Randomised controlled trial of AI-assisted fetal ultrasound and anomaly detection – a health economic evaluation

Background Antenatal detection of fetal malformations is known to result in better outcomes for affected babies, and offers choice to parents. Unfortunately, current screening programmes, including in the UK, fail to detect a large number of cases. There is also wide regional variation in detection rates, worsening inequity. Recent work has demonstrated that artificial intelligence (AI) may be an effective way of improving fetal detection rates, but this has not yet been studied in a large-scale prospective clinical trial. Aims This project aims to assess the efficacy, cost-effectiveness, and acceptability of AI-assisted workflow tools and a tech-enabled remote reporting service in the context of fetal ultrasound anomaly screening. In partnership with university spin-out company, Fraiya, novel tools will be used that automate data collection (images/measurements) and also facilitate an efficient second review of the scan by a human expert (remote reporting), something which has previously been unfeasible. Methods We propose a multi-centre randomized clinical trial and embedded process evaluation, comparing standard ultrasound screening with our AI-assisted ultrasound scans coupled with remote AI-powered human expert second review. The trial will take place in the obstetric screening departments of four NHS trusts. 9566 participants will be randomized and followed up to determine birth outcome and postnatal diagnosis. The main outcome measures of the trial will be the detection rates of congenital anomalies in the two balanced groups, as well as data to allow health economic modelling. We also propose a mixed methods study that will assess acceptability of the intervention via surveys and focused interviews with service users, frontline healthcare professionals and other antenatal healthcare service providers. There will be extensive PPIE work throughout the project, with a new project-specific advisory group being set up and led by our PPI lead, meeting regularly in both the planning and execution stages of our work. Timelines for delivery The total study duration will be 27 months. The trial set up will begin on notification of award success (approximately 3 months before and 3 months after the start date), and include ethics applications, standard operating agreements and data sharing agreements, as well as initiation of the IT integration project supported by all co-applicants and participating NHS Trusts with advertisement of new team roles also beginning at this time. Trial data collection will take place over 12 months, and the following nine months will be used for data cleaning, analysis and reporting. The last 3 months of the project will be solely for the final health technology evaluation using trial results. Six-monthly update reports will be produced. Anticipated impact and dissemination If the efficacy and cost-effectiveness of our AI tools can be demonstrated by the health technology evaluation at the end of the project, then this will facilitate widespread uptake across the UK, and internationally. This would reduce health inequality, by levelling up the worst-performing regions, and increasing detection rates overall. We will also publish our findings (both the randomized trial and the process evaluation study) in high-impact peer-reviewed journals, and present them at international conferences.