Show simple item record

dc.contributor.advisorRoger G. Mark.en_US
dc.contributor.authorShavdia, Dewangen_US
dc.contributor.otherHarvard University--MIT Division of Health Sciences and Technology.en_US
dc.date.accessioned2008-09-03T15:23:00Z
dc.date.available2008-09-03T15:23:00Z
dc.date.copyright2007en_US
dc.date.issued2007en_US
dc.identifier.urihttp://hdl.handle.net/1721.1/42338
dc.descriptionIncludes bibliographical references (leaves 87-89).en_US
dc.descriptionThesis (M. Eng.)--Harvard-MIT Division of Health Sciences and Technology, 2007.en_US
dc.description.abstract(cont.) The EWS models were then tested in a forward, casual manner on a random cohort of 500 ICU patients to mimic the patients' stay in the unit. The model with the highest performance achieved a sensitivity of 0.85 and a positive predictive value (PPV) of 0.70. Of the 35 episodes of hypotension despite fluid resuscitation present in the random patient dataset, the model provided early warnings for 29 episodes with a mean early warning time of 582 ± 355 minutes.en_US
dc.description.abstractEarly goal-directed therapy (EGDT) in severe sepsis and septic shock has shown to provide substantial benefits in patient outcomes. However, these preventive therapeutic interventions are contingent upon an early detection or suspicion of the underlying septic etiology. Detection of sepsis in the early stages can be difficult, as the initial pathogenesis can occur while the patient is still displaying normal vital signs. This study focuses on developing an early warning system (EWS) to provide clinicians with a forewarning of an impending hypotensive crisis-thus allowing for EGDT intervention. Research was completed in three main stages: (1) generating an annotated septic shock dataset, (2) constructing multivariate logistic regression EWS models using the annotated dataset, and (3) testing the EWS models in a forward, causal manner on a random cohort of patients to simulate performance in a real-life ICU setting. The annotated septic shock dataset was created using the Multi-parameter Intelligent Monitoring for Intensive Care II (MIMIC II) database. Automated pre-annotations were generated using search criteria designed to identify two patient types: (1) sepsis patients who do not progress to septic shock, and (2) sepsis patient who progress to septic shock. Currently, manual review by expert clinicians to verify the pre-annotations has not been completed. Six separate EWS models were constructed using the annotated septic shock dataset. The multivariate logistic regression EWS models were trained to differentiate between 107 high-risk sepsis patients of whom 39 experienced a hypotensive crisis and 68 who remained stable. The models were tested using 7-fold cross validation; the mean area under the receiver operating characteristic (ROC) curve for the best model was 0.940 ± 0.038.en_US
dc.description.statementofresponsibilityby Dewang Shavdia.en_US
dc.format.extent89 leavesen_US
dc.language.isoengen_US
dc.publisherMassachusetts Institute of Technologyen_US
dc.rightsM.I.T. theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. See provided URL for inquiries about permission.en_US
dc.rights.urihttp://dspace.mit.edu/handle/1721.1/7582en_US
dc.subjectHarvard University--MIT Division of Health Sciences and Technology.en_US
dc.titleSeptic shock : providing early warnings through multivariate logistic regression modelsen_US
dc.typeThesisen_US
dc.description.degreeM.Eng.en_US
dc.contributor.departmentHarvard University--MIT Division of Health Sciences and Technology.en_US
dc.contributor.departmentHarvard University--MIT Division of Health Sciences and Technology
dc.identifier.oclc233639098en_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record