Predicting the implementation impact of RAPIDx AI in South Australian emergency departments

Background: There were 75,900 presentations to Australian public hospital Emergency Departments (EDs) with a principal diagnosis of Coronary Heart Disease in 2020–21. RAPIDx AI is a randomised controlled trial to test whether computer algorithms in hospital EDs can help doctors provide better care for patients with symptoms that may be due to their heart.

Objective: To develop an evaluation and prediction method to measure stakeholders' perspectives about the implementation impact of RAPIDx AI. This methodological innovation is necessary because person-centred healthcare services require effective technology integration within clinical workflows to provide better patient care while considering the needs of all end-users involved and affected by such types of tech/practices/service changes.

Methods: We introduce an evaluation framework and method based on complexity science and participatory action research (PROLIFERATE). Using Bayesian statistics, we created a protocol and produced computer-simulated results to demonstrate the evaluation and prediction capabilities of the method concerning RAPIDx AI. Ethical approval was granted by the Southern Adelaide Human Research Ethics Committee (SACHREC): OfR no.272.20

Results: Our methodological innovation is informed by 95% probability prediction and credible intervals on these domains of stakeholders' perspectives: Comprehension, Emotional response; Uptake barriers; Motivation and Optimisation. Computer-simulated responses to a PROLIFERATE online survey predicted an Average Impact for RAPIDx AI. The simulation results imply that motivational and emotional knowledge-translation strategies must be implemented for clinicians and the community to improve RAPIDx AI sustainability. (PROLIFERATE constructs benchmarked at 50%—algorithm and data analysis developed in R.)

Conclusion: PROLIFERATE considers the non-linear characteristics of complex and adaptive workflows of acute care environments from an end-user perspective. It can monitor real-world clinical settings, research outcomes, and technological products by assessing their fitness via person-centred parameters and a transdisciplinary approach.