What is it about?
Machine learning methods offer a way to track diseases through automated analysis of Emergency Department triage notes, to improve public health officials’ chances of earlier detection and control of public health problems. However, they are limited by the scarcity of positive examples for model training. In this study we explored several methods for increasing the number of training examples by making new ones based on the existing ones. The best method used a GPT model to create ‘synthetic’ triage notes based on what it can learn from the existing examples. The new data were used to retrain our original models, and improved our accuracy by more than 10 percentage points, without the need for any new ‘real’ data.
Featured Image
Photo by Felipe Salgado on Unsplash
Why is it important?
Febrile convulsions in small children are an uncommon but important side-effect of vaccination. There is a need to monitor the large-scale deployment of new vaccines to ensure they are not causing a lot of convulsions. Computer-based detection of febrile convulsions in emergency department settings can allow faster detection of problems with new vaccines, leading to timely interventions and better health outcomes. Further, the methods we developed will be useful in a range of other disease-detection applications where training examples are scarce. The results will be interesting to anyone who cares about how technological improvements might contribute to better health outcomes.
Perspectives
Conducting the experiments that led to this publication has expanded our perspective on the potential of automatically detecting health issues from triage notes in the emergency department. We are thrilled to explore the various applications of this technology and its potential to enhance our ability to improve public health. The findings of this study have sparked new ideas and possibilities for future research and development in this field. We are excited about the impact this work can have on the health industry and look forward to contributing further to this field.
Sedigh Khademi
University of Melbourne
Read the Original
This page is a summary of: Data augmentation to improve syndromic detection from emergency department notes, January 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3579375.3579401.
You can read the full text:
Contributors
The following have contributed to this page
