1 Introduction
2 How is an indication obtained?
2.1 Sensing
2.2 Signal conditioning
2.2.1 Amplification
2.2.2 Filtering
2.2.3 Linearization
2.2.4 Analog-to-digital conversion
2.3 Calibration
2.4 Averaging
3 Validation procedures
3.1 Reference measurement procedure
3.2 Constancy of the measurand
3.3 The test device: estimation of measurements properties
3.3.1 Instrumental precision
3.3.2 Instrumental bias
3.3.3 Sensitivity
3.3.4 Uncertainty
3.3.5 Stability
4 Practical considerations for validations procedures
4.1 Instrumental precision of test method (random error)
-
On a unique patient in a steady state, collect as much as possible repeated indications using the test method during a short period of time. Since during this short period of time, the true cardiac output is supposed to be relatively constant, the 2σ and 2σ/µ of the test method indications estimate the random error of measurement (repeatability). However, even in clinically stable patients, the true value of any hemodynamic variable is never completely constant over several cardiac and respiratory cycles. Therefore, from a strict statistical point of view, we have to deal with repeated measurements of multiple (changing) true values and not with repeated measurements of a single (constant) true value [3].
-
Then, restart the same process with different patients for deriving the inter-patient random error as mean 2σ’ and mean 2σ’/µ’ (intermediate precision).
-
Then, in theory, one may restart with different operators and different devices for deriving the inter-operator, and inter-device random error as mean 2σ’’ and mean 2σ’’/µ’’ (reproducibility).
4.2 Instrumental bias of test method (systematic error)
-
Choose a reference method with a 2SEMref ideally < 2σ/4. For example, if the test method precision obtained in step 1 is 2σ/µ = 16%, the reference method should reach at least a SEMref of 4%. Hence, if a continuous thermodilution method with 2σ/µref = 20% is chosen as reference method for measuring cardiac output, then 25 measurements (indications) should be obtained (since 20/\(\sqrt{25}\) = 4). If a bolus thermodilution method with 2σ/µref = 10% is chosen as reference method, then 6 measurements should be obtained (since 10/\(\sqrt{6}\) = 4). If a steady state cannot be obtained during the period of time necessary to collect the appropriate number indications, another reference method must be found.
-
Compare the test method indications and the reference method indications for at least one low, one medium, and one high value of cardiac output. Then derive the slope and compare it with the identity line (sensibility and linearity).
-
Repeat this procedure in a series of different patients to derive the mean bias, mean sensitivity, and mean linearity and their variabilities.
4.3 Step response time
-
Choose a reference method with a fast step response time for measuring cardiac output (not necessarily the same method used for point 2). Indeed, the reference method is not only specific of the measurand but also of the quality criteria. Therefore, any hemodynamic variable closely linked to cardiac output with a very good step response time can be used to study the time delay of a cardiac output indication, if physiologically linked, for example invasive arterial blood pressure. Of course, during the short time of the experiment, other components of the blood pressure must be constant.
-
Choose as start time the sudden change in blood pressure following a sudden hemodynamic intervention (for example a lung recruitment test with high level of PEEP). This change is also indicative of the cardiac output change.
-
Measure the time delay between the change in blood pressure and the change of the test method indication.
-
Repeat this in a series of different patients to derive the mean time and variability.
4.4 Stability of precision, trueness, and step response time
-
Restart 2), 3) and 4) during a specified period of time to evidence an eventual drift in the quality criteria mentioned above.