Increasing Quality Through Better Predictive Modeling
Todd McNutt, PhD, and his team at Johns Hopkins University have been building the Oncospace database and website since 2007 to collect a large and ever-growing amount of clinical, treatment, imaging, toxicity, oncologic and patient-reported data. The system has been designed so that every radiation oncology patient can be treated as if they are in a clinical trial by collecting the data as part of the normal work flow.
With his ROI grant, Dr. McNutt is using the data from the head and neck cancer patients in Oncospace to build machine learning models that make personalized, evidence-based predictions of treatment toxicities related to weight loss. Together, Dr. McNutt, radiation oncologist Harry Quon, MD, who specializes in treating head and neck cancers, and others on the Oncospace team are working to:
- Better understand how the spatial distribution of the radiation dose influences toxicities.
- Improve current toxicity prediction models by accounting for image-based features of a given structure’s anatomy and more complex spatially dependent features of the dose distribution based on dose gradient, symmetry and analysis of which regions of the structure have the most influence on function.
- Begin developing a decision support tool that will assist with treatment planning and clinical interventions based on these predictions.
Oncospace is designed to support a learning health system in which patient outcomes improve over time -- the data from each patient entered in the system, including his or her outcomes, becomes part of the knowledge base that the next patient’s predictions are based on.
Six institutions currently participate in the Oncospace consortium, and there is room for it to grow. The bigger the dataset, the better the predictive models will become. The ultimate goal is to be able to make these types of tools available to any physician through an Oncospace web portal. Learn more about Oncospace.