Session S24.6

Combining Multiple Cardiac Output Estimators to Reduce Error

OT Abdala*, T Chen, GD Clifford

Massachusetts Institute of Technology
Cambridge, MA, USA

Previous work explored the accuracy of cardiac output (CO) estimators on a database of 1,497 thermodilution measurements taken from 121 patients. In this work we applied linear regression to determine if a combination of CO estimators improves accuracy. We found that all individual estimators underestimated CO at low CO levels and overestimated at high CO levels. A linear combination of the tested techniques reduced the gross error from a MSE of 0.66 for the best individual estimator to 0.60 for the combined estimator. Most significantly, we see improvements in error at both high and low estimated CO levels. We also tested the performance of estimators as a function of the number of available calibration points. We found that the CO estimate was further improved by incorporating previous calibration values into each sequential estimate and that the Liljestrand estimator alone was the most accurate when calibrating with at least 13 thermodilution measurements. When less than 13 calibration values are available, our new combination of estimators provided the lowest error.

(Abstract Control Number: 282)