Abstract
BACKGROUND AND PURPOSE: Bioselection with induction chemotherapy in larynx cancer is associated with excellent larynx preservation and disease-specific survival but requires visual inspection of the primary tumor. We retrospectively compare clinical and imaging response in bioselected patients to develop predictive models of surgeon-assessed response (SR), laryngectomy-free survival (LFS), and overall survival (OS) in bioselected patients. MATERIALS AND METHODS: In a secondary analysis of patients on two single-institution bioselection trials, model building used a regularized regression model (elastic-net) and applied nested cross-validation. Logistic regression-based model was used to predict SR and Cox proportional hazard-based models were used to predict LFS and OS. RESULTS: In 115 patients with a median age of 57 years, most patients had supraglottic tumors (73.0%) and T3/T4 disease (94.8%). Definitive treatment was chemoradiation in 76.5% and laryngectomy in 23.5%. Change in primary tumor (OR = 5.78, p < 0.001) and N-classification (OR = 1.64, p = 0.003) predicted SR (AUC 0.847). Change in tumor volume (HR = 0.58, p < 0.001) predicted LFS (c-index 0.724). N-classification (HR = 1.48, p = 0.04) and pre-chemotherapy tumor volume (HR = 1.30, p = 0.174) predicted OS (c-index 0.552). CONCLUSIONS: Imaging offers a non-invasive opportunity to evaluate response to induction chemotherapy, complementary to surgeon assessment. Further evaluation of approaches to bioselection that optimize generalizability of this paradigm are needed, and clinical trials utilizing imaging to predict outcomes including LFS are warranted.