Learning and Model-Based Approaches to Improved Patient Monitoring, Assessment and Treatment in Capnography and Procedural Sedation

SHARE:

Event Speaker: 

Rebecca J. Mieloszyk

Event Location: 

36-428 (Haus Room)

Event Date/Time: 

Thursday, August 11, 2016 - 10:30am

Full Title of Doctoral Thesis: Learning and Model-Based Approaches to Improved Patient Monitoring, Assessment and Treatment in Capnography and Procedural Sedation

Abstract:

Capnography is the continuous measurement of the partial pressure of carbon dioxide in exhaled air, through a non-invasive and effort-independent process. This monitoring modality is highly informative, with the resulting capnogram rapidly reflecting changes in ventilation, perfusion, and metabolism. In current practice, however, clinicians assess the capnogram through chiefly qualitative, visual means that only capture a small fraction of the information embedded in the measurement. This thesis provides an objective, quantitative methodology for extracting information from the time-based capnogram. The framework includes automated preprocessing, segmenting of the capnogram, feature extraction, classification through an ensemble of voters when labeled training data are available, and clustering in the absence of labeled data. We develop and apply this framework to characterize the capnogram in healthy subjects, to provide diagnostic information in disease conditions, and to assess patient state during procedural sedation.

Procedural sedation refers to the use of sedation agents such as propofol to relieve pain and anxiety associated with diagnostic and therapeutic procedures – for example, lumbar puncture, colonoscopy, and fracture reduction – that are conducted by a diverse group of specialists outside the operating room. There is no objective measure of sedation level to guide the titration of medications, leading to the risk of under- or over-sedation. Drug titration is currently guided by visual observation, basic vital sign monitoring, and response to stimulus, which are inadequate surrogates for sedation level. Capnography is the earliest indicator of the respiratory compromise that can occur with over-sedation. We conduct online clustering of features extracted from the capnograms of patients undergoing procedural sedation and find separable clusters, or states, whose transition times are aligned with those of relevant clinical events. Additionally, through pharmacokinetic modeling, we compute continuous estimates of drug plasma concentrations over the course of the procedure. The outputs of the feature-based capnogram analysis and of the pharmacokinetic models provide continuous patient assessments and can allow for more quantitatively guided drug titration.

Thesis Supervisors: Profs. George C. Verghese, Thomas Heldt