Skip to Content

Second Phase of a Study Aimed at Predicting Patient Instability Receives Funding

Fri, 10/07/2016

When a patient is moved from an intensive care unit (ICU) to a step-down unit (SDU) that provides an intermediate level of care, patients’ vital signs are continuously monitored by nurses for any sign of cardiorespiratory insufficiency, or CRI. Data has shown that CRI is difficult to detect, however, and even more difficult to anticipate. Moreover, even though nurses monitor the vital signs of all patients in SDUs for signs of CRI, fully three-fourths of patients will not become unstable, meaning that resources and life-saving interventions are often not delivered as quickly as they could be to at-risk patients.

Michael Pinsky, MD, CM, Dr hc Marilyn Hravnak, PhD, RN

What if, however, it were possible to improve the intelligence of the detection technology, to the extent that nurses would not only be notified when CRI is occurring in real time, but also when and in which patients it may occur in the future? Such technology could not only have important implications for reducing preventable mortality, but could also eliminate alarm fatigue, and improve the quality of nursing care and the efficiency of treatment delivery systems.

Those are the goals of a project called Predicting Patient Instability Noninvasively for Nursing Care-Two (PPINNC-2), co-led by Marilyn Hravnak, PhD, RN, and Michael Pinsky, MD, CM, Dr hc, which was recently funded by the National Institute of Nursing Research (NINR), a part of the National Institutes of Health (NIH).

PPINNC-2 builds off of an earlier study in which Hravnak, a Professor in the Acute and Tertiary Care Department of the University of Pittsburgh School of Nursing, and Pinsky, Professor of Critical Care Medicine, Bioengineering, Anesthesiology, Cardiovascular Diseases, and Clinical & Translational Sciences, developed a prototype of a smart clinical decision support system (SDSS) that was able to identify with high sensitivity and specificity real-time instances of CRI (“nowcasting”), as well as potentially “forecast” future occurrences. This model was also importantly able to discriminate between actual CRI and artifacts, or errors, which occurred during data processing.

To develop this model, Hravnak, Pinsky, and their co-investigators assembled and mined a set of data that had been gathered from a step-down unit over the course of a year, and made use of Machine Learning (ML), a technique in which an algorithm can process data and detect patterns so as to learn and even make predictions.

In this iteration of the study, the researchers will be able to expand on these results and focus on strengthening the model’s ability to predict CRI. To do this, the researchers will assemble a wider data set to strengthen the model’s capacity for detection, and will test the model’s prediction capabilities in a pilot study in an actual SDU.

Co-investigators on the project (R01NR01391) include Gilles Clermont, MD, MSc, Professor of Critical Care Medicine, Industrial Engineering, Mathematics, Chemical Engineering, and Clinical and Translational Science, and Artur Dubrawski, PhD, MSc, Director of the Auton Lab at Carnegie Mellon University. The project will run until June 2020.