Abstract 20: Individualized outcome prognostication for patients with larynx cancer

Abstract

Background: Accurate and individualized prognostication is obligatory to the optimal management of larynx cancer. The wide variety of treatment options has made decision making complex and challenging for patients and physicians. Predictive models have been developed to calculate the risk of oncologic outcomes using individualized patient data. Implementing these predictive models into clinical practice may improve patient counseling and clinical management, but preemptive validation of their accuracy and reliability is essential. Aim: To utilize an independent patient cohort to characterize and externally validate existing prognostic calculators for larynx cancer. Methods: Four published prognostic calculators that use individualized information to predict 5-year overall survival for patients with larynx cancer were utilized. Analyses were performed using an independent cohort of 246 patients with previously untreated larynx cancer diagnosed between 2002 and 2012. Clinical and oncologic factors were extracted from a prospective epidemiology database and confirmed via chart abstraction. Survival estimates were calculated using each of the four models and subsequent analyses compared the calculators to one another and in reference to observed outcomes. Median follow-up was 48 months. Results: The observed mean 5-year overall survival within the entire cohort was 72%, whereas each of the calculators predicted significantly worse outcomes, with a mean 5-year overall survival of 47%. One model predicted notably lower survival estimates compared to the others, particularly when assigning outcomes to patients with higher risk disease; otherwise, correlation coefficients were fair (0.62-0.81). Analyses of discrimination suggested that two calculators underperform in their capacity to stratify low- and high-risk patients, whereas two others underperform in their discrimination of mid-stage risk. The AUC ranged from 0.65 to 0.73. C-index values were similar for each of the four models (0.66 to 0.69). Multivariate modeling for calculator-estimated survival estimates identified surgical intervention and induction chemotherapy as predictive of better-than-expected outcomes. Conclusion: Prognostic calculators varied in their ability to accurately predict survival in an external cohort of patients with larynx cancer. Suboptimal reliability and accuracy limit the potential integration of existing individualized prediction tools into routine clinical decision-making. The calculators significantly underestimated survival among patients treated with induction chemotherapy and primary surgery, suggesting that treatment modality may need to be better integrated into revised prediction tools. Deficiencies in calculator performance may be further explained by institutional variation in oncologic outcomes. Citation Format: Connor W. Hoban, Emily L. Belille, Lauren Beesley, Gregory T. Wolf, Jeremy M.G. Taylor, Andrew G. Shuman. Individualized outcome prognostication for patients with larynx cancer [abstract]. In: Proceedings of the AACR-AHNS Head and Neck Cancer Conference: Optimizing Survival and Quality of Life through Basic, Clinical, and Translational Research; April 23-25, 2017; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2017;23(23_Suppl):Abstract nr 20.