The RAPIDxAI trial, presented at the European Society of Cardiology Congress, investigated the efficacy and safety of an artificial intelligence (AI)-based clinical decision support tool in managing emergency department (ED) patients with elevated high-sensitivity cardiac troponin (hs-cTn). The multicenter, cluster-randomized trial across twelve South Australian EDs aimed to determine if the AI tool could improve outcomes compared to standard care in patients with suspected cardiac etiology. The study revealed that while the AI tool did not significantly improve 6-month cardiovascular outcomes, it did enhance adherence to evidence-based care without increasing adverse events.
The trial enrolled 3,029 patients with at least one hs-cTn value above the 99th percentile and suspected cardiac cause. The mean patient age was 75 years, and 58% were female. Participants were randomized to either the AI-based decision support tool (n = 1,568) or standard care (n = 1,461). The primary efficacy outcome, a composite of cardiovascular (CV) death, myocardial infarction (MI), and unplanned CV readmission at 6 months, occurred in 26.0% of the AI group and 26.4% of the control group (hazard ratio 0.99, 95% confidence interval 0.86-1.14, p = 0.872).
The primary safety outcome, a composite of all-cause death or MI at 30 days, was observed in 0.86% of the AI group and 1.1% of the control group (p for noninferiority < 0.001). This indicated that the AI tool was non-inferior to standard care in terms of short-term safety.
Impact on Secondary Outcomes
Interestingly, the study found differences in secondary outcomes. Invasive coronary angiography, when not classified as type 1 MI, was performed less frequently in the AI group (5.2%) compared to the control group (9.4%). Conversely, statin use in patients classified as type 1 MI was higher in the AI group (81.8%) compared to the control group (68.0%).
Interpretation of Findings
According to Dr. Derek Chew, who presented the findings, the AI algorithm provided diagnostic probabilities of type 1 versus 2 MI and acute versus chronic myocardial injury, associated prognostic assessments, and evidence-based management recommendations based on the Fourth Universal Definition of MI. The similar event rates between the groups, coupled with higher rates of evidence-based therapies in the AI arm, suggest that AI-based decision support tools may reinforce adherence to guideline-directed medical therapies in critical diagnoses such as type 1 MI without an increased risk for harm.
Future Directions
The researchers noted that the overall rate of antiplatelet use in type 1 MI-classified patients was still relatively low, potentially reflecting subsequent reclassification based on additional diagnostic data. They suggest that AI algorithms will likely be refined over time. Future studies should also evaluate the cost-effectiveness and scalability of such tools.
