Computational Modeling to Predict Demand for Chest Pain Management
Authors: Ruoming Jin PhD, Wei Xiong PhD, Brigitte Piniewski MD
Abstract— Each time an individual presents at the emergency room with chest pain, this is an expensive problem which uses multiple high cost hospital units such as the emergency room, radiology, the cath lab, cardiac intensive care unit and so on. Currently, most hospitals are poorly able to predict the demand for these services and thus may under staff or over staff these departments on any given day. Imagine if a computational model could deliver a ―community chest pain index‖ and enable services to gear up or gear down in anticipation of a more predictable demand for their expensive services.
To solve this problem, PeaceHealth Laboratories connected live laboratory testing volumes with the computational expertise of Kent state University. A reasonable surrogate quantifying the level community chest pain management demand is troponin laboratory testing rates per date per location. The Kent State team used the first two years of troponin volumes as training data and the last half year as test data for three communities. The error measure was as low as 13.2% in one community. Males were consistently more predictable than females. The day shift variability was greater than night shift, and the week days were more predictable than weekends. A number of adaptive modeling and stratified modeling incorporating a sliding rule was used to further shrink the error measure which could also be broken down by the individual day.
Finally, non-traditional community influencers such as sport games, weather changes, unemployment trends, and crime rates may influence the ―community chest pain index‖. Our work is currently exploring the additional predictive capacity afforded to the models when these non- traditional yet potentially high yield dynamic contributors are also taken into account.
In the future, the ability to predict a ―high or low chest pain index‖ will enable evidence-based or ―computational model-based‖ staffing for many expensive hospital areas thus significantly reducing costs. Also new theories of individual human behavior will depend upon a solid understanding of the contextual ebb and flow of dynamic background group behaviors.
This work will help lay the ground work for Experience- based medicine which will be fundamental to supporting experience-based theories of human health behaviors at large.
R. Jin is with the Department of Computer Science, Kent State University, Kent, OH 44242 USA
W. Xiong is with the Department of Computer Science, Kent State University, Kent, OH 44242 USA
B. Piniewski is Chief Medical Officer with Peacehealth Laboratories USA
The authors of the following paper emphasize that this is a work in progress and is provided here as an example of the types of predictive insights and new business models available when laboratory systems partner with data analysts.
Download Article (.PDF):2014 Computational chest pain indexes FINAL