling. A prospective
biomarker study involving 146 NSCLC patients assessed the predictive value of a novel 10-gene expression signature
(CEI), quantified via PCR in tumor tissue. Microbiome profiling and immune microenvironment characterization were
carried out in 63 patients using 16S rRNA sequencing and immune gene expression analysis, with response defined
as lack of progression within 6 months. Baseline radiomic features from CT scans of 220 patients were analyzed using
a comprehensive machine learning ensemble framework to predict long-term benefit (24-month overall survival).
Results: The clinical study identified six statistically significant prognostic factors (including lymph node status, NLR, and
sex), which allowed classification into prognostic subgroups; patients with favorable profiles had significantly improved
survival with checkpoint inhibitors (p=0.045). The gene expression study confirmed that high CEI values were associated
with improved immunotherapy efficacy, with a 6-month progression-free survival rate of approximately 75% versus 35%
in the low CEI group (p<0.05). Microbiome diversity indices (Chao1, Shannon) and specific taxa abundance correlated
with clinical benefit, although no single genus or phylum was universally predictive. The radiomics ensemble model,
combining clinical and imaging features, achieved an AUC of 0.86 for predicting 24-month survival, outperforming
models based on either data type alone.
Conclusions: Our multidisciplinary results emphasize the inadequacy of PD-L1.

104