0
Critical Care |

Automated Intensive Care Unit Mortality Prediction Using Approximate Motifs and SOFA Scores FREE TO VIEW

Sean McMillan, BS; Zeeshan Syed, PhD; Satinder Singh, PhD; Ilan Rubinfeld, PhD
Author and Funding Information

University of Michigan, Ann Arbor, MI


Chest. 2013;144(4_MeetingAbstracts):401A. doi:10.1378/chest.1705096
Text Size: A A A
Published online

Abstract

SESSION TITLE: Quality Improvement in the ICU I

SESSION TYPE: Original Investigation Slide

PRESENTED ON: Monday, October 28, 2013 at 01:45 PM - 03:15 PM

PURPOSE: Existing ICU scoring metrics do not rely upon minute-to-minute vital sign measurements over the course of the first 24 hours in the ICU and require the use of trained annotators.

METHODS: We analyzed 19,685 consecutive ICU admissions from January 2011 through November 2012 which met our inclusion criteria of having 24 hours worth of vital sign data analyzed at rate of once per minute. We used a locality-sensitive hashing (LSH) method to discover approximate patterns in the vital sign sequences of heart rate, respiratory rate, oxygen saturation, and systolic, diastolic, and arterial blood pressures. Motif frequency counts corresponding to motifs that were shown to have good discrimination through rank sum testing were used in conjunction with a patient's demographic information (age/sex/race) and sepsis-related organ failure assessment component scores as feature vectors. A support vector machine (SVM) with an area under receiver operating characteristic curve (AUROC) loss function was used to predict patient mortality.

RESULTS: Of the 19,685 ICU admissions analyzed there were 1,438 in-hospital deaths. Results were averaged over 10 runs of random two-thirds training, one-thirds testing patient cohorts. Our base model that just included the motif frequency counts had an average(standard deviation) AUROC of 0.753(0.007). The combination of our base model with a patient's demographic data yields an AUROC of 0.809(0.007). The base SOFA component scores as a feature vector had an AUROC of 0.808(0.006). The complete model using our approximate motifs and patient demographic data and SOFA component scores had an AUROC of 0.870(0.005).

CONCLUSIONS: Our model offers an interesting and completely automated alternative to existing scoring systems for the prediction of mortality based on the first full day in the ICU, while attaining AUROCs comparable to the best reported AUROCs for existing scoring systems.

CLINICAL IMPLICATIONS: The short vital sign patterns may potentially provide some clinical insight into physiological activity beyond what summary scores or data snapshots can.

DISCLOSURE: The following authors have nothing to disclose: Sean McMillan, Zeeshan Syed, Satinder Singh, Ilan Rubinfeld

No Product/Research Disclosure Information


Figures

Tables

References

NOTE:
Citing articles are presented as examples only. In non-demo SCM6 implementation, integration with CrossRef’s "Cited By" API will populate this tab (http://www.crossref.org/citedby.html).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

Find Similar Articles
CHEST Journal Articles
PubMed Articles
Pulmonary artery catheter. Best Pract Res Clin Anaesthesiol 2014;28(4):323-335.
A history of neonatal medicine in iran. Arch Iran Med 2014;17(12):855-61.
  • CHEST Journal
    Print ISSN: 0012-3692
    Online ISSN: 1931-3543