A. W. Bogan1, D. M. Routman2, M. R. Buras1, M. A. Golafshar1, K. M. Van Abel3, K. Price4, M. L. Hinni5, M. A. Neben-Wittich2, R. L. Foote2, L. A. McGee6, J. C. Rwigema6, M. Gamez2, S. C. Lester2, A. V. Chintakuntlawar4, E. J. Moore3, D. L. Price3, J. W. Gregor7, T. A. DeWees8, D. J. Ma9, and S. H. Patel6; 1Department of Qualitative Health Sciences, Section of Biostatistics, Mayo Clinic, Scottsdale, AZ, 2Department of Radiation Oncology, Mayo Clinic, Rochester, MN, 3Department of Otolaryngology-Head and Neck Surgery, Mayo Clinic, Rochester, MN, 4Division of Medical Oncology, Mayo Clinic, Rochester, MN, 5Department of Otolaryngology, Mayo Clinic, Scottsdale, AZ, 6Department of Radiation Oncology, Mayo Clinic, Phoenix, AZ, 7Department of Otolaryngology, Mayo Clinic, Phoenix, AZ, 8Department of Computational and Quantitative Medicine, City of Hope National Medical Center, Duarte, CA, 9Mayo Clinic Comprehensive Cancer Center, Rochester, MN
Purpose/Objective(s): This study combines data from three multi-site phase II/III clinical trials focusing on HPV mediated oropharyngeal squamous cell cancer (OPSCC) to develop a predictive model for 12-month post radiotherapy treatment (RT) aspiration. The trials reported outcomes for standard of care (SOC) adjuvant concurrent chemotherapy and intensity-modulated radiation therapy (IMRT), de-escalated adjuvant RT and chemotherapy (DART), and mucosal sparing proton beam therapy (MSPBT). We hypothesized that patient reported outcomes (PROs), clinical, and pathologic factors could identify patients with preserved swallow function 12 months post-RT. Materials/
Methods: Patients were assessed for swallowing dysfunction using Penetration-Aspiration Scale (PAS) at baseline and 12 months post-RT from a modified barium swallow study (MBSS). Aspiration was defined as PAS >= 6. Patient-reported outcomes (PROs) were measured using the EORTC QLQ HN35 instrument at baseline and 12-month post-RT. Age at RT, sex, ECOG performance status, tumor laterality/location, staging, treatment modality, fractions, dose, and disease recurrence were considered as potential predictors. A multivariate Bayesian logistic regression model with a lasso prior for feature selection was employed to predict 12-month post-RT aspiration. The model was validated using a 25% holdout test cohort, with training and test AUC serving as key performance metrics. An optimal predicted probability cut point calculated using Youden’s J based on the training data was used to predict 12-month post-RT aspiration (PAS >=6) on the test cohort. Results: 12-month aspiration rates were similar for the training (8/238=3.4%) and test cohort (2/88=2.3%). Age at RT, baseline PAS, obesity (BMI > 31), pharyngeal constrictor V4Gy, and 12-month EORTC QLQ HN35 swallowing score were positively associated with the likelihood of aspiration based on 95% credible intervals. Model training AUC was 0.97 (95% CI 0.97-1.0), and test AUC was 0.90 (0.70-1.0). Using the training-optimized predicted probability cut point to predict aspiration on the test cohort, prediction accuracy was 0.97 (0.90-0.99), specificity was 0.98 (0.92-1.0), sensitivity 0.5 (0.01-0.99), PPV 0.33 (0.01-0.91), and NPV 0.99 (0.92-1.0). Conclusion: OPSCC 12-month post-RT aspiration can be reliably predicted using patient demographics, treatment characteristics, and PROs. The model’s test NPV of 0.99, provides a novel tool to tailor the intensity of patient intervention during the surveillance phase. Using this validated model could have reduced the number of 12-month MBSS ordered from 88 to 3 (97% reduction) in the test cohort. Abstract 3613 – Table 1: Test and training confusion matrices
