Two Heads Better Than One? the Challenges of Artificial Intelligence in Postmastectomy Radiotherapy for Node-negative Breast Cancer Patients

Locally advanced breast cancer usually complicates with nodal metastasis and several studies suggest adjuvant radiotherapy could improve tumor control. However, there is limited literature exploring the need of postmastectomy radiotherapy (PMRT) for patients with T3N0 breast cancer. It would be significant benefit to clinical practitioners if artificial intelligence (AI) clinical decision-support systems (CDSS) could integrate all and latest related literature with each patient’s clinical features, and then provide the treatment suggestions ranking according to the level of evidence. Therefore, we investigated the AI CDSS in clinical scenarios and the concordance of real practice in PMRT for T3N0 breast cancer. We retrospectively included female patients aged 18 years and older with pT3N0M0 breast cancer, who received total mastectomy and axillary node dissection, at a single institute from 2006 January to 2018 December. All clinical data, information, and attributes were input into the AI CDSS with de-identification process. A study panel would review whether the patients received radiotherapy or not and the advice from the AI CDSS. Literature provided from the AI CDSS and the clinicians, searching from the PubMed and Cochrane Library databases, would also be investigated. A total of 19 patients were included into this study and all of them received PMRT. However, the AI CDSS suggested observation over PMRT for all patients; it recommended observation while advised PMRT as “for consideration.” The AI CDSS also provided 3 articles supporting its advice. One meta-analysis in 2014 suggested the benefit of PMRT for T3N0 breast cancer is unclear while another one study in 1981 suggested the importance of PMRT for nodal-positive breast cancer and the other one in 1992 focused on radiotherapy after lumpectomy. The radiation oncologists in the panel listed 17 relevant articles from 1997 to 2017. Besides the meta-analysis mentioned above and smaller studies with polarized opinions, two latest SEER database analyses in 2014 and 2017 demonstrated the need of PMRT for advanced breast cancer. For most clinical setting, some studies suggest AI CDSS could benefit to patients, physicians, and even health system. However, criticism also comes along. Our experience suggests AI CDSS has its advantage, yet currently also has certain limitations. With AI assistance or not, well-experienced physicians are still required in making real clinical decision.