Scott Robertson, Radiation Oncology and Molecular Radiation Sciences
Purpose: To develop a framework for automatic extraction of clinically meaningful dosimetric-outcome relationships from an in-house, analytic oncology database.
Methods: Dose-volume histograms (DVH) and clinical outcome-related structured data elements have been routinely stored to our database for 592 head and neck cancer patients treated from 2007 to 2014. SQL queries were developed to extract outcomes assessed for at least 100 patients, as well as DVH curves for organs-at-risk (OAR) that were contoured for at least 100 patients. DVH curves for paired OAR (e.g., left and right parotids) were automatically combined and included as additional structures for analysis. For each OAR-outcome combination, DVH dose points, D(Vt), at a series of normalized volume thresholds, Vt=[0.00,1.00], were stratified into two groups based on toxicity severity. The probability, P[D(Vt)], of an outcome was modeled at each Vt by logistic regression. Notable combinations, defined as having an odds ratio ≥1.05 (p<0.05), were further evaluated for clinical relevance using a custom graphical interface. Results: A total of 57 individual and combined structures and 97 outcomes were queried. From all combinations, 17% resulted in significant regression models with an odds ratio ≥1.05. Further manual inspection revealed a number of reasonable models based on either reported literature or proximity between neighboring OAR. The data mining algorithm confirmed the following well-known OAR-dose/outcome relationships: dysphagia/larynx, voice changes/larynx, esophagitis/esophagus, xerostomia/parotid glands, and mucositis/oral mucosa. Other notable relationships included nausea/brainstem and nausea/spinal cord. Several surrogate relationships, defined as OAR not directly attributed to an outcome, were also observed, including esophagitis/larynx, mucositis/mandible, and xerostomia/mandible. Conclusions: Our database platform has enabled large-scale analysis of dose-outcome relationships. The current data-mining framework revealed both known and novel dosimetric and clinical relationships, underscoring the potential utility of this analytic approach. Multivariate models may be necessary to further evaluate the complex relationship between neighboring OARs and observed outcomes.