How to Measure Cardiac Output: An Overview of Methods

Cardiac output, the volume of blood your heart pumps per minute, can be measured by more than a dozen techniques ranging from invasive catheter-based methods to completely hands-off wearable sensors. The three historically dominant approaches are the Fick method, dye dilution, and thermodilution, with thermodilution becoming the bedside standard after the pulmonary artery catheter gained popularity in the early 1970s.¹ Over the past few decades, though, non-invasive and less-invasive alternatives have multiplied, each trading some accuracy for safety and convenience.

The Fick Method

The Fick principle is conceptually the simplest approach. It says that the amount of oxygen your body consumes per minute equals the cardiac output multiplied by the difference in oxygen content between arterial blood (heading out to the body) and venous blood (returning to the heart). If you measure oxygen consumption and sample blood from an artery and from the pulmonary artery, you can solve for cardiac output with basic arithmetic. Because it relies on fundamental physiology rather than any particular technology, the Fick method has long been treated as a reference standard, especially in settings like heart-transplant eligibility evaluations where precision matters greatly.²

The catch is practicality. You need a pulmonary artery catheter in place to get a true mixed venous blood sample, and measuring oxygen consumption accurately requires either a metabolic cart with a mouthpiece or a hood system. That makes Fick measurements labor-intensive and hard to repeat frequently at the bedside. In many hospitals, clinicians use an “assumed” oxygen consumption based on the patient’s age, sex, and body size rather than measuring it directly, which introduces error. Despite these hassles, Fick remains the benchmark against which newer methods are compared.

Pulmonary Artery Thermodilution

Thermodilution is the method most clinicians think of first. A pulmonary artery catheter (often called a Swan-Ganz catheter) is floated through the right side of the heart until its tip sits in the pulmonary artery. A known volume of cold saline is injected through a port near the catheter’s base in the right atrium, and a thermistor at the tip records how the blood temperature changes over time. A larger temperature change that washes out slowly indicates low flow; a small, brief change indicates high flow. The area under that temperature-time curve gives you cardiac output.

The traditional version uses bolus injections, each one producing a single snapshot. A newer generation of catheters has a thermal filament that gently heats blood in a repeating pattern, allowing continuous measurement without repeated injections.³ Early validation work showed that continuous thermodilution correlated well with the bolus method across cardiac outputs ranging from about 2.8 to 10.8 liters per minute, with a correlation coefficient of 0.94 in a study of 54 ICU patients.⁴

Continuous thermodilution sounds like a clear upgrade, but the “continuous” label deserves a caveat. Because the thermal filament’s signal is small and requires averaging over time, the displayed number can lag behind reality during rapid hemodynamic swings. One study found that during graded hemorrhage and resuscitation, continuous cardiac output showed a delayed response compared with direct pulmonary blood flow measurement, only matching reliably at baseline and at the end of each phase.⁵ In other words, if the patient’s circulation is crashing or recovering minute to minute, the continuous number on the screen may be telling you where things were a few minutes ago, not where they are now.

Transpulmonary Thermodilution and Pulse Contour Analysis

Transpulmonary thermodilution takes the same cold-bolus concept but routes it differently. Instead of reading the temperature change in the pulmonary artery, a cold injection in a central vein is tracked by a thermistor on an arterial catheter, typically placed in the femoral or axillary artery. The indicator travels through the entire cardiopulmonary circuit before being detected, which means the technique also provides additional information about cardiac preload by estimating the volume of the four heart chambers.⁶

A major advantage of this setup is that the arterial catheter, once in place, can also perform pulse contour analysis. The shape of your arterial pressure waveform contains information about how much blood the heart is ejecting with each beat. By calibrating the pulse contour algorithm against a transpulmonary thermodilution measurement, devices like the PiCCO monitor can then track cardiac output beat by beat in real time, something a pulmonary artery catheter with bolus injections cannot do.⁷ That continuous pulse contour readout needs periodic recalibration, though, because arterial tone and compliance shift over time, especially with vasopressor drugs or evolving sepsis.

Several pulse contour algorithms exist that do not require external calibration at all. These “uncalibrated” systems estimate cardiac output purely from the arterial waveform’s shape using mathematical models of the vasculature, including Windkessel models and multi-beat analysis approaches.⁸ The trade-off is accuracy. In critically ill patients with impaired heart function, one prospective study found that an uncalibrated pulse contour device had a mean error of about 55% compared with pulmonary artery thermodilution, well above the commonly accepted threshold.⁹ Uncalibrated systems tend to perform better in hemodynamically stable patients and worse when vascular tone changes rapidly.

Echocardiography and Doppler Ultrasound

Ultrasound offers a completely non-invasive route to cardiac output. With transthoracic echocardiography, a clinician measures the diameter of the left ventricular outflow tract (the channel between the heart and the aorta) and uses Doppler to record the velocity of blood flowing through it. Multiplying the cross-sectional area by the velocity-time integral gives stroke volume, and multiplying stroke volume by heart rate gives cardiac output.¹⁰

This approach is genuinely useful at the bedside, especially in emergency departments and ICUs where quick hemodynamic assessments are needed without placing a catheter. The main limitation is operator dependence. Small errors in measuring the outflow tract diameter get squared when you calculate the area, so a millimeter off can shift the cardiac output estimate meaningfully. And you cannot leave the probe in place and walk away; each measurement is a snapshot that requires a skilled hand.

Esophageal Doppler addresses the “snapshot” problem by placing a thin ultrasound probe into the esophagus, where it sits next to the descending aorta and continuously tracks blood flow velocity. Changes in aortic blood flow tracked this way closely followed thermodilution-measured cardiac output both in animal models and in patients undergoing liver transplantation.¹¹ Its value lies in goal-directed fluid therapy during surgery: rather than guessing whether a patient needs more fluid, the surgical team watches the Doppler signal in real time and optimizes stroke volume. The limitation is that it measures only descending aortic flow and uses a correction factor to estimate total cardiac output, which introduces assumptions about how much blood goes to the head and arms.

Bioimpedance and Bioreactance

If ultrasound requires a skilled operator, thoracic electrical bioimpedance requires almost none. Electrodes placed on the chest send a tiny, high-frequency electrical current through the thorax and measure how the resistance (impedance) to that current changes with each heartbeat. When the aorta fills with blood during systole, thoracic impedance drops, and the magnitude of that change relates to stroke volume.¹² The appeal is obvious: stick some electrodes on, press start, and get beat-by-beat numbers with zero risk to the patient.

Bioreactance is a refinement of the same idea. Instead of looking at changes in the amplitude of the electrical signal, it analyzes shifts in the phase of the signal, which proponents argue is less sensitive to electrode placement and body composition.¹³ A prototype bioreactance device applied a 75-kHz current and derived cardiac output from the product of peak phase shift, heart rate, and ventricular ejection time.¹⁴

The accuracy picture for these technologies is mixed. In adults, the reported agreement with thermodilution is sometimes acceptable and sometimes not. One study comparing electrical cardiometry (a bioimpedance-family device) with continuous pulmonary artery thermodilution found the percentage error exceeded 30% at every measured time point.¹⁵ That 30% threshold is a widely used benchmark: if a new device’s percentage error against a reference method exceeds it, the agreement is generally considered insufficient for interchangeable clinical use.¹⁶ Still, even when absolute accuracy falls short, trending ability (detecting whether cardiac output is going up or down) can be clinically useful for guiding fluid and drug therapy.

Rebreathing Methods

The partial CO₂ rebreathing technique is an indirect, non-invasive application of the Fick principle. Instead of measuring oxygen consumption and sampling blood, it uses carbon dioxide. A rebreathing loop briefly causes you to re-inhale some of your exhaled CO₂, which changes the amount of CO₂ your lungs eliminate and the concentration of CO₂ at end-expiration. By comparing those values before and during the rebreathing period, the device calculates pulmonary capillary blood flow and then adds a shunt estimate to arrive at total cardiac output.¹⁷

The technique is limited to patients who are mechanically ventilated or who can cooperate with a mouthpiece, which excludes most awake, spontaneously breathing patients in the emergency department. It also relies on adequate gas exchange, so large intrapulmonary shunts or significant ventilation-perfusion mismatch can degrade accuracy. Pediatric applications have been explored, with devices adapted for children, though the smaller tidal volumes and higher respiratory rates in young patients add technical complexity.¹⁸

Cardiac MRI

Phase-contrast cardiac MRI can measure blood flow velocity in the aorta or pulmonary artery with high spatial resolution. In patients with pulmonary hypertension, cardiac output measured this way correlated well with invasive Fick measurements, and MRI actually contributed the least measurement variation among the compared techniques.² MRI-derived cardiac index has also been characterized across a range of healthy adults: in people aged 20–29, the average cardiac index was about 3.3 liters per minute per square meter of body surface area, declining gently to about 3.0 in those over 60. Athletes had the same cardiac index as age-matched healthy controls, but they got there with a lower heart rate and a higher stroke volume per beat. Patients with congestive heart failure, by contrast, averaged a cardiac index of about 2.3.¹⁹

Cardiac MRI is the closest thing medicine has to a true gold standard for volumetric heart measurements. But it requires a trip to the scanner, the patient must be able to hold still and follow breath-hold instructions, and the setup is far too expensive and slow for real-time hemodynamic monitoring. It is best suited for research, baseline characterization, and situations where precision matters more than speed.

When Thermodilution Measurements Go Wrong

No method works perfectly in every patient. Thermodilution, as the most widely used invasive technique, has well-cataloged pitfalls. The most clinically important is tricuspid regurgitation, where the valve between the right atrium and right ventricle leaks. When the cold indicator gets sloshed backward through a leaky tricuspid valve, it takes longer to wash out, and the resulting temperature curve is distorted. In patients with severe (third-degree) tricuspid regurgitation, the difference between thermodilution and Fick cardiac output averaged about 1.9 liters per minute, a large discrepancy in someone whose total output might only be four or five liters per minute.²⁰ The direction of the error is consistent: thermodilution underestimates true cardiac output when the tricuspid valve is regurgitant.²¹

Other sources of error in thermodilution include intracardiac shunts, where blood crosses between the left and right sides of the heart, and technical factors like the speed and volume of the saline injection. Bolus thermodilution is operator-dependent in subtle ways: injecting too slowly, using injectate that has warmed up, or timing the injection relative to the respiratory cycle can all shift the result. An animal-model study found systematic error of about 26% for pulmonary artery thermodilution compared with a direct aortic flow probe, with that error widening at higher cardiac outputs.²² Averaging three or more consecutive measurements helps, but clinicians pressed for time sometimes rely on a single injection, which adds noise.

How Validation Studies Are Judged

When you read that a new device “agrees well” with a reference method, the quality of that claim depends on how it was tested. The standard tool is the Bland-Altman plot, which graphs the average of two paired measurements against their difference. This reveals both the systematic bias (does the new method consistently read high or low?) and the spread of disagreement.¹⁶ Percentage error, defined as the limits of agreement divided by the mean cardiac output, is then compared against the 30% threshold. A device under 30% is considered interchangeable with the reference; above 30%, it may still be useful for trending but not for trusting the absolute number.

The problem is that the reference method itself (usually thermodilution) carries its own error. If thermodilution has a precision of roughly 10–15%, then any device compared against it starts with a built-in floor of disagreement that is not the new device’s fault. This means the 30% threshold is generous in theory but hard to meet in practice, especially in hemodynamically unstable patients where both methods are at their worst. A trending analysis, often done with four-quadrant plots that ask whether both methods moved in the same direction, can be more forgiving and arguably more clinically relevant. If your device reliably tells you that cardiac output just dropped, even if the exact number is off, you still have actionable information.

Measuring Cardiac Output in Children and Neonates

Small patients present unique challenges. Pulmonary artery catheters are risky and technically difficult to place in neonates, so non-invasive options are especially appealing. Echocardiography is the most common tool for assessing neonatal cardiac output, but it shares the same snapshot and operator-dependence problems as in adults, amplified by tiny anatomical structures.²³ Bioimpedance devices have been studied in children and infants, but the results are uneven. A systematic review and meta-analysis found that in older children and adolescents, bioimpedance percentage error approached acceptability (a median of about 31% for cardiac output), but in neonates and infants the error was higher, around 45%, placing it well outside the interchangeability threshold.²⁴

Part of the difficulty is that newborns, especially premature ones, have a fundamentally different cardiovascular system in the first days of life. Transitional shunts through the ductus arteriosus and foramen ovale mean that “cardiac output” is not a single number the way it is in an older child or adult; left and right ventricular outputs may differ substantially.²⁵ No monitoring technology perfectly accounts for that transitional physiology, which is one reason neonatal cardiac output monitoring remains an active area of research rather than a solved problem.

Finger Cuff Photoplethysmography

A small but growing family of devices estimates cardiac output from the arterial pressure waveform captured at the finger. These use a technique called the volume-clamp method: a tiny cuff around the finger inflates and deflates rapidly to keep the finger arteries at a constant volume, which lets the device reconstruct a continuous arterial pressure waveform without an arterial line.²⁶ From that waveform, pulse contour algorithms estimate stroke volume and cardiac output.

The appeal is obvious for settings like ambulatory monitoring, outpatient procedures, or research where placing an arterial catheter is not justified. Accuracy depends heavily on peripheral perfusion, though: cold fingers, vasoconstriction from shock, or heavy vasopressor use can all degrade the signal. These devices are increasingly validated in specific clinical niches but are not yet standard tools for high-acuity ICU monitoring.

Cardiac Output During Exercise Testing

Measuring cardiac output during exercise poses special problems because the patient is moving, breathing heavily, and pushing flow rates far above resting values. Impedance cardiography has been evaluated in this setting and performs surprisingly well. In one study comparing impedance cardiography with simultaneous direct Fick measurements across outputs ranging from 3.5 to 18 liters per minute, no systematic error was found, and random error stayed under 5% per subject. Average maximal cardiac output in those subjects was about 27 liters per minute.²⁷ A newer impedance device showed similarly good reproducibility across repeated maximal exercise tests, with a correlation of 0.94 against Fick.²⁸

Foreign gas rebreathing using acetylene has also been recognized as meeting the criteria for being non-invasive, easy to use, reliable, and valid during maximal exercise.²⁹ In sports science and exercise physiology labs, these non-invasive methods have largely replaced catheter-based techniques for routine testing, reserving the Fick method for research protocols where the highest possible accuracy justifies the invasiveness. Bioimpedance devices designed for exercise are also used to assess children’s cardiovascular fitness, where invasive methods are ethically difficult to justify.³⁰

Wearable Sensors and Deep Learning

The frontier of cardiac output monitoring is moving toward wearable devices paired with machine learning. A recent study tested a lightweight deep learning model that fused single-lead ECG data from a chest patch with finger-based pulse oximetry signals to predict cardiac output in cardiac surgery patients. The model was trained against pulmonary artery catheter measurements, and the researchers found that normalizing output to cardiac index (cardiac output divided by body surface area) improved agreement with the catheter reference.³¹ This is still early-stage work validated in a specific surgical population, but it points toward a future where a patch on your chest and a clip on your finger could continuously estimate how much blood your heart is pumping, no catheter required.

Whether wearable-derived cardiac output will be accurate enough to guide critical clinical decisions remains to be seen. The signal quality from consumer-grade sensors is inherently noisier than from hospital monitors, and the models will need validation across a wide range of body types, disease states, and activity levels before they can be trusted outside research settings. For now, they represent the most ambitious end of a decades-long trend: steadily moving cardiac output monitoring from the catheterization lab to the bedside to, eventually, the patient’s own body.