Evaluation of AI-assisted chart review for acute myocardial infarction and cardiogenic shock.

INTRODUCTION

The Danish RETROSHOCK registry has provided valuable insights into patients with acute myocardial infarction complicated by cardiogenic shock (AMICS) and their clinical trajectories. However, continuously updating the registry through manual chart review is time-consuming.

METHODS

Electronic medical records were retrieved from patients admitted to Odense University Hospital from 2018 to 2022, assigned an ICD-10 code suggestive of AMICS. A random sample of 100 consecutive patients was selected for testing the AI-assisted chart review. The AI tool was a natural language processing model identifying AMICS-related keywords. Electronic medical records were initially screened using AI-assisted chart review, with time recorded until AMICS diagnosis or exclusion due to an alternative diagnosis. One week later, a manual chart review was performed. AI-assisted and manual screening times were compared using the Wilcoxon signed-rank test.

RESULTS

AI-assisted chart review identified the same 26 AMICS patients as manual review (Cohen's kappa = 1). Median manual inclusion time was 2:20 minutes, compared with 1:16 minutes with AI-assisted chart review (p less-than 0.001). Median manual exclusion time was 2:45 minutes, and AI-assisted exclusion was 0:59 minutes (p less-than 0.001).

CONCLUSIONS

AI-assisted chart review significantly reduced screening time for including AMICS patients in the RETROSHOCK registry and excluding non-eligible patients, while preserving diagnostic accuracy.

FUNDING: The RETROSHOCK database is supported by Johnson and Johnson Med Tech and the Novo Nordic Foundation.

TRIAL REGISTRATION: Not relevant.