Predicting Novasure surgery outcomes based on patient characteristics and treatment activities

Abstract

For 30% of the women, their daily life is influenced by heavy menstrual bleeding, meaning that their energy level, mood, work productivity, social interaction, family life, and sexual functioning alternate due to their menstruation cycle. Endometrial ablation, like the Novasure surgery, can be used as definitive surgical treatment. Endometrial tissue is vaporised during this process, preventing the flow possibility. Such surgery has failed when tissue grows back, flow continues or complaints reoccur. Current medical research has found multiple prognostic factors associated with the failure of the Novasure surgery. While these factors are purely focused on the patient’s characteristics, the question arises whether process features as well point at the failure of the Novasure surgery. This research investigates in the use of historical data of Novasure patients to provide evidence-based insights into current treatments and their impacts on the outcome of the Novasure surgery per patient. Using six machine learning algorithms in four experiments with patient characteristics and process features as potential prognostics factors, the most important features are predicted and the most effective algorithms is determined. Adenomyosis, age, BMI, cavity length, and cavity width are the patient characteristics which have the most influence on the outcome of the Novasure surgery. The addition of process features led to the awareness that investigating in care activities and appointments brings new insights in predicting reinterventions. Random forest, extreme gradient boosting and neural network are the algorithms which can be used best for predicting which patients are likely to undergo a reintervention.