Artificial Intelligence to Improve Detection and Risk Stratification of Acute Pulmonary Embolism (AID-PE)

Overview

The goal of this exploratory observational study is to assess the feasibility and real-world clinical impact of implementing Artificial Intelligence (AI) software for the detection of acute Pulmonary Embolism (PE) in patients who undergo Computed Tomography Pulmonary Angiogram (CTPA). The main questions that this study aims to answer are: \[Question 1\] What is the real-world impact of AI on the clinical outcomes and decision making by radiologists and clinicians in the management of acute PE? \[Question 2\] Is AI software for the detection of acute PE acceptable to use in clinical practice and do they have a favourable impact on clinical workload? \[Question 3\] Is it cost-effective to implement AI software for the detection of acute PE in clinical practice? Patients having a CTPA for the detection of acute PE will have their imaging analysed by AI software in combination with a human radiologist. Researchers will aim to compare the clinical and radiology specific outcomes with a retrospective cohort of patients who have had standard routine radiology reporting.

Acute Pulmonary Embolism (PE) results from partial or total occlusion of the pulmonary blood vessels by thrombus, which can cause right ventricular failure and death if not diagnosed and treated early. Acute PE is a common condition with rising mortality. Patients with acute PE are often poorly risk stratified despite clear guidelines. In fact, the 2019 National Confidential Inquiry into Patient related Outcome and Death (NCEPOD) for acute PE highlighted the need to address worsening mortality rates through appropriate risk stratification of the condition. ESC/ERS guidelines for the diagnosis and management of acute PE also advise on the importance of risk stratification. An increased right ventricle: left ventricle (RV:LV) ratio \>1.0 on Computed Tomography Pulmonary Angiogram (CTPA) is associated 2.5-fold increased risk of all-cause mortality, and 5-fold risk for PE-related mortality. This metric is intended to help clinicians distinguish between patients with high and low risk acute PE. Patients stratified as high risk (RV:LV ratio \>1.0) necessitate closer monitoring within an inpatient setting. Whereas, patients stratified as low risk (RV:LV ratio \<1.0) are suitable for early discharge through ambulatory pathways. Therefore, the provision of RV:LV metrics within radiology reporting has potentially important clinical implications. If clinicians are not provided with any quantifiable evidence of RV dysfunction on which to base their treatment decisions, patients with high risk acute PE may be unintentionally considered 'low risk' and discharged home. Furthermore, patients with low risk acute PE may be subject to longer, and potentially unnecessary, inpatient stays which undoubtedly contributes to the cost of healthcare. The integration of Artificial Intelligence (AI) technology within radiology reporting of CTPAs for acute PE could be a potential solution to address this challenge. AI is an increasingly attractive technology within healthcare. It describes a number of computer software techniques which mimic human cognitive function. AI shows promise in ability to detect and risk stratify acute PE. However, most studies have been conducted in retrospective cohorts. Furthermore, no study current has addressed the health economic impact of implementing AI technology within the real-world reporting of acute PE. This observational study will be led by Royal United Hospital Bath NHS Trust (RUH). The aim of this study is to integrate Artificial Intelligence and machine learning technology within the reporting of CTPAs for acute PE. The investigators hypothesise that AI technology can improve the prompt diagnosis, risk stratification, and management of acute PE within a real-world clinical setting. The investigators also hypothesis that integration of AI technology is cost-effective, and acceptable to radiologists and clinicians. Patients whose scans will be included in the study will be all those consecutively presenting to the RUH with a possible diagnosis of acute PE for 12 months before (comparator cohort) and 12 months after (intervention cohort) 'live' introduction of integrated AI technology reporting. For all recruited participants, an anonymised clinician case report form will be used to capture details relating to their demographics, clinical-radiological PE severity, their management, and outcomes including mortality at 12 months. At the point of analysis, the investigators will perform adjustments/matching between the two cohorts for patient baseline characteristics. The investigators will also adjust for calendar time of recruitment, to account for temporal trends. Analysis between both cohorts will also allow development of a decision analysis model to assess the cost-effectiveness of integrated AI technology within CTPA report for acute PE. Clinician and radiologist questionnaires will be used to assess user acceptability.