Low cardiac output syndrome (LCOS) occurs in the post-operative period after cardiac surgery and can affect 25% of patients (1, 2). LCOS results in an insufficient supply of oxygenated blood to the tissues, which leads to an increase in tissue oxygen extraction assessed by the oxygen extraction ratio (O₂ER). This hemodynamic impairment is associated with an increase in mortality, in duration of mechanical ventilation and PICU length of stay (1, 3). LCOS diagnosis is defined as fall of the cardiac index below 2 L/min per m², but invasive measurement of cardiac index is not usually performed in pediatric cardiac intensive care units (PCICU). Thus, the clinical diagnosis of LCOS mostly relies upon caregivers' qualitative assessment including skin warmth of the extremities and capillary refill time (1, 4).To improve LCOS diagnosis in PCICU, a clinical-biological LCOS score has recently been developed and was correlated with composite measures of postoperative morbidity (5). This score includes the qualitative assessment of recoloration time and the quantitative assessment of toe temperature. It was designed to predict adverse events potentially related to low cardiac output but has not been compared with an objective measure of cardiac output. Instead of toe temperature, several authors have proposed to use a thermal gradient between core and peripheral temperature with a threshold value of 5°C being correlated with decreased cardiac output and increased peripheral vascular resistance (6–8).

Skin warmth can be assessed qualitatively by caregiver touch or quantitatively via a skin thermometer or infrared thermography (IRT). IRT is a non-invasive, non-contact thermal measurement tool. The quantification of a thermal gradient by IRT should be more objective and reproducible than clinical assessment. The thermal resolution rises up to 0.02°C. This technology creates an image from the temperature obtained after the conversion of the infrared radiation emanating from a body or an object. IRT is used in many medical conditions (9–12) including assessment of local microcirculation in the extremities in the context of hemodynamic instability (13).

The use of the thermal gradient assessed by IRT to detect a decrease in cardiac output has not been extensively studied in PCICU. The aim of our pilot study was to assess the correlation between thermal gradient obtained by IRT and O₂ER in children aged 0 to 18 years, following cardiac surgery with cardiopulmonary bypass. We hypothesized that an increase in the thermal gradient obtained by IRT would be positively correlated with the increase in O₂ER after cardiac surgery, and therefore could be used for early detection of low cardiac output syndrome.

One hundred twenty-five patients were screened, 50 patients were included and had thermographic data acquisition. A total of 41patients were included in the analysis (Figure 1).

The demographic and baseline clinical characteristics are summarized in Table 1. The median age was 6 months (IQR = 3–47). There was no significant difference for sex. Cardiac malformations were mainly right ventricular outflow tract anomalies (n = 12), ventricular septal defect (n = 11) (see Table 1). The median PICU length of stay in our cohort was 5 days (IQR = 4–7), with a median duration of mechanical ventilation (MV) of 7 h (IQR = 4–23). Two patients developed a complication potentially related to LCOS such as necrosis enterocolitis. One patient required prolonged MV (2 832 h).

Four patients had vasopressors and 11 patients a vasodilator drug. The capillary refill time evaluated by the nurses was in 100% of the cases lower than 2 s.

The average lactate level was 2.1 mmol/L (±1.9) and the average cerebral rSO₂ was 68% (±12).

In the post-operative period following cardiac surgery, we collected thermal gradient measurements in 41 of 50 included patients. Despite a standardized protocol to acquire IRT, we identified several clinical situations that interfere with thermal gradient between the internal canthus of the eye and the hallux measurement including heated incubators and non-invasive ventilation. We found a weak correlation between the thermal gradient obtained by IRT and the O₂ER. This correlation was higher for patients with O₂ER values above 30%. The thermal gradient was not correlated with any clinical and biological variables that can be altered in the context of LCOS.

The use of thermal gradient as a tool for monitoring cardiac output is controversial. Older studies found an increase in thermal gradient correlated with an increase in systemic vascular resistance, a decrease in cardiac output, stroke index, and urine output (7, 18, 19). An increase in gradient preceded or was concomitant with changes in hemodynamic parameters (18). Such a correlation has also been found in the adult population (6). Conversely, several other authors did not find any clinically significant correlation (20–23). In most of these studies, thermodilution was used to measure cardiac output, more rarely doppler ultrasound was used (22). However, thermodilution is an invasive technique and is no longer routinely used to assess cardiac output in PICU. O₂ER is a well described marker of cardiac output (15).We found a weak correlation between O₂ER and the thermal gradient. Thus, as found in several studies, the thermal gradient correlated poorly with markers of cardiac output. In our study, the median O₂ER was 35%, only 11 O₂ER values were above 50% and 1 above 60%. Thus only a few patients had significant evidence of low cardiac output. For the group of highest O₂ER values, the correlation with thermal gradient increased from weak to medium, suggesting that the thermal gradient becomes more significant as the mechanisms for compensating decreased cardiac output begin to be overwhelmed and when more pronounced peripheral vasoconstriction occurs.

The thermal gradient measurement technique has not always been well described in the literature (6). Only in one study was the thermal measurement performed by infrared thermography. Our pilot study highlights the technical difficulties of using the infrared camera in an intensive care unit, especially the use of incubators and the numerous medical devices used on the face and hallux. Due to the discontinuous nature of blood gas sampling, we chose to perform discontinuous measurements of the thermal gradient. Those technical difficulties resulted in the exclusion of many photos (46%). Quality and consistency of the thermography imaging could be improved by the exclusion of patients with non-invasive ventilation, positioning the camera on the ceiling above the bed (to standardize patient distance and avoid interactions with nursing care), improvement of the spatial resolution of the camera and the automation of thermal picture capture.

There was no correlation between the thermal gradient and serum lactate, base excess, and urine output. Unlike the thermal gradient, these are all late markers of tissue hypoperfusion in the context of LCOS (15). Some authors suggest that global hemodynamic parameters may not be sensitive enough to reflect changes in peripheral blood flow in critically ill patients (24).

However, NIRS allows a real time evaluation of oxygenation and regional vascular perfusion, and a real time estimation of central venous saturation and cardiac output (25–27). This may explain why we found a moderate correlation between cerebral rSO₂ and the thermal gradient. In one patient with several assessments of the gradient over time, we observed a parallel evolution of the thermal gradient and cerebral rSO₂, suggesting that the follow-up of the thermal gradient could also give an information on the evolution of the hemodynamic pattern (Figure 4).

Younger patients had a higher thermal gradient. This may be due to greater severity of the cardiac lesions in those operated in the neonatal period. Indeed, neonatal cardiac surgeries often have a higher RACH-1 score and are therefore more at risk of postoperative hemodynamic instability. However, as most of the neonates were placed in a heated incubator, the number of neonates included in this study is too small to allow further analysis of the interaction between thermal gradient and younger age.

Ryan et al. found a statistically significant correlation between the thermal gradient (core temperature assessed by a pulmonary artery thermistor and the foot temperature assessed by thermography) and cardiac output assessed by thermodilution (21).They hypothesized that modifications in systemic vascular resistance with use of vasoactive drugs, for example, may impact this correlation. In our study, we did not find a significant relationship between vasoactive drugs used and the thermal gradient. Neither milrinone, known for its vasodilatory effects, or, on the contrary, vasoconstrictor agents, had a significant impact on the thermal gradient. However, a few patients had high doses of vasoactive agents and showed a trend towards a higher thermal gradient. Cardiopulmonary bypass can also modify systemic vascular resistance, especially in case of deep hypothermic circulatory arrest, but in our study, CPB duration had no influence on thermal gradient (28–30). In a recent study, we observed an inverse gradient in a patient with high dose of milrinone suggesting that the impact of the drug can be quantitatively assessed using temperature gradient by thermography (31).

This study has several strengths. As a single center study there was consistency in protocols for patient management that might not be observed across multiple units. Moreover, all the patients were operated by the same surgeon. We also used a standardized protocol to capture thermal pictures in agreement with medical thermology recommendations and the American Academy of Thermology guidelines. This was crucial to study the correlation between thermal gradient and O₂ER.

However, there were several limitations. The performance limitations of our thermal camera are the first limitation. The third generation thermal cameras used in the medical field must meet specific criteria (11). The cameras currently used have a thermal resolution of 0.05 °C. Our camera FLIR One Pro may not have been powerful enough, with a thermal sensitivity of 0.07 °C and a rather weak infrared and spatial resolution. We positioned the camera between 80 and 100 cm from the patient to compensate for this, but depending on the patient's age, the area of interest (internal epicanthus) was very small, with a risk of decreased precision of the temperature's measurement. This site has been described as having a good correlation with core temperature, but Teunissen et al. found that it was unreliable in cases of fever or in conditions of physical effort when compared with esophageal temperature (14). However, this was not the case in our study. The second limitation is due to sample size. We had estimated that we needed 120 thermal pictures (3 photos per patient), but several thermal pictures did not meet sufficient quality criteria to be analyzed. A third limitation was the small number of patients with low cardiac output i.e., high O₂ER. This was probably related to the exclusion of newborns weighing less than 3.5 kg. We excluded them because they were placed in heated incubators that were interfering with thermal recording. Indeed, only a few patients with cardiac malformations typically addressed surgically in the neonatal period, such as conotruncal lesions, or anomalies of the aortic arch were included in our study. However, neonates undergoing surgery for these lesions often have more hemodynamic instability postoperatively and thus higher O₂ER (32). Another issue related to small sample size was the inability to classify the thermal pictures according to the elapsed time after surgery. Since low cardiac output syndrome generally occurs between 6 h–18 h post-surgery, it would have been interesting to assess the evolution of the correlation between O₂ER and thermal gradient over time, but there were not enough thermal photos taken in order to do so. Finally, few patients had significant doses of vasoactive drugs, so we could not evaluate their impact on the thermal gradient.

In this pilot study, we determined thermal gradient in 41 pediatric patients following cardiac surgery. The data collection and image analysis barriers suggest that the image acquisition method needs to be improved before being clinically incorporated into noninvasive multimodal monitoring after cardiac surgery. As technology evolves, we plan to repeat the study using a more powerful camera that will be permanently installed at the patient's bedside. We will then be able to automatically monitor the evolution of the thermal gradient over time and according to changes in the patient's management, reflecting more accurately what is done in clinical practice.