JMIR Publications

ABSTRACT

Background:

Patient monitoring is vital in all stages of care. Especially, ICU Patient Monitoring (ICUPM) has the potential to reduce complications and morbidity and increase the quality of care.

Objective:

In this article, we propose a novel approach to enable hospitals to deliver higher quality, cost-effective patient care, and improve the quality of medical services in the Intensive Care Unit (ICU).

Methods:

We use Machine Learning (ML) algorithms along with statistical methods to predict ICU patients’ length of stay and mortality without prior knowledge about their medical conditions or diagnosis. Building accurate reliable predictive modeling using ML algorithms requires having access to representative features that can describe a model. The features used to build these models are minimal and rely on the most commonly used physiological variables that are often obtained inside and outside of hospitals (e.g., vital signs and lab values).

Results:

This study demonstrates how to achieve predictive modeling with minimal features while maintaining reasonable accuracy. The model’s accuracy for the mortality model is around 89% while the accuracy of the length of stay model, based on a population median ICU patient stay of 2.636 days, is around 65 %.

Conclusions:

The research shows that utilizing and processing time-series features helps in better understanding of the patient's condition. It will lead to enhance predictive modeling results.