A Bayesian network and heuristic approach for systematic characterization of radiotherapy receipt after breast-conservation surgery

Abstract

Abstract Background Breast-conservation surgery with radiotherapy is a treatment highly recommended by the guidelines from the National Comprehensive Cancer Network. However, several variables influence the final receipt of radiotherapy and it might not be administered to breast cancer patients. Our objective is to propose a systematic framework to identify the clinical and non-clinical variables that influence the receipt of unexpected radiotherapy treatment by means of Bayesian networks and a proposed heuristic approach. Methods We used cancer registry data of Detroit, San Francisco-Oakland, and Atlanta from years 2007–2012 downloaded from the Surveillance, Epidemiology, and End Results Program. The samples had patients diagnosed with in situ and early invasive cancer with 14 clinical and non-clinical variables. Bayesian networks were fitted to the data of each region and systematically analyzed through the proposed Zoom-in heuristic. A comparative analysis with logistic regressions is also presented. Results For Detroit, patients under stage 0, grade undetermined, histology lobular carcinoma in situ, and age between 26−50 were found more likely to receive breast-conservation surgery without radiotherapy. For stages I, IIA, and IIB patients with age between 51−75, and grade II were found to be more likely to receive breast-conservation surgery with radiotherapy. For San Francisco-Oakland, patients under stage 0, grade undetermined, and age >75 are more likely to receive BCS. For stages I, IIA, and IIB patients with age >75 are more likely to receive breast-conservation surgery without radiotherapy. For Atlanta, patients under stage 0, grade undetermined, year 2011, and primary site C509 are more likely to receive breast-conservation surgery without radiotherapy. For stages I, IIA, and IIB patients in year 2011, and grade III are more likely to receive breast-conservation surgery without radiotherapy. Conclusion For in situ breast cancer and early invasive breast cancer, the results are in accordance with the guidelines and very well demonstrates the usefulness of the Zoom-in heuristic in systematically characterizing a group receiving a treatment. We found a subset of the population from Detroit with ductal carcinoma in situ for which breast-conservation surgery without radiotherapy was received, but potential reasons for this treatment are still unknown.

Keywords

• Bayesian network
• Systematic
• Heuristic