Computerized "Learn-As-You-Go" classification of traumatic brain injuries using NEISS narrative data

Area of Science:

• Injury research
• Public health surveillance
• Machine learning applications

Background:

• Classifying injury circumstances from narrative text is crucial for research.
• Manual classification is time-consuming.
• Existing automated systems can be complex for new users.

Purpose of the Study:

• To evaluate the effectiveness of a user-friendly, interactive machine learning program called DUALIST.
• To assess its ability to classify injury circumstances in narrative text.
• To determine if it improves efficiency compared to traditional methods.

Main Methods:

• Utilized a "Learn-As-You-Go" machine learning program (DUALIST).
• Trained classification models interactively to achieve desired accuracy thresholds.
• Applied the program to classify traumatic brain injuries (TBIs) from the National Electronic Injury Surveillance System (NEISS) into sport and non-sport categories.

Main Results:

• The DUALIST program demonstrated effectiveness in classifying injury narratives.
• User training and interactive model refinement were key to its success.
• Classification of tens of thousands of NEISS TBI narratives was reduced from days to minutes after about 60 minutes of training.

Conclusions:

• The DUALIST "Learn-As-You-Go" program offers an efficient and user-friendly solution for injury narrative classification.
• It significantly accelerates the process of categorizing injury circumstances.
• This tool has potential for improving injury research and surveillance data analysis.