Machine learning insights on optimal timing for re-suturing postpartum perineal wound dehiscence.

OBJECTIVES

To identify factors influencing re-suturing decisions for perineal wound dehiscence, and develop predictive models to assist clinical decision making.

METHODS

Five machine learning models; Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost), analyzed 129 cases of postpartum perineal wound dehiscence (October 2020-September 2023). Variables included clinical symptoms, wound characteristics, degree of perineal trauma, and clinician seniority. Models were validated on 20% of the data set as a hold-out set.

RESULTS

Re-sutured wounds were significantly deeper, longer, and more disrupted, with higher rates of rectal mucosal involvement (p = 0.031), subspecialist involvement (50% versus 0%, p < 0.001), and follow-up needs. The SVM model showed the best performance with area under the receiver operating characteristic curve (AUC) 0.86 (95% confidence interval 0.81-0.91) and accuracy 0.86, identifying wound depth, wound length, and pain score as top predictors. Decision Tree and XGBoost models achieved 0.82 accuracy (AUC 0.83 and 0.85, respectively), with subspecialist involvement, wound depth, and discharge identified as importance features. Random Forest achieved 0.80 accuracy (AUC 0.81). Across models, seniority and wound depth emerged as critical predictors of re-suturing decisions.

CONCLUSION

Machine learning models can reliably predict re-suturing decisions in perineal wound dehiscence. Wound severity and clinician seniority were the strongest predictors supporting data-driven standardization of perineal wound management. Future work should focus on prospective validation, improving interpretability, and clinical integration.