Machine learning Model may Enhance Risk Assessment in Breast Reconstruction: JAMA

A new research published in the Journal of the American Medical Association revealed that a machine learning model which integrates both structured and unstructured clinical data is feasible and effective for supporting patient risk assessment and clinical decision-making. This is effective in both implant-based and autologous breast reconstruction in postmastectomy breast reconstruction (PMBR) patients.

PMBR has long been associated with improved quality of life for patients recovering from breast cancer surgery. However, understanding the risk of complications has remained a challenge, often relying on generalized statistics rather than individualized risk profiles.

This recent prognostic study conducted across two academic medical centers in the US, developed and validated machine learning models capable of predicting major complications within one year of reconstruction surgery. The study analyzed retrospective data from 411 women, aged 18 and older, who underwent either implant-based or autologous reconstruction between 2012 and 2022.

This study used electronic health records and compiled a dataset that included both structured variables (age, body mass index, and treatment type) and unstructured data extracted manually from clinical notes. This combination allowed for a more nuanced analysis of patient risk factors.

Two machine learning models such as extreme gradient boosting (XGBoost) and random forest were tested. The models were trained on 80% of the dataset and validated on the remaining 20%. The primary outcome measured was the occurrence of major complications like unplanned reoperations or hospital readmissions within one year.

The findings showed that 25.8% of patients experienced major complications. Among the models, XGBoost demonstrated superior predictive performance, which achieved an area under the receiver operating characteristic curve (AUROC) of 0.83, when compared to 0.74 for the random forest model. It also outperformed in precision-recall metrics, which indicated stronger reliability in identifying high-risk patients.

The key predictors of complications like the smoking status, receipt of adjuvant radiotherapy, higher body mass index, older age, and diabetes were included. The model maintained consistent performance across different types of reconstruction procedures.

These findings highlight the potential for machine learning to transform clinical decision-making. By providing individualized risk assessments, such models can help patients better understand their options and engage in more informed discussions with their care teams. Future work focusing on testing the model in broader populations should be elaborated.

Source:

Shaheen, M. S., McManus, B. T., Cullen, C. M., Chen, J.-S., Momeni, A., Kuo, C.-F., Tsai, P.-H., & Chung, K. C. (2026). Machine learning model to predict postmastectomy breast reconstruction complications. JAMA Network Open, 9(4), e267232. https://doi.org/10.1001/jamanetworkopen.2026.7232