Validation of a Machine Learning Model for Predicting Postmastectomy Radiotherapy Recommendation Following Immediate Breast Reconstruction.

Background: Postmastectomy radiotherapy (PMRT) in the context of immediate implant-based breast reconstruction (IIBBR) is associated with long-term morbidity. The likelihood of PMRT may influence the type and timing of breast reconstruction chosen in the preoperative setting. This study aimed to validate a machine learning (ML) model for predicting the probability of PMRT recommendations in IIBBR patients, in accordance with the transparent reporting of studies on prediction models for individual prognosis or diagnosis guidelines. Methods: The study cohort comprised 224 breast cancer patients who underwent mastectomy with IIBBR from January 2021 to December 2022. Data were collected on 12 patient characteristics identified as predictive in our ML model. Preoperative characteristics were recorded from clinical history, physical examination, diagnostic imaging, and pathology reports. Results: Of the 224 patients, 37% (n = 84) were recommended PMRT. Our ML model demonstrated high predictive performance, with an area under the receiver operating characteristic curve score of 0.80 (0.74-0.86). The variables most predictive of PMRT recommendation included the size of suspicious lymph nodes, the presence of carcinoma in axillary lymph node biopsies, tumor size, and the use of ultrasound as the initial diagnostic modality. Conclusions: An ML model for predicting PMRT recommendations following IIBBR was validated. This prediction model may be helpful in the preoperative clinical setting to inform the discussion of reconstructive options.