Highlight
This study showcases the development and validation of machine learning algorithms capable of accurately predicting patient-reported outcomes one year following breast reconstruction surgery. Leveraging large multicenter datasets, the models highlight key factors influencing outcomes, facilitating enhanced shared decision-making and tailored patient care.
Study Background
Breast reconstruction surgery is a critical component of the multidisciplinary treatment for breast cancer patients, offering not only physical restoration but also psychological benefit. However, patients’ satisfaction and quality of life post-reconstruction vary significantly due to multiple factors, including surgical technique, patient characteristics, and adjuvant treatments such as radiation. Patient-reported outcome measures (PROMs), particularly the BREAST-Q instrument, have become essential tools to quantify the subjective success of reconstructive procedures. Despite their importance, predicting individual patient outcomes remains challenging. This knowledge gap limits a surgeon’s ability to counsel patients effectively and tailor treatments to optimize long-term satisfaction.
Machine learning (ML) techniques offer promising opportunities to analyze complex, multivariate clinical data and capture nonlinear relationships to predict outcomes more accurately than traditional statistical methods. Applying ML to predict PROMs after breast reconstruction could personalize preoperative counseling and improve shared decision-making by setting realistic expectations and guiding reconstructive strategy choices.
Study Design
This retrospective study analyzed data collected from women undergoing breast reconstruction at Memorial Sloan Kettering Cancer Center (MSKCC) between January 2010 and March 2024. The dataset included patient demographics, clinical variables (such as timing of radiation, body mass index, age), reconstructive techniques, and BREAST-Q score domains collected preoperatively and at 1-year post-operation. A total of 2,687 patients from MSKCC were used for model development.
Five distinct machine learning algorithms were developed to predict whether patients would show improvement in specific BREAST-Q domains at 1 year: physical well-being of the abdomen, satisfaction with the breast, sexual well-being, physical well-being of the chest, and psychosocial well-being. For external validation, the models were tested on a separate cohort of 2,089 patients from the multicenter Mastectomy Reconstruction Outcomes Consortium dataset.
Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and the Brier score, which together assess discrimination and calibration of predictive accuracy.
Key Findings
The machine learning models achieved impressive predictive performance across various patient-reported outcome domains. The highest accuracy was noted for predicting physical well-being of the abdomen with an AUC of 0.97, indicating excellent discrimination. Satisfaction with the breast was also robustly predicted (AUC 0.86), followed by sexual well-being (AUC 0.79), physical well-being of the chest (AUC 0.78), and psychosocial well-being (AUC 0.74).
Variables consistently identified as the most influential predictors included:
- Preoperative BREAST-Q scores, emphasizing the importance of baseline patient status
- Timing and administration of radiation therapy
- Body mass index (BMI)
- Age at reconstruction
- Type and technique of breast reconstruction performed
These findings underscore the multifactorial nature of patient satisfaction and recovery after breast reconstruction.
The external validation in an independent, multicenter cohort demonstrated the models’ generalizability and robustness beyond a single institution setting.
Expert Commentary
This study exemplifies the growing integration of artificial intelligence into surgical oncology and reconstructive medicine. Predicting patient-reported outcomes before surgery represents a substantial advance toward personalized medicine in breast reconstruction. The use of well-validated PROM instruments like BREAST-Q enhances clinical relevance by directly addressing patient-centered endpoints rather than surgeon-centered or anatomical metrics alone.
Limitations inherent to retrospective design include potential selection bias and missing data, which the authors likely mitigated with appropriate preprocessing techniques, though these details warrant scrutiny in full-text review. Future prospective validation and incorporation of additional psychosocial and lifestyle variables could further refine predictive accuracy.
Implementing such algorithms in clinical workflows would necessitate user-friendly interfaces and safeguards against algorithmic bias. Moreover, elucidating mechanistic insights behind predictive factors—such as the impact of radiation timing on tissue viability and patient perception—could complement data-driven predictions with biological plausibility.
Conclusion
Machine learning algorithms can accurately predict patient-reported outcomes one year following breast reconstruction, utilizing readily available clinical and patient-reported data. These predictive models have the potential to revolutionize preoperative counseling and shared decision-making by personalizing risk-benefit profiles, thus improving patient satisfaction and quality of life. Integrating such tools into routine clinical practice could foster a new paradigm of data-driven, patient-centered reconstructive care.
Future research should focus on prospective validation, integration into decision-support systems, and exploring the impact on clinical outcomes and patient satisfaction when used in real-world settings.
Funding and ClinicalTrials.gov
The provided abstract and citation did not explicitly mention funding sources or clinical trial registration numbers. Given the retrospective nature of the study, it may be an observational cohort without specific interventional registration.
References
1. Chen J, Gabay A, Boe LA, Shammas RL, Stern C, Pusic A, Mehrara BJ, Gibbons C, Nelson JA. Machine Learning Accurately Predicts Patient-reported Outcomes 1 Year After Breast Reconstruction. Ann Surg. 2025 Mar 5;284(1):176-183. PMID: 40040622.
2. Pusic AL, Klassen AF, Scott AM, Klok JA, Cordeiro PG, Cano SJ. Development of a new patient-reported outcome measure for breast surgery: the BREAST-Q. Plast Reconstr Surg. 2009 Apr;124(4):345-53. doi:10.1097/PRS.0b013e3181b5e7ee.
3. Maroulakos M, Lafreniere D, Kim J, et al. The impact of radiation therapy on breast reconstruction: a systematic review and meta-analysis. Ann Plast Surg. 2020;84(3):298-305. doi:10.1097/SAP.0000000000002243.
