Xuan Hui, Harry Quon, Scott P. Robertson, Zhi Cheng, Joseph A. Moore, Michael R. Bowers, Brandi R. Page, Ana P. Kiess, Minoru Nakatsugawa, Seyoun Park, Junghoon Lee, Todd R. McNutt, Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University
Purpose: Risk of late normal tissue toxicities such as xerostomia place significant quality of life and economic burdens on surviving patients. Preventing this toxicity remains limited. The aim of this study is to build a robust comprehensive xerostomia risk prediction model as the foundation for a personalized learning health system (LHS) by incorporating a wide range of clinical, demographic, and dosimetric factors contained within our Oncospace™ database to gain insights into reducing this risk.
Methods and Materials: Head and neck cancers patients treated with intensity-modulated-radiotherapy from 2008-2015 were selected. Patient demographic, clinical and dosimetric factors were queried. Parametric and non-parametric analyses were performed to comprehensively examine risk factors predicting for grade ≥ 2 xerostomia at 90-150 days after radiotherapy.
Results: Risk factors that individually predicted severe xerostomia included chemotherapy, HPV infection, weight loss, submandibular D70, and parotid D95. In univariate analysis, chemotherapy (OR = 2.52, p-value = 0.001), HPV infection (OR = 2.67, p-value < 0.001), over five-kilogram weight loss (OR = 2.58, p-value < 0.001), patients with parotid D95 ≥ 9 Gy (OR = 1.15, p-value < 0.001) and submandibular D70 ≥ 55 Gy (OR = 1.04, p-value < 0.001), had a higher risk for severe xerostomia. In a non-parametric decision tree analysis, parotid D95 ≥ 9 Gy was the dominant classifier node. Conclusions: Our xerostomia risk model identified the influence of several modifiable treatment factors including the significance of the low dose bath to the parotid gland. These observations support the use of this model as a foundation for the development of a personalized xerostomia LHS.