Machine Learning Miscarriage Management Clinical Decision Support Tool Study

Overview

Machine learning used to develop an algorithm to determine chance of success with expectant or medical management for an individual patient. Taking into account the following objective measures: * Demographics: Maternal Age, Parity * History: Previous CS, Previous SMM/MVA, Previous Myomectomy * Gestation by LMP * Presenting symptoms: Bleeding score, Pain score * USS Measurements: CRL, GS, RPOC 3 dimensions, Vascularity * Discrepancy between gestation by CRL and LMP Audit to collate 1000 cases and identify features contributing to an algorithm that can predict outcome of miscarriage management for individualized case management.

* Artificial intelligence discovery science: Algorithm Development based on a retrospective Audit of approximately 1000 cases of miscarriage * To determine the reliability of the tool with test data sets * To increase the sensitivity and specificity of the decision aid by widening the data collection to multiple sites and testing the algorithm with prospective data The study will be conducted at Queen Charlotte's and Chelsea Hospital at Imperial College Healthcare NHS Trusts (Primary Centre of the study). This is a multi-centre retrospective, cohort observational study. The study will be conducted over a minimum of three years to enable sufficient time to go through the retrospective data and collate test data sets. Retrospective annonymised cases of missed miscarriage and incomplete miscarriage managed at Imperial College Healthcare NHS Trust will be analyse: For each case the following clinical features will be collated and outcomes: * Demographics: Maternal Age, Parity * History: Previous CS, Previous SMM/MVA, Previous Myomectomy * Gestation by LMP * Presenting symptoms: Bleeding score, Pain score * USS Measurements: CRL, GS, RPOC 3 dimensions, Vascularity * Discrepancy between gestation by CRL and LMP All data will be collected retrospectively and annonymised. Following data collection, machine learning models and feature reduction methods will be applied to determine the best performing model to predict success or failure of expectant or medical management of miscarriage respectively. The next phase will include a prospective audit to collect data and test the predictive power of the MLM clinical decision support tool.