Zhi Cheng MD, MPH1, Minoru Nakatsugawa PhD1, 2, Chen Hu PhD3, Ana P. Kiess MD, PhD1, Scott P. Robertson PhD1, Joseph A. Moore PhD1, Michael R. Bowers BS1, Xuan Hui MD, MS1, Brandi R.Page MD1, Laura Burns BSN1, Mariah MuseBSN1, Amanda Choflet MS, RN, OCN 1, Kousuke Sakaue4, Shinya Sugiyama4, Kazuki Utsunomiya4, John W. Wong PhD1 , Todd R. McNutt PhD1, and Harry Quon MD, MS1,  Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University, Baltimore, MD,  Toshiba America Research, Inc., Baltimore, MD,  Oncology Center – Biostatistics/Bioinformatics, Johns Hopkins University, Baltimore, MD,  Toshiba Medical Systems Corporation, Otawara, Japan
Background and Purpose:
To evaluate a prediction model of weight loss in head and neck cancer (HNC) patients treated with radiotherapy (RT) by Classification and Regression Tree (CART) algorithm as a component of learning health system (LHS).
Material and Methods:
From a prospectively collected database, 391 HNC patients from 2007 to 2015 were identified. The data contains 3,015 variables, including patient demographic, delineated dose data, planning target volume – organs at risk shape relationships, on-treatment toxicities and Quality of Life. Weight loss ≥ 5kg at 3 months post-RT was predicted by the CART. Two prediction models at the time of RT planning and at the end of treatment (EOT) were developed.
Weight loss predictors during RT planning were diagnosis, dose to masticatory, superior constrictor muscle, larynx, parotid and age. At EOT, patient reported oral intake, diagnosis, N stage, nausea, skin toxicity, pain, dose to larynx, parotid, and low dose PTV-larynx distance were significant predictive factors. The area under the curve (AUC) of the models at RT planning and EOT were 0.773 and 0.821 respectively.
The established Oncospace® informatics infrastructure can facilitate large-scale analysis predicting weight loss and CART showed potential for the development of a LHS to reduce the risk of radiation toxicities.