Real alerts and artifact classification in archived multi-signal vital sign monitoring data: implications for mining big data
Data from vital sign monitoring devices can be stored in hospital information systems or health data warehouses. This data is usually averaged and archived within the data repositories, sometimes with clinical data or with administrative data for the purpose of developing clinical decision support systems or to capture new knowledge. Artifact data can present both technical and methodological challenges and human experts can be costly to determine the accurate classification of real data or artifact captured to present as real data. 
Continuous vital sign data streams were collected over an eight-week period in a surgical-trauma unit. Artifact rules and definitions were created using annotation by human experts. Features were extracted and a machine-learning model was developed.
It was determined there were human inconsistency in the definition of real and artifactual data and too few were captured for real comparison from one institution and one intensive care unit. Additionally, data was captured from only one monitor manufacturer.
Continuous vital sign capture is common in healthcare settings and it was concluded that the majority of artifact captured was specific to one vital sign. With this differentiation in the data, real alerts and artifacts could be identified with in the data repository and through machine learning, the data could be used to build meaningful prediction models in big data.
Area of interest and comments
This is an important breakthrough in perfecting the data capture from electronic health records (EHR) so that the data is meaningful for assessment and implementation of patient quality measures. This is particularly valuable when the data is used to provide alerts to providers within the clinical decision support (CDS) system.
Related Article Reviews
- Hravnak, M., et al. Real alerts and artifact classification in archived multi-signal vital sign monitoring data: implications for mining big data, http://www.ncbi.nlm.nih.gov/pubmed/26438655