Oxygenation and pulmonary mechanics in critically ill children receiving mechanical ventilation: a computational phenotype and machine learning approach.

Permanent URL: http://hdl.handle.net/2047/D20316411
Title:
Oxygenation and pulmonary mechanics in critically ill children receiving mechanical ventilation : a computational phenotype and machine learning approach
Creator:
Smallwood, Craig David (Author)
Contributor:
Gouldstone, Andrew (Thesis advisor)
Ruberti, Jeffrey (Committee member)
Kamarthi, Sagar (Committee member)
Language:
English
Publisher:
Boston, Massachusetts : Northeastern University, 2018
Date Accepted:
April 2018
Date Awarded:
May 2018
Type of resource:
Text
Genre:
Dissertations
Format:
electronic
1 unnumbered page, ix, 142 pages
Digital origin:
born digital
Abstract/Description:
Mechanical ventilation (MV) is a lifesaving therapy applied to critically ill children. A component of MV, positive end-expiratory pressure (PEEP) is often increased to improve ventilation efficiency thereby improving gas exchange. PEEP may ameliorate or exacerbate lung injury. PEEP optimization is clinically important and predicting physiologic response is desirable. Despite this, there is a paucity of literature to guide the clinician at the bedside. Importantly, time-series physiologic data are available for MV patients in the pediatric intensive care unit. However, these data have not been adequately explored in the literature. Therefore, we sought to 1) quantify the time required for oxygenation and pulmonary system compliance changes in children requiring mechanical ventilation, 2) quantify the empirical probability of PEEP changes implemented by expert clinicians that result in positive effects on oxygenation, pulmonary mechanics and dead-space fraction and identify clinical features associated with positive response and 3) extract computational phenotypes from time-series physiologic data and develop a model to predict PEEP response during MV in children. Equilibration time for oxygenation and compliance was computed. PEEP changes were quantified as either a responder (condition improvement) or a non-responders (condition worsening) and the empirical probability of a clinician implementing a change that improved the patient's condition within was quantified. Features from continuous mechanical ventilation variables were extracted and used to train a model in order to predict improvements in ventilation efficiency subsequent to changes in PEEP. A total of 265 subjects were enrolled and 1327 PEEP change cases were analyzed. Equilibration requires 38 and 71 minutes for respiratory system compliance and oxygenation respectively; the latter was directly observed to be dependent upon severity of illness. The empirical probability of improving a patient's condition following a PEEP change by clinicians was ~59% and ~48% for PEEP increases and decreases respectively. Responders to increased PEEP had higher oxygen requirements and ventilator support. A total of 27 computable phenotypes were identified and incorporated into a prediction model. These phenotypes incorporated features that are not typically assessed by clinicians at the bedside. The area under the receiver operator characteristic curve was 0.82 and 0.90 for classifying response to PEEP increases and decreases respectively. In the future, these methods may play an important role in optimizing care during pediatric MV.--Author's abstract
Subjects and keywords:
Artificial respiration -- Research
Respiratory therapy for children -- Physiological aspects -- Data processing
Lungs -- Wounds and injuries -- Treatment -- Methodology
Bioinformatics
computational phenotype
medicine
pediatric
respiratory failure
ventilation
DOI:
https://doi.org/10.17760/D20316411
Permanent URL:
http://hdl.handle.net/2047/D20316411
Use and reproduction:
In Copyright: This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the right-holder(s). (http://rightsstatements.org/vocab/InC/1.0/)
Copyright restrictions may apply.

Downloads