JMIR Publications

ABSTRACT

With the increasing rates of cesarean sections (C-sections) and their impact on maternal health and financial support to the hospitals, we propose an interoperable machine-learning-based clinical decision support system to assist in the detection of unnecessary C-sections. Data from nine hospitals were collected. The results demonstrate the potential effectiveness of the model, with gradient-boosting trees performing best among the tested algorithms, achieving AUROC of around 90%. We also tested the model in a clinical setting to estimate the rate of warnings triggered (3.8%). Additionally, we paved the way for an in-depth clinical validation by making a questionnaire to healthcare professionals in order to compare human assessment and the algorithm developed. A simulation study was developed in order to assess the financial implications of the proposed system on Portuguese public hospitals. The findings indicate that implementing the system could result in financial benefits for a significant number of public hospitals, with some hospitals experiencing between 50% to 100% increase in financial support. In conclusion, our clinical decision support system has an AUROC superior to most of the models and scores existent in the area. It is also interoperable and EHR agnostic. This is promising to improve both health outcomes and the financial well-being of hospitals.