Interpreting free-text cardiac catheterisation reports: A machine learning approach informed by focused ethnography

Comprehensive Summary

The aim of this study was to develop a machine learning model that can extract accurate information for post-procedure reports in cardiac catheterisation. The study utilized medical results from a group of patients from a Taiwan medical centre who underwent cardiac catheterisation between December 2010 and 2020. Additionally, observation sessions were done to examine the reporting practices of healthcare workers after a Percutaneous Coronary Intervention (PCI), to understand the consistency and structure of clinical documentation; this information guided the researchers to identify a strategy to extract structured data from the medical reports. The use of these observations allowed that model to perform significantly better than previous models, as the resulting extraction was similar to what a healthcare worker would have reported. The model’s extraction had about a 90% accuracy for the sections of procedure, vascular access and complication. However, in sections such as post-cath diagnosis and angiography the accuracy dropped to about 50%. For this reason, it is best to use this program as an assistance to healthcare workers, to make the process more efficient while still ensuring accuracy. Overall, the creation of the machine learning model provides a useful tool for post procedural reports in cardiac catheterization that can benefit the medical community.

Outcomes and Implications

Currently, nurses are the ones extracting the information from the results of the cardiac catheterisation which is highly time consuming and allows for an increased chance of an error occurring. The use of this model, developed in this study, can help make this process more efficient and even more accurate. It was seen that this model will produce the best results when being used alongside healthcare workers and can be a very beneficial tool in hospitals. However, this study was done in only one location and should be tested in the country for use before being implemented clinically.