A Fast and Accurate Diagnostic Test for Severe Sepsis Using Kernel Classifiers
dc.contributor.author | Parente, J.D. | |
dc.contributor.author | Lee, D.S. | |
dc.contributor.author | Lin, J. | |
dc.contributor.author | Chase, Geoff | |
dc.contributor.author | Shaw, Geoff | |
dc.date.accessioned | 2011-05-09T22:01:45Z | |
dc.date.available | 2011-05-09T22:01:45Z | |
dc.date.issued | 2010 | en |
dc.description.abstract | Severe sepsis occurs frequently in the intensive care unit (ICU) and is a leading cause of admission, mortality, and cost. Treatment guidelines recommend early intervention, however gold standard blood culture test results may return in up to 48 hours. Insulin sensitivity (SI) is known to decrease with worsening condition and inflammatory response, and could thus be used to aid clinical treatment decisions. Some glycemic control protocols are able to accurately identify SI in real-time. A biomarker for severe sepsis was developed from retrospective SI and concurrent temperature, heart rate, respiratory rate, blood pressure, and SIRS score from 36 adult patients with sepsis. Patients were identified as having sepsis based on a clinically validated sepsis score (ss) of 2 or higher (ss = 0–4 for increasing severity). Kernel density estimates were used for the development of joint probability density profiles for ss = 2 and ss < 2 data hours (213 and 5858 respectively of 6071 total hours) and for classification. From the receiver operator characteristic (ROC) curve, the optimal probability cutoff values for classification were determined for in-sample and out-of-sample estimates. A biomarker including concurrent insulin sensitivity and clinical data for the diagnosis of severe sepsis (ss = 2) achieves 69–94% sensitivity, 75–94% specificity, 0.78–0.99 AUC, 3–17 LHR+, 0.06–0.4 LHR-, 9–38% PPV, 99–100% NPV, and a diagnostic odds ratio of 7–260 for optimal probability cutoff values of 0.32 and 0.27 for in-sample and out-of-sample data, respectively. The overall result lies between these minimum and maximum error bounds. Thus, the clinical biomarker shows good to high accuracy and may provide useful information as a real-time diagnostic test for severe sepsis. | en |
dc.identifier.citation | Parente, J.D., Lee, D.S., Lin, J., Chase, J.G., Shaw, G.M. (2010) A Fast and Accurate Diagnostic Test for Severe Sepsis Using Kernel Classifiers. Coventry, UK: UKACC International Conference on CONTROL 2010, 7-10 Sep 2010. | en |
dc.identifier.uri | http://hdl.handle.net/10092/5176 | |
dc.language.iso | en | |
dc.publisher | University of Canterbury. Mathematics and Statistics | en |
dc.publisher | University of Canterbury. Mechanical Engineering | en |
dc.rights.uri | https://hdl.handle.net/10092/17651 | en |
dc.subject | sepsis | en |
dc.subject | insulin sensitivity | en |
dc.subject | model-based control | en |
dc.subject | non-parametric | en |
dc.subject | classification | en |
dc.subject | characteristic curves | en |
dc.subject | discrimination | en |
dc.subject | likelihood | en |
dc.subject | accuracy | en |
dc.subject | decision support systems | en |
dc.subject.anzsrc | Fields of Research::40 - Engineering::4003 - Biomedical engineering::400308 - Medical devices | en |
dc.subject.anzsrc | Fields of Research::32 - Biomedical and clinical sciences::3202 - Clinical sciences::320212 - Intensive care | en |
dc.title | A Fast and Accurate Diagnostic Test for Severe Sepsis Using Kernel Classifiers | en |