919P - Predictive multi-omic signature in locally advanced laryngeal/hypopharyngeal (LH) squamous cell carcinoma (SCC) treated with induction chemotherapy (IC)

Background

Only limited evidence exists about predictive factor(s) in locally advanced (LA) LHSCC. We aim at assessing a multi-omics signature of response to IC and laryngeal-esophageal dysfunction (LED).

Methods

A retrospective, multicentric, case series of stage III-IV LHSCC patients (pts) treated with IC followed by locoregional treatments was collected. Clinical data (comorbidities, smoking, primary site, T and N categories, performance status), baseline histological samples and radiological imaging were retrieved and correlated with response to IC and LED. Transcriptomic data were derived through RNA-sequencing of biopsy samples. Genes were filtered based on gene-level false-discovery rate comparing response to IC or LED. Radiomic features were entered using 24 principal components. Several classification algorithms were fit and compared using different metrics (e.g. Classification error-CE), using a repeated 5-fold cross-validation. Three different settings were compared: only clinical data, clinical + genomic and clinical + genomic + radiomic.

Results

We retrieved 282 pts treated in Italy, Spain and Germany until 2022. Data for clinical, genomic and radiomic analysis were available for 282, 197 and 80 cases, respectively. Median follow-up was 89.4 months (IC_95% 77.7-108.8). 88% of pts were male, 64% current smokers and 68% had larynx primary subsite; 25% were T4, while 54% had a N category >1. After IC, 78% of pts achieved a partial/complete response. Upon RNA-sequencing based transcriptomic analysis, we generated a multi-omic model based on clinical + genomic information, fitted using Support Vector Machine, that achieved a cross-validated AUC > 85% with CE < 30%. A similar approach applied to LED achieved an AUC > 90% with CE < 20%. Adding radiomic features failed to improve the model further.

Conclusions

Using an integrated clinical and omics data, we were able to define the performances of classification algorithms in assessing response according to either IC or LED in LA LHSCC. These algorithms will be applied within a prospective, pilot phase II trial of tailored IC treatment.

Clinical trial identification

Editorial acknowledgement

Funding

Fondazione Regionale per la Ricerca Biomedica (Regione Lombardia), Project ERAPERMED2020-283, GA 779282, and Sächsische Aufbaubank (SAB WI413, Antragsnummer 100610496) and PRIN 20178S4EK9.

Disclosure

P. Bossi: Financial Interests, Personal, Advisory Board: MSD, Merck, Sanofi, SunPharma, GSK, Molteni, Angelini, Nestlè; Financial Interests, Institutional, Coordinating PI: MSD, GSK, Pfizer, Kyowa Kyrin; Non-Financial Interests, Leadership Role, National research group - Board of Directors: GONO; Non-Financial Interests, Leadership Role, Board of Directors: MASCC. All other authors have declared no conflicts of interest.