ICHNO-ECHNO 2022 - Abstract Book

S29

ICHNO-ECHNO 2022

Poster: Imaging, radiomics and artificial intelligence

PO-0059 Positive predictive value of post radiotherapy FDG PET-CT is affected by treatment and HPV-status

Abstract withdrawn

PO-0060 Predicting two-year survival in patients with head and neck cancer in clinical and research settings

D. Kotevski 1,2 , R. Smee 1,2,3 , C. Vajdic 4 , M. Field 5,6

1 The Prince of Wales Hospital, Radiation Oncology, Sydney, Australia; 2 The University of New South Wales, Prince of Wales Clinical School, Sydney, Australia; 3 Tamworth Base Hospital, Radiation Oncology, Tamworth, Australia; 4 The University of New South Wales, Centre for Big Data Research in Health, Sydney, Australia; 5 Ingham Institute for Applied Medical Research, Medical Physics, Sydney, Australia; 6 The University of New South Wales, South Western Sydney Clinical School, Sydney, Australia Purpose or Objective In New South Wales (NSW), Australia, one in three patients diagnosed with head and neck cancer (HNC) between 2012 and 2016 died within five years following diagnosis, with a 5% improvement in overall survival to 62.9% during this timeframe. We investigated parameters that may be associated with survival through comparison of two datasets collated at our facility for clinical and research purposes. The aim was to identify prognostic factors and predict two-year cancer-specific survival (CSS) using these two datasets. Materials and Methods Adult patients with newly diagnosed, previously untreated squamous cell carcinoma of the head and neck, treated definitively with radiotherapy (±chemotherapy ± surgery) at the Prince of Wales Hospital between 2000-2017, with known stage, and nil distant metastasis at presentation, were eligible. The research and clinical datasets contained 23 and 9 variables, respectively, with death data obtained via linkage to the National Death Index. Prognostic factors were investigated using multivariate Cox regression. For machine learning prediction analysis only, a third dataset was also compared; the research dataset containing the 9 variables in the clinical dataset. Five machine learning models (logistic regression, gradient boosted trees, random forest, support vector machine, and artificial neural network) were trained with 5-fold cross-validation and hyperparameter tuning in Python for each dataset. Models were compared on accuracy, sensitivity/recall, specificity, precision, F1 score, and area under the curve (AUC). Results Data on 529 patients in the research dataset, and 430 in the clinical dataset were analysed. In the research dataset, 87 (16%) deaths were due to HNC at two years, and 66 (15%) in the clinical dataset. Hypothyroidism, and higher T and overall stage were found to be negative prognostic factors, and fitness for operation, and nasopharynx and oropharynx primary sites (compared to hypopharynx) as favourable prognostic factors in the research dataset. In the clinical dataset, higher T stage was identified as a negative prognostic factor, and larynx primary site (compared to hypopharynx) as a favourable prognostic factor. Machine learning analysis demonstrated highest performance using the research dataset (23 variables), with the gradient boosted tree and random forest models predicting two-year CSS with the best metrics: 90% test accuracy, 86% and 85% sensitivity/recall, 93% and 94% specificity, 93% and 94% precision, 89% F1 score, and 96% and 97% AUC, respectively. Conclusion Datasets designed for research purposes with more variables demonstrated greater insight into prognostic factors and better model performance for predicting two-year CSS. Hospitals should consider improving data entry and using tree-based models in clinical decision support systems to improve understanding of the factors impacting on and predicting CSS to drive advancements in patient care. 1 Inselspital Bern, Bern University Hospital, and University of Bern, Department of Otorhinolaryngology, Head and Neck Surgery, Bern, Switzerland; 2 Inselspital Bern, Bern University Hospital, and University of Bern, Department of Radiation Oncology, Bern, Switzerland; 3 University of Bern, Institute of Pathology, Bern, Switzerland; 4 Inselspital Bern, Bern University Hospital, and University of Bern, Department of Otorhinolaryngology, Head and Neck Surgery , Bern, Switzerland Purpose or Objective Lymph node metastases are associated with poor outcome in recurrent laryngeal squamous cell carcinoma. Elective neck dissection is therefore performed along with salvage laryngectomy. In this study, we assessed the rate of occult lymph node metastases and diagnostic accuracy of PET/CT and MRI to detect them in recurrent laryngeal cancer. Materials and Methods Retrospective study including patients with recurrent laryngeal cancer after primary radio(chemo)therapy, re-staged by PET/CT and/or MRI treated with salvage laryngectomy and elective neck dissection between 2004 and 2019. Histopathology of neck dissection samples was used as reference. PO-0061 PET/CT and MRI: diagnostic value in neck re-staging before salvage total laryngectomy R. Giger 1 , J. Galli 1 , O. Elicin 2 , M. Wartenberg 3 , L. Anschuetz 1 , L. Nisa 4