Controlling testing volume for respiratory viruses using machine learning and text mining

Area of Science:

• Pediatric Infectious Diseases
• Medical Informatics
• Health Services Research

Background:

• Viral testing for pediatric respiratory infections is frequent and costly.
• Identifying children who do not require broad viral testing can optimize resource allocation.

Purpose of the Study:

• To develop predictive models using admission data to identify pediatric inpatients with a low probability of viral respiratory infections.
• To assess the potential for reducing viral testing volumes without compromising diagnostic accuracy.

Main Methods:

• Collected clinical data from 1,685 pediatric inpatients undergoing respiratory virus testing (2010-2012).
• Applied machine learning techniques to construct pre-test predictive models for viral infection.
• Utilized text mining to enhance model performance for specific viral tests.

Main Results:

• Machine learning models accurately predicted viral infection status based on available clinical data.
• Text mining improved prediction accuracy for certain viral tests.
• Cost-sensitive models demonstrated the potential to reduce viral test volumes by up to 46% for individual viral assays while maintaining acceptable specificity.

Conclusions:

• Electronic medical record data can be effectively leveraged to build predictive models.
• These models can assist clinicians in reducing unnecessary viral testing in pediatric inpatients.
• Implementing such data-driven strategies can lead to significant cost savings and improved healthcare efficiency.