AI-Powered CURAᵀᴹ Application for Identifying At-Risk Pregnancies in Obstetric Management

Overview

It is important to identify high pregnancies early through screening so that appropriate care and intervention may be instituted. An AI-assisted risk categorisation approach may be advantageous compared with traditional means of screening. The purpose of this study is to determine if the adoption of an AI-assisted approach in general pregnancy risk screening will improve the accuracy of antenatal risk categorization into high- and low- risk pregnancy groups, ultimately resulting in fewer poor maternal and fetal/neonatal outcomes.

High-risk pregnancies refer to pregnancies at risk of an adverse maternal outcomes (e.g., gestational diabetes, pre-eclampsia) or fetal/neonatal (e.g. preterm birth, still birth, hypoxic-ischemic encephalopathy). In most healthcare facilities, antenatal care is delivered through a general obstetric clinic. The initial screening of risk is guided by the patient's past medical history, past obstetric history for multigravida patients, and the individual provider's knowledge, which may vary across years of experience in the field. Therefore, the triaging of patients into appropriate antenatal care pathways is inconsistent and often inaccurate. AI technology, particularly Machine Learning (ML) has potential to develop predictive models that are able to segregate low-risk from high-risk pregnancies using complex interactions and relationships. The investigators propose a novel AI-assisted risk stratification model in pregnancy that can help to overcome the current gaps. The AI model considers maternal history and simple biophysical measurements performed in pregnancy. The primary objective of the CURAte trial is to compare the composite incidence of maternal and fetal/neonatal adverse outcomes between participants who were randomised to the AI-assisted risk stratification intervention arm and participants who were randomised to the no-AI assisted control arm. The secondary objective is to test the feasibility and acceptability of an AI-assisted antenatal risk stratification approach in a real-life patient-care system. The study will adopt a parallel arm single-blinded, pragmatic randomised controlled trial design. Women presenting at the subsidised antenatal clinics in the first trimester will be approached and assessed for eligibility. A total of 1444 participants (722 in each arm) will be recruited in this study. All participants will be randomised via block randomisation in a 1:1 ratio into two groups (AI-assisted arm versus non-AI assisted arm (standard of care)) which will be done through an electronic programme prepared by the trial statistician. Enrolled participants will be required to complete a questionnaire about their sociodemographic, obstetric and medical history on the FormSG platform prior to consultation with the clinician. The AI-assisted risk stratification will be deployed twice in each participant's pregnancy- at the first trimester visit before 13 weeks' and 6 days' gestation, followed by after the results of the oral glucose tolerance test and third trimester growth scan are available, usually between 31- and 33-weeks' gestation. The results of the AI-assisted risk stratification will not be disclosed in the no-AI intervention arm, until the end of the study. Other study data (i.e. pre-specified study outcomes) will be extracted from medical records at or after 6 weeks from delivery (or at the end of pregnancy) to assess the primary and secondary outcomes. The primary analysis will be conducted on an intention-to-treat basis, for the binary primary composite outcome of maternal/fetal and neonatal morbidity and mortality. For improved precision, a further multiple regression adjusting for factors known to be prognostic of pregnancy and neonatal outcomes including maternal age, BMI, parity, ethnicity will also be conducted. The investigators' proposed new AI-assisted screening model will address the current gaps in the stratification approach and improve the clinical relevance of antenatal screening in the long run, with downstream positive impact on maternal and neonatal well-being, as well as potential cost savings to the healthcare system. By testing this AI-assisted model in an actual clinical setting in a public healthcare institution, the investigators will be able to identify challenges relating to real-world logistics and enablers for translating this digital innovation into clinical practice. The investigators can use the findings to elicit specific modifications to both the AI-assisted model workflow and CuraTM application, ultimately optimising the future implementation as well as acceptability and uptake amongst healthcare providers and pregnant women.