Date Awarded


Document Type


Degree Name

Doctor of Philosophy (Ph.D.)




John B. Delos

Committee Member

David Armstrong

Committee Member

Dennis M. Manos

Committee Member

Leah Shaw


In the intensive care unit (ICU) of a hospital, patients are monitored continuously and the data on those patients provide powerful diagnostic tools for the medical community. However, the patient data creates incredibly large data sets with instruments measuring multiple signals simultaneously. This work seeks to improve monitoring techniques through analysis of large data sets from former ICU patients. By knowing the outcomes of patients in the past, can we detect patterns to diagnose future patients while also reducing the amount of recorded information? This thesis first seeks to improve methods of storing infant electrocardiograms (EKGs) by reducing the full EKG signal to only a vector of timestamps between heart beats. This work then focuses on improvements to estimating adult cardiac output (CO) from radial blood pressure waveforms. CO is estimated using nine previously proposed algorithms applied to radial blood pressure waveforms and applied to aortic waveforms estimated using three transfer functions. Results are compared with 3966 thermodilution measurements for 440 patients. Predictions based on the various algorithms are combined using linear regression and reduced linear regression methods. The method with the highest correlation coefficient was the Liljestrand method with the general transfer function waveform, with a mean squared error (MSE) of 0.50 (L/min)2. The MSE when all predictors are used is 0.44 (L/min)2, and the MSE after elastic net reduction is 0.49 (L/min)2. The reduced model is then applied to patients that have received a blood transfusion, and it is shown that the model has potential to detect a change in a patient’s Stroke Volume leading up to the time a doctor has ordered a blood transfusion.




© The Author

Included in

Physics Commons