SCCM Resource Library-21-Trends in Vital Signs Predict Critical Deterioration in Pediatric Intensive Care Unit Transfers

Video Transcription

Hello, everyone, and welcome to this presentation. I wanted to thank SCCM for selecting this study for a Star Research Achievement Award.  My name is Anupmayan Bharat, and I'm a research assistant professor in the Department of Pediatrics at University of Chicago.  This talk is about how trends in vital signs can be used to predict critical deterioration in pediatric patients that have been transferred to the ICU.  The term critical deterioration was first coined by Christopher Bonafide in 2012, and as an important pediatric patient outcome measure.  It is defined by a direct ward to ICU transfer, where the patient goes on to experience non-invasive  ventilation, intubation, administration of vasopressors, or mortality within 12 hours of transfer.  Subsequent studies have shown that critical deterioration is associated with a higher risk of mobility and mortality.  It also has a significant financial impact, where critical deterioration has been associated with increased cost of up to $100,000 US for pediatric patients.  On the ward, there are several pediatric early warning scores that have been developed, from a Nagin score in 2005 to the Pew score and Bedside Pew score in 2011.  Recent work from our lab has also shown how electronic health record data can be combined with machine learning to identify patients at risk for deterioration on the ward.  However, all of these scores and models can be used to predict deterioration on the ward. In other words, they are used to predict the actual ICU transfer event.  Currently, there are no scores or models that can be used to predict critical deterioration. In other words, there are no models that say once a patient is transferred to the ICU,  who is at risk for experiencing a critical event. Thus, the objective of this study was to determine whether vital sign data can be utilized in  a machine learning framework to predict critical deterioration in patients that have been transferred to the pediatric ICU. Here are our study details.  We considered a prospective cohort for pediatric ward to ICU direct transfers at Comer Children's Hospital at University of Chicago Medicine from 2009 to 2018.  Within approximately 1,200 patients, 6% of patients went on to experience a critical deterioration event.  As predictors, we considered patient age, as well as the last recorded vital sign readings on the ward before ICU transfer.  We also considered vital sign trends over a 24-hour period prior to ICU transfer. We envision our model to be utilized at the point of ICU transfer.  In other words, once the decision has been made to transfer the patient to the ICU, we basically take vital sign recordings over a 24-hour period prior to the decision point.  For each vital sign, we calculate the mean, median, standard deviation, min and max values, the most recent change, as well as slope over the 24-hour look-back period.  The model utilizes this data to predict who is at risk for experiencing a critical event  once transferred to the ICU.  We developed a logistic regression model as well as a random forest machine learning model to predict critical deterioration.  We derived and validated these models using a cross-validation approach. We compared the performance of our models against a modified version of bedside views.  Briefly, we utilized the objective scoring elements within bedside views and calculated a score and back-calculated a score for a retrospective study period.  As a primary measure of discrimination, we utilized AUC and compared model performances with that of bedside views and against each other.  Here is a table 1 that describes the study population. As you can see, patients with critical deterioration did not differ from patients who did not experience  critical deterioration in terms of sex, age, race, ethnicity, as well as time spent in the ward before the actual ICU transfer.  Our study demonstrated that vital sign trends can be used to predict critical deterioration. Here is a table that shows the AUC for all our models.  Using the modified bedside views scoring scheme would give us an AUC of 0.64 for predicting critical deterioration.  Using the last observed vital sign was suboptimal, where the AUC decreased to 0.60 for predicting critical deterioration.  However, once we included trends from the prior 24-hour period, the AUC jumped from 0.60 to 0.69 for predicting critical deterioration.  Extending our mission from homologistic regression to a random forest framework further increased our AUC from 0.69 to 0.73.  Our study also shows that using 24-hours is vital to predicting critical deterioration.  Here is a plot that shows the model performances when using 12-hour trends and 24-hour trends.  As you can see, regardless of what model you use, for both the logistic regression as well as the random forest model, using 24-hour trends improved the performance of our models  in predicting critical deterioration as opposed to using 12-hour trends. As a final proof that trends are important for predicting critical deterioration, we  further looked at the top 10 variables most important for predicting critical deterioration by the random forest model.  As you can see, 9 of these 10 variables have something to do with trend, with age being the only non-trend predictor deemed important for predicting critical deterioration.  In conclusion, we show that trends and vital signs before ICU transfer can be used to predict critical deterioration.  We further show that it is important to use data from 24-hour period and not anything shorter. We show that our random forest model that utilized the 24-hour trends can be used to  predict critical deterioration, and this model outperformed a modified version of a bedside pew scoring scheme as well as a model that you only use as vital signs.  I would like to thank my co-authors, in particular I would like to thank my students Yasra and Cindy, my collaborators Nelson and Kyle, and my mentors Preeti and Matt.  I also would like to thank the NIH for funding this research. Thank you for listening to my talk.

Video Summary

The video presentation discusses a study on how trends in vital signs can be used to predict critical deterioration in pediatric patients transferred to the ICU. Critical deterioration refers to patients needing non-invasive ventilation, intubation, vasopressors, or experiencing mortality within 12 hours of transfer. The study analyzed data from a cohort of pediatric patients at Comer Children's Hospital and developed models using vital sign trends over a 24-hour period to predict critical deterioration. The random forest model outperformed existing scoring schemes, showing that trends in vital signs before ICU transfer can be used to predict critical deterioration in pediatric patients.

Asset Subtitle

Pediatrics, Quality and Patient Safety, 2021

Asset Caption

The Society of Critical Care Medicine's Critical Care Congress features internationally renowned faculty and content sessions highlighting the most up-to-date, evidence-based developments in critical care medicine. This is a presentation from the 2021 Critical Care Congress held virtually from January 31-February 12, 2021.

Anoop Mayampurath

Introduction/Hypothesis: Critical deterioration events, defined as transfer of a pediatric patient from the ward to the intensive care unit (ICU) followed by mechanical ventilation or vasopressor infusion within 12 hours, is associated with increased risk of in-hospital death. While scores exist to assess the risk of deterioration in a patient, there is currently no model that predicts the likelihood of a patient experiencing critical deterioration. Here, we utilize ward-level vital sign data recorded prior to ICU transfer to predict the risk of experiencing critical deterioration.

Methods: We conducted a retrospective study of all pediatric patients transferred from ward to ICU from 2009-2018 at the University of Chicago Comer Children's Hospital. The primary outcome was critical deterioration, defined as mechanical ventilation, vasopressor infusion, or mortality within 12 hours of ward to ICU transfer. Predictors include patient characteristics, last observed vital signs before ICU transfer as well as their prior 24-hours trends. We constructed a logistic regression (LR) and a random forest (RF) model to determine if vital signs and their prior 24-hour trends can predict risk of critical deterioration within 12 hours of ICU transfer. We used the area under the curve (AUC) metric to compare model performances.

Results: Our cohort had 1281 pediatric patients, among which 85 (6%) experienced a critical event after ICU transfer. Our RF model utilizing last observed vital signs and 24-hour trends predicted risk of experiencing a critical deterioration event better than (a) the LR model using only vital signs (AUC 0.72 vs. 0.61, P < 0.05), (b) the modified Bedside PEWS model utilizing physiological variables (AUC 0.72 vs. 0.64 P < 0.05), and (c) the LR model using vital signs and 24-hour trends (AUC 0.72 vs. 0.68, P < 0.05). Temperature change, mean temperature, and minimum diastolic blood pressure were important to predicting critical deterioration events.

Conclusions: We developed a random forest model that utilized trends in vital signs to determine the likelihood of a pediatric patient experiencing a critical deterioration event upon transfer to the ICU. Our model fulfills an important need in determining risk of developing critical deterioration patients who are transferred from ward to ICU.

Meta Tag

Content Type Presentation

Knowledge Area Pediatrics

Knowledge Area Quality and Patient Safety

Knowledge Level Intermediate

Knowledge Level Advanced

Membership Level Select

Tag Scoring Systems

Tag Mortality

Year 2021

Keywords

vital signs

critical deterioration

pediatric patients

ICU transfer

random forest model

	Society of Critical Care Medicine 500 Midway Drive Mount Prospect, IL 60056 USA Phone: +1 847 827-6888 Fax: +1 847 439-7226 Email: support@sccm.org	Contact Us About SCCM Newsroom Advertising & Sponsorship DONATE	MySCCM LearnICU Patients & Families Surviving Sepsis Campaign Critical Care Societies Collaborative	GET OUR NEWSLETTER
© Society of Critical Care Medicine. All rights reserved. \| Privacy Statement \| Terms & Conditions The Society of Critical Care Medicine, SCCM, and Critical Care Congress are registered trademarks of the Society of Critical Care Medicine.