Site Editor
William J. Gradishar, MD, FACP, FASCO
William J. Gradishar, MD, FACP, FASCO
Posted: Monday, January 8, 2024
In the journal Plastic and Reconstructive Surgery, study authors asserted that the use of artificial intelligence (AI) may feasibly and effectively predict infection and explantation after implant-based reconstruction following mastectomy. Their results suggest potential benefits in incorporating machine learning into patients’ perioperative evaluations, stated Charles E. Butler, MD, of The University of Texas MD Anderson Cancer Center, Houston, and colleagues. Machine learning provides data-driven, patient-specific risk assessment that should be disclosed during presurgical counseling and optimization, contributing to shared decision-making, the researchers contend.
“AI has the potential to reshape the field of [cancer-related] plastic surgery by more precisely and accurately identifying factors that lead to poor outcomes than can conventional statistical analysis,” the investigators predicted.
Data regarding 694 reconstructions performed in 2018 and 2019 (481 patients) were randomly divided into training (80%) and testing (20%) sets; the training set used readily available perioperative clinical data. Nine machine learning algorithms were developed to predict periprosthetic infection and explantation. The median follow-up was 16.1 months.
“Periprosthetic infection developed in 113 reconstructions (16.3%), and explantation was required with 82 (11.8%),” which are relatively high rates “despite improvements in prosthesis design and surgical techniques,” noted Dr. Butler and co-investigators. However, machine learning was able to identify 9 and 12 significant predictors of periprosthetic infection and explantation, respectively, demonstrating good discriminatory performance, compared findings identified by multivariate logistic regression. According to the study authors, machine learning outperformed multivariate logistic regression in predicting periprosthetic infection (AUC = 0.73 vs 0.66) and explantation (AUC = 0.78 vs 0.67).
Machine learning “revealed interactions among risk factors that arise in a nonlinear fashion,” emphasized the researchers. For instance, regression analysis failed to identify such important predictors of periprosthetic infection as adjuvant radiotherapy and chemotherapy; further, machine learning identified higher body mass index, older age, and postoperative radiotherapy as factors leading to a higher risk of infection.
Disclosure: The study authors’ disclosure information can be found at journals.lww.com.
Plastic and Reconstructive Surgery
JNCCN Spotlights
