A Machine Learning Approach to Predicting Donor Site Complications Following DIEP Flap Harvest.

[BACKGROUND]  The additional donor site incisions in autologous breast reconstruction can predispose to abdominal complications. The purpose of this study is to delineate predictors of donor site morbidity following deep inferior epigastric perforator (DIEP) flap harvest and use those predictors to develop a machine learning model that can identify high-risk patients.

[METHODS]  This is a retrospective study of women who underwent DIEP flap reconstruction from 2011 to 2020. Donor site complications included abdominal wound dehiscence, necrosis, infection, seroma, hematoma, and hernia within 90 days postoperatively. Multivariate regression analysis was used to identify predictors for donor site complications. Variables found significant were used to construct machine learning models to predict donor site complications.

[RESULTS]  Of 258 patients, 39 patients (15%) developed abdominal donor site complications, which included 19 cases of dehiscence, 12 cases of partial necrosis, 27 cases of infection, and 6 cases of seroma. On univariate regression analysis, age ( = 0.026), body mass index ( = 0.003), mean flap weight ( = 0.006), and surgery time ( = 0.035) were predictors of donor site complications. On multivariate regression analysis, age ( = 0.025), body mass index ( = 0.010), and surgery duration ( = 0.048) remained significant. Radiographic features of obesity, such as abdominal wall thickness and total fascial diastasis, were not significant predictors of complications ( > 0.05). In our machine learning algorithm, the logistic regression model was the most accurate at predicting donor site complications with the accuracy of 82%, specificity of 0.93, and negative predictive value of 0.87.

[CONCLUSION]  This study demonstrates that body mass index is superior to radiographic features of obesity in predicting donor site complications following DIEP flap harvest. Other predictors include older age and longer surgery duration. Our logistic regression machine learning model has the potential to quantify the risk of donor site complications.