Enhancing decision-making in tubal ectopic pregnancy using a machine learning approach to expectant management: a clinical article

Objective: To refine decision-making regarding expectant management for ectopic pregnancy (EP) using machine learning. Methods: This retrospective study addressed expectant management in stable patients with ampullar EP, 2014–2022. Electronic medical record data included demographics, medical history, admission data, sonographic findings, and laboratory results. Follow-up data on βhCG levels and success rates were collected. Statistical analysis incorporated a Decision Tree Classifier, a decision tree-based machine learning model. The cohort was divided into training and testing groups for the machine learning model. This model was evaluated for accuracy, precision, recall, and F1 score to predict success of expectant management. Results: Among 878 cases of EP, the expectant management cohort, comprising 221 cases, exhibited a success rate of 79.6%, with 20.4% requiring subsequent intervention. Mean βhCG levels on admission were 1056.8 ± 1323.5 mIU. The Decision Tree Classifier demonstrated an accuracy of 89%, with precision, recall, and F1 scores of 92%, 95%, and 94%, respectively. Factors for predicting success included clinical symptoms such as pain, the percentage decrease in βhCG levels, gestational age and βhCG level at decision day. Moderate impactful features were white blood cell count, gravidity and maximum tubal dimensions. Smoking status, duration (hours) from time of EP diagnosis to second βhCG test and marital status were minimal significant predictors of success. Conclusion: The Decision Tree-Based classifier model, with 92% precision and 95% recall, may be a valuable tool for predicting treatment success in hemodynamically stable patients with EP, particularly within the initial 24 h of βhCG follow-up.