Method comparison – a new approach to implementing the Bland–Altman analysis to estimate the precision of a new method: tested on 30 critically ill patients monitored with pulse pressure analysis and continuous cardiac output vs intermittent thermodilution

Introduction

The Bland–Altman plot is the standard way of determining agreement between two methods of measuring the same variable. Following work by Critchley and Critchley [1], ± 30% is now accepted as the cutoff point when calculating the percentage error. This study estimates the coefficient of variation (CV) of three different cardiac output (CO) techniques with the aim of assessing the relative contributions to the percentage error.

Materials and methods

Thirty critically ill patients had their CO measured every hour with continuous cardiac output by Vigilance (CCO), pulse pressure analysis by LiDCO™plus (PulseCO) and intermittent thermodilution (ITD) (average of four ITD curves). Data were analysed with Bland–Altman plots, calculation of the percentage error, determination of the CV of ITD, and calculation of the overall CV for CCO and PulseCO.

Results

Two hundred and forty (eight per patient) measurements of CO were obtained. CCO vs ITD had an overall bias (± 2SD) of 0.2 ± 2.4 l/min (error 31%), mean CO (ITD + CCO) 7.7 l/min. PulseCO vs ITD had an overall bias of -0.1 ± 2.4 l/min, mean CO (PulseCO + ITD) 7.5 l/min (error 33%). According to the above criteria (without measuring the CV for ITD), CCO performed well when compared with ITD (31%) but PulseCO (33%) was outside clinically acceptable levels of agreement. The CV for a single ITD CO measurement was 15%, and this decreased to 7.5% when averaging four thermodilution curves. Using the CV for ITD of 7.5%, the relative CVs for the CCO and PulseCO were determined. The CV for CCO was 13.6% and for PulseCO was 14.7%.

Conclusion

In trying to understand the relative contributions of error when testing two techniques to measure the same variable it is vital to understand the CV of the reference technique. Using this approach, both the PulseCO and Vigilance perform in a clinically acceptable fashion.