Machine Learning Pipeline

Background: Unplanned hospitalizations may diminish quality of care among cancer patients receiving radiotherapy (RT). Researchers found that median health care costs from radiotherapy initiation to 30 days post-radiotherapy were $69,108 in non-hospitalized patients vs $119,844 in hospitalized patients. Unplanned hospitalization is a costly toxicity and can also delay patient treatment. Studies have also found correlations between acute care events and worsened survival. 

After analyzing over 700 clinical variables, Oncora's machine learning model successfully identified patients with gastrointestinal cancers undergoing radiotherapy who are at low vs high risk of 30-day unplanned hospitalization. High-risk patients may be hospitalized more than 30% of the time and may have worse outcomes compared with low-risk patients.

‍

Status: This model performed well both retrospectively and prospectively, and is now deployed in MDACC's radiation therapy department as a clinical quality improvement project to help reduce unplanned hospitalizations after RT.

Background: The purpose of this model was to predict overall survival at 6 months, 1 year, and 3 years following radiation of brain metastases. The model was trained with clinical and treatment data from 915 patients, extracted from a variety of institutional sources. After retrospective analysis, Oncora's machine learning methods demonstrated a clinically valid predictive model of overall survival for brain metastases.

Quantitative prognosis tools such as these, in the hands of expert physicians, has the potential to guide treatment decisions and inform end-of-life discussions.

Status: This model performed well retrospectively. Additional efforts are underway to validate these models prospectively.