Meta-Modeling for ICU Contamination Transmission Simulations: Using Smart Sampling and Machine Learning to Link Data to Simulation Parameters

• Activity Number: 600 - Less Can Be More: Smart Sampling in Data and Engineering Sciences
• Type: Topic Contributed
• Date/Time: Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Ben Haaland* and Damon Toth and Molly Leecaster
• Companies: University of Utah and University of Utah and University of Utah
• Keywords: Meta-model; Simulation; ICU contamination; Calibration

Abstract

ICU patients are especially susceptible to infection, and over time several successful interventions aimed at blocking transmission of pathogens, such as changes in decolonization and hand hygiene practices, have substantially reduced ICU infections. While the underlying stochastics and physics of the micro-systems involved in the overall transmission and contamination system are relatively well-understood, the underlying system parameters are difficult to estimate, and the overall dynamics of the system are not well-characterized. Accurate and computationally inexpensive models of ICU contamination transmission would enable examination of outcomes associated with proposed system manipulations relatively freely.

A meta-modeling and machine learning framework is proposed for linking pathogen transmission simulation models to actual data. We discuss a framework for constructing accurate, computationally efficient, and interpretable emulators for representing the distribution of simulation outcomes, given proposed input parameters and an efficient routine for estimating the underlying system parameters, with appropriate quantification of uncertainty.