Prediction of unsuccessful endometrial ablation: Random forest vs logistic regression

Abstract

Background: Five percent of pre-menopausal women experience abnormal uterine bleeding. Endometrial ablation (EA) is one of the treatment options for this common problem. However, this technique shows a decrease in patient satisfaction and treatment efficacy on the long term. Study objective: To develop a prediction model to predict surgical re-intervention (for example re-ablation or hysterectomy) within 2 years after endometrial ablation (EA) by using machine learning (ML). The performance of the developed prediction model was compared with a previously published multivariate logistic regression model (LR). Design: This retrospective cohort study, with a minimal follow-up time of 2 years, included 446 pre-menopausal women (18+) that underwent an EA for complaints of heavy menstrual bleeding. The performance of the ML and the LR model was compared using the area under the receiving operating characteristic (ROC) curve. Results: We found out that the ML model (AUC of 0.65 (95% CI 0.56–0.74)) is not superior compared to the LR model (AUC of 0.71 (95% CI 0.64–0.78)) in predicting the outcome of surgical re-intervention within 2 years after EA. Based on the ML model, dysmenorrhea and duration of menstruation have the highest impact on the surgical re-intervention rate. Conclusion: Although machine learning techniques are gaining popularity in development of clinical prediction tools, this study shows that ML is not necessarily superior to the traditional statistical LR techniques. Both techniques should be considered when developing a clinical prediction model. Both models can identify the clinical predictors to surgical re-intervention and contribute to the shared decision-making process in the clinical practice.