Causes of inaccurate thermodilution cardiac output measurements

The thermodilution measurement of cardiac output is notoriously error-prone, and the values generated by this method are only trustworthy if various conditions are met.  This came up in Question 19 from the second paper of 2007, and has never been asked about again, which suggests that CICM are moving away from this subject matter. However, one cannot exclude the possibility that some Swan Ganz fanboys among the examiners are biding their time, waiting to release a curveball at the Fellowship candidates. Moreover, it would appear that PA catheters are still in use around the world, even in this echosonography-obsessed age, and so conceivably CICM trainees and fellows might at some stage find themselves faced with thermodilution cardiac output measurements in their actual patients, agonising over whether or not they are reliable. 

The raw awesomeness of the PA catheter is discussed in greater detail elsewhere. For some theoretical background, thermodilution measurement of cardiac output by the PA catheter and the PiCCO device is described and critiqued in the CICM First Part revision section. For something even more theoretical, a discussion of the foundational principles behind indicator dilution measurement of blood flow is also available, irrelevant though it might be to the practical needs of the end-stage CICM trainee.

In summary:

One might notice that a lot of these sources of error have been eliminated by improvements in measurement technique and equipment, to say nothing of the near-total abandonment of invasive cardiac output monitoring in the Age of Echo. The reader will notice that most of the studies quoted below are from the early to mid-1980s. Modern devices and strategies have banished most of these concerns to the pages of exam papers and the hallways of major teaching institutions, where academic mastodons stomp and trumpet. For the rest of us, the contents of this grey box is probably enough, at least for a standard CICM question answer. If for whatever more detail is required, one may be referred to any number of excellent articles, such as Nishikawa & Dohi (1993) or Nadeau & Noble (1986).

Errors of technique

There's the right way and there's the wrong way of performing a thermodilution measurement, with the wrong way generally generating numbers which will not be reproducible and which can give rise to errors in clinical decisionmaking. Much of this, over decades, has been ameliorated by the advent of automated thermodilution catheters and standardised techniques.

• Temperature and volume of the injectate: this has to be standardised, and the same for every measurement. For that matter, the rate of infusion also matters. In short, too much injected cold stuff causes an underestimation of cardiac output, and too little injected cold stuff causes an overestimation of cardiac output. The reason is the AUC - the small volume produces a small curve, similar to what is seen with a normal or increased cardiac output.  The difference is not massive, and is not always in the direction described by the textbooks - for example Elkayam et al (1983) found that smaller volumes (5ml vs 10ml) understimated the cardiac output by a non-statistically-significant margin.  Apart from having to remain the same, this injectate should not be too cold, as the flush of icy fluid into the ventricle can cause a change in cardiac output, or even an arrhythmia. However, it should not be too warm, either; or rather you should know how warm it is. Rewarming the injectate (eg during transport to the bedside or during passage through the catheter) can introduce an error into calculation if the degree of warming is unpredictable. The most practical solution for this, as far as most authors recommend, is to use room temperature injectate.
• Rapid infusion of room temperature IV fluid occurring at the same time as the thermodilution measurement is likely to produce an error by introducing a random decrease in the ambient blood temperature which is impossible to correct for. This will generally underestimate the calculated cardiac output, potentially by a massive margin. Wetzel & Latson (1985) managed to drop a patient's apparent cardiac output from 6.0L/min down to 3.5 L/min by concurrently bolusing them with a manual hand-pumped fluid giving set. 
• Poor timing with respiratory cycle: by doing thermodilution measurements at different points in a normal respiratory cycle, you will end up with markedly different measurements. Stephens et al (1985) tried randomly diluting cold injectate into their PA catheter patients, and found that the standard deviation of CO measurements was about three times greater. Moral of the story: always perform the thermodilution measurement at end-expiration.
• Poor catheter position: not only wedge pressure measurements but also cardiac output measurements by the PA catheter rely on the tip being positioned in West's Zone 3. The reason for this is the need for the thermistor to be in a position where a decent amount of flow will be occurring at the time of thermodilution. To use a crude exaggeration, if the tip is in the middle of a big alveolar dead space, there will be no blood flow past the thermistor, and therefore none of the cold injectate will flow there either. 
• The PAC thermistor tip is up against the wall. If the thermistor is sensing the temperature of a pulmonary artery wall instead of the pulmonary arterial blood, the temperature change due to the passing of the injectate will not be fully appreciated, and the cardiac output measurement may be underestimated.
• The PiCCO thermistor is in a vessel which is alongside the injectate vessel. With transpulmonary thermodilution,  where the arterial catheter sits alongside the central venous line (eg. where both are in neighbouring femoral vessels), there can be "crosstalk" between the lines. The arterial catheter will detect the drop in local temperature which occurs due to the injection of cold fluid into the nearby venous line. The result is a preposterously low cardiac index and a "camel back" biphasic thermodilution curve (Keller et al, 2012).
• Concurrent use of electrocautery can be described as an error of technique, in the sense that it would be good technique to ask the surgeons to stop burning and cutting for like ten seconds while you measure the cardiac output (surely that's not too much to ask). Electrocautery can apparently interfere with the measurement of baseline pulmonary artery temperature. How exactly? Nadeau & Noble (1986), who are the only people who mention this, do not specifically say, nor do they have a reference to back this up ("electrical noise created by cautery" is blamed).

Technical errors

In contrast to the errors of technique, these are errors inherent in the measurement process. No matter how you standardise and control the technique, these errors will always be present. 

• Reproducibility: In laboratories, constant flow though various tubes or in instrumented animals yields excellent reproducibility, but obviously in critical care there are numerous factors which cannot be controlled for, and which give rise to random error. This is amplified by the number of directly measured and estimated variables in the Stewart-Hamilton equation. Owing to the inherent instability of the critical care environment,  Stetz et al (1982) recommended that an average of three measurements should be taken, and that only a change of 12-15% from the previous measurement average should be accepted as a genuine change.
• Thermistor sensitivity: As with all medical sensor equipment, the manufacturer must take a compromise between accuracy and cost. The need to mass manufacture vast quantities of such devices means that no individual sensor can possibly be perfectly accurate to within a million decimal points. Routine calibration testing may or may not be carried out, and the sensor may or may not drift from its calibration during transport and storage, adding to the possibility that each PA catheter and PiCCO sensor will have a unique and unpredictable personality.
• Thermistor sampling time:  Probably not so much a problem these days, but in days gone by the computational power of bedside electronics was so poor that only a finite number of measurements could be collected and processed per second. This would produce a jagged series rather than a smooth curve, adding an element of inaccuracy into the task of integrating the area under the temperature/time curve. 

Influence of cardiac abnormalities

• Extremely low cardiac output: low flow states can give rise to errors which overestimate cardiac output. van Grondelle et al (1983) found that cardiac output values less than 2.5L/min (i.e. a cardiac index of around 1.2-1.3) were associated with a big difference between the Fick and the thermodilution method (up to 35%). Of course, one might argue that this would be completely meaningless clinically, as you would still end up with a cardiac index of 1.75; i.e. nobody would be lulled into a false sense of security, thinking that their patient was perfectly fine. 
• Atrial fibrillation:  No so much a problem for the PA catheter measurements, atrial fibrillation (as well as IABP, for that matter) can interfere with cardiac output measurements estimated by pulse contour analysis on the basis of transpulmonary thermodilution. This is one of the limitations of the PiCCO device.
• Intracardiac shunt: Shunt will obviously interfere with thermodilution measurements. In the case of right-to-left shunt Freed & Keane (1978) found spuriously elevated cardiac output measurements, as the cold injectate rapidly escapes into the left side of the circulation without being measured, giving an impression of a faster pulmonary blood flow rate. In the case of a left-to-right shunt, the cardiac output monitoring software becomes terribly confused because the cold injectate circulates back into the right heart, giving a second peak to the thermodilution curve and increasing the area under it (as demonstrated in human subjects by Pearl & Siegel, 1991). Each scenario leads to the overestimation of the cardiac output. 
• Valvular regurgitation,  specifically tricuspid regurgitation, can allow some volume of cold injectate to be blown backwards out of the right ventricle in systole, thereby reducing the amount of injectate which flows past the pulmonary artery thermistor. The rest of the injectate will eventually catch up, but only slowly. The result will be an underestimated cardiac output. 

Extracardiac influences

These are unexpected or unaccounted-for changes in patient physiology which can create errors in the thermodilution measurement of cardiac output. This could be because they interfere in some way with the empirically derived (i.e. guessed) estimated variables plugged into the Stewart-Hamilton equation, or because they introduce an unexpected or unpredictable variation which derails the reliability of the measurement process. 

• An erratic respiratory pattern can influence thermodilution by randomly changing the delivery of venous preload to the right heart. Cardiac output is also erratic, so this is only a source of error insofar as the measurement of cardiac output during that specific instant will not be reflective of the average cardiac output over time.
• Unexpected haematocrit, eg. where the patient is anaemic or polycythaemic, is likely to produce an error because the specific heat and specific gravity of blood is incorporated into the Stewart-Hamilton equation (usually as the coefficient k₁). 
• Extremely small patients: the paediatric population is a variant of the aforementioned low flow scenario, except the flow is low because it is supposed to be low (as the patient is one-tenth the size of an adult, for example). Again, cardiac output will be overestimated; though apparently, you can restore accuracy by using a smaller (weight-adjusted) volume of injectate (Wyze et al, 1975).
• Hypothermia obviously interferes with the measurement of temperature as the patient's baseline temperature is lower than expected, which tends to decrease the amplitude of the curve. Or at least that's what happens when you use cold injectate (presumably, the opposite sort of error occurs if you use hot injectate in a hypothermic patient). Moreover, as the patient is being defrosted, the uptrend of the temperature can also produce errors in calculation. This is a very 1970s problem, as by 1980 Merrick et al established a correction factor to restore accuracy, and modern equipment makes all the adjustments in the background by measuring ambient temperature trends. 
• Pulmonary oedema presents a volume of fluid which interfaces closely with the pulmonary circulation and which can potentially act as an added mixing volume into which heat can be exchanged. This is in fact the intended effect in transpulmonary thermodilution, which uses this effect to measure extravascular lung water (EVLW). The PA catheter measurements should not be affected by this, unless there is a truly preposterous amount of pulmonary oedema. Nishikawa et al (1992) demonstrated that small amounts of alveolar water do not interfere with thermodilution measurements by PA catheter.