Near-infrared Spectroscopy and Skin Tone in Children: A Prospective Cohort Study

Abstract

Background:

Retrospective studies suggest that pulse oximetry overestimates saturation in children from races that may be associated with darker skin tone. Near-infrared spectroscopy (NIRS) relies on similar optical technology, but less is known about the effect of skin tone on NIRS. This study aimed to quantify the effect of skin tone on NIRS performance.

Methods:

Consecutive patients under 21 yr old undergoing cardiac catheterization were enrolled (N = 110). Skin tone was measured using spectrophotometry. Regional oxygen saturation was recorded from a Medtronic (USA) INVOS 5100C NIRS device placed on the forehead and was compared to the mixed venous saturation. Multivariable linear regressions were used to determine the effect of skin tone measured by individual typology angle (ITA).

Results:

Mean bias was larger for patients with darker skin in ITA categories 5 and 6 at −12.8% compared to ITA categories 3 and 4 at −2.5% with a difference of 10.3% (95% CI, 4.4 to 16.3; P < 0.001) and ITA categories 1 and 2 at 0.3% with a difference of 13.1% (95% CI, 7.5 to 18.7; P < 0.001). ITA was associated with NIRS bias in multivariable regression analysis (coefficient, 0.173; P < 0.001).

Conclusions:

Darker skin tone is associated with worse NIRS performance and lower NIRS values compared to mixed venous saturation for the INVOS 5100C system. This may lead to differences in care and contribute to inequities in outcomes. Better validation guidelines are needed to ensure equity in performance across varying skin tones.

When using the Medtronic (USA) INVOS 5100C system in children, darker skin tone is associated with worse near-infrared spectroscopy performance and lower near-infrared spectroscopy values compared to mixed venous saturation.

Editor’s Perspective

What We Already Know about This Topic

• Pulse oximetry overestimates saturation in children from races that may be associated with darker skin tone

• Near-infrared spectroscopy (NIRS) relies on similar optical technology, but less is known about the effect of skin tone on NIRS

What This Article Tells Us That Is New

• When using the Medtronic (USA) INVOS 5100C system in children, darker skin tone is associated with worse NIRS performance and lower NIRS values compared to mixed venous saturation

Near-infrared spectroscopy (NIRS) is a noninvasive technology for measuring regional oxygen saturation. It is used widely in pediatrics to monitor cerebral tissue oxygenation before, during, and after cardiac surgery¹ and to monitor brain² and organ tissue oxygenation³ in intensive care settings. Similar to pulse oximetry, it uses the differential absorption of light by oxygenated and deoxygenated hemoglobin to quantify saturation.

NIRS is often used as a noninvasive surrogate for mixed venous oxygen saturation (Svo₂). Both low absolute values and decrease from baseline have associated with worse outcomes, although there is variation in the results of individual studies.⁴ Importantly, NIRS measures regional saturation below the probe and not direct Svo₂. There is no accepted gold standard measurement for regional saturation. NIRS validation studies have used a variety of measurements as the gold standard.⁵ Pediatric studies have used both Svo₂^6,7 and a weighted average of Svo₂ and arterial oxygen saturation (Sao₂).⁵ Compounds with similar absorption to hemoglobin may interfere with NIRS measurements, including meconium⁸ and bilirubin.⁹

The absorption spectra of melanin overlaps the wavelengths used by NIRS and pulse oximetry.^10,11 This has formed the basis of extensive recent literature regarding inaccuracy of pulse oximetry in patients with darker skin.^12,13 Despite this growing evidence, relatively little is known about the effect of skin tone on NIRS. Early lab-based studies suggested that skin color could make NIRS inaccurate or cause loss of signal.^14,15 More recent laboratory studies have found correlation between skin tone and NIRS measurements in healthy patients^16,17 and a lab phantom.¹¹ Some NIRS devices use multiple wavelengths to attempt to adjust for additional chromophores like melanin.^18–20

A retrospective clinical study identified lower cerebral NIRS in African American patients compared with White patients undergoing cardiac surgery,²¹ but this finding has not been universal, depending on the machine used.¹⁹ A recent systematic review identified only these two studies examining NIRS and race.²² Importantly, both studies were retrospective and used race as a surrogate measure for skin tone. There is significant variability in skin color within races, and this introduces uncertainty of effect size and makes potential device improvement impractical. Additionally, these studies did not directly measure Sao₂ or Svo₂, making determination of accuracy impossible. There have been no studies of NIRS and skin tone in children, and there have been no prospective clinical studies of NIRS that included quantitative measurement of skin tone.

We aimed to quantify the effect of skin tone on NIRS performance. We hypothesized that darker skin tone would lead to underestimation of mixed venous saturation.

Materials and Methods

Compliance with Ethical Standards

This study was approved by the Institutional Review Board at Vanderbilt University Medical Center (approval No. 231423). Written informed consent was obtained from all participants or their parents, and age-appropriate written assent was also obtained.

Study Population and Sample Size

Consecutive patients less than 21 yr of age undergoing cardiac catheterization were screened for enrollment from May 28, 2024, to December 3, 2024. To be eligible, a systemic arterial and mixed venous oxygen saturation had to be obtained as a part of the catheterization. Exclusion criteria included (1) methemoglobinemia or carbon monoxide poisoning, (2) direct hyperbilirubinemia, and (3) emergent catheterization before which consent was not able to be obtained. Patients were also enrolled in a concurrent pulse oximetry study,¹³ so patients with profound anemia with hemoglobin less than 8 g/dl and nonpulsatile cardiovascular circulation were also excluded. Patients undergoing multiple catheterizations during the study period were enrolled once. Details of patients approached and enrolled can be found in figure E1 (Supplemental Digital Content, https://links.lww.com/ALN/E220).

A retrospective clinical study found that African American race was associated with a decrease in cerebral NIRS by 8%.²¹ We estimated our population to be approximately 20% non-White, based on historical data. Using a t test, power of 0.8, alpha of 0.05, a difference of 8%, published mean and SD for NIRS, and a 4:1 allocation yielded a sample size of 110 (22 children from races associated with darker skin tone and 88 White children). Because data were not available regarding NIRS and individual typology angle (ITA) skin types, we assumed that White children would generally resemble ITA types 1 and 2, while children from races that tend to have darker skin tone would be more likely to have ITA types 5 and 6 skin tone. Based on this assumption and our inability to account for multiple comparisons based on available data, our sample size may be underestimated.

Skin Tone Measurement

Skin tone was measured at the dorsal aspect of the distal phalanx (between the joint and fingernail), the palmar distal phalanx, and the forehead. As cerebral NIRS was the focus of this study, forehead measurements were used for analyses. Measurements were taken using the perceived Fitzpatrick scale²³ based on the original question-based scale,²⁴ Monk scale,²⁵ and a Konica Minolta (USA) CM-700d spectrophotometer. Measurements were taken by trained study personnel (J.R.S., W.W., and S.R.) under consistent fluorescent lighting conditions in the perioperative holding area. Visual scales were obtained online and printed using a Canon ImagePRESS C700 printer (Canon, USA).

Spectrophotometer measurements were obtained with the Konica Minolta CM-700d spectrophotometer using the 3-mm aperture, 2-degree observe angle, and D65 illuminant setting. White calibration was conducted before each patient using the manufacturer-supplied standard. Measurements were obtained in triplicate without lifting the spectrophotometer, and the average of each anatomic site was used. The device was paired with the CM-SA skin analysis software for data processing (Konica Minolta Sensing Americas, USA). This software directly calculates a melanin index. ITA was also calculated for each site. ITA correlates with melanin content²⁶ and is defined using L* (lightness) and b* (blue–yellow axis) from the CIELAB color space. Importantly, ITA becomes more negative as skin becomes darker, while melanin index becomes higher as skin becomes darker. ITA is defined as

$$\text{ITA}=\arctan \left(\frac{\text{L}*-50}{\text{b}*}\right)*\frac{180}{\pi }$$ (1)

NIRS and Saturation Measurement

NIRS measurements were recorded at the exact time as directly observed arterial and mixed venous blood sampling. The Medtronic (USA) INVOS 5100C regional oximeter was used for all NIRS measurements. The INVOS uses two wavelengths of infrared light (730 and 810 nm). Infant regional saturation sensors were used for patients weighing less than 40 kg, and adult sensors were used for larger patients according to manufacturer recommendations. Sensors were placed by study personnel over the right forehead following manufacturer guidance. Fractional Sao₂ and Svo₂ measurements were obtained immediately after sampling using an Avoximeter 1000E co-oximeter (Werfen, USA). Svo₂ was considered the superior vena cava saturation if left-to-right intracardiac shunting was present and the distal pulmonary artery saturation if no shunting was present.

Additional Data Collection

Race, ethnicity, and sex were reported by patients and their parents at the time of enrollment. Demographic data were collected from the medical record, including age, height, weight, cardiac anatomy, and previous repairs. Laboratory values, including hemoglobin and blood gas values, were recorded from the electronic medical record. The most recent hemoglobin value within the 72 h preceding the catheterization was used for analyses. Data regarding vitals and medication use at the time of blood sampling were gathered from the anesthesia record. These included heart rate, blood pressure, temperature, fraction of inspired oxygen, and vasopressor use.

Statistical Analysis

The original INVOS validation was performed using a 75:25 weighted average of jugular venous bulb saturation and Sao₂.²⁷ Jugular venous saturation is rarely available in clinical practice. Because NIRS is often used as a surrogate for Svo₂, many studies have used superior vena cava or pulmonary artery saturations for comparison.^28,29 Similarly, jugular venous bulb saturations are not routinely collected during cardiac catheterization in our laboratory. Our primary analysis therefore uses Svo₂ as the standard for comparison. We conducted a secondary analysis using a 75:25 weighted average of Svo₂ and Sao₂.

Descriptive statistics are reported as median (interquartile range [IQR]) and number (%). Three measures typically used for pulse oximetry evaluation studies—bias, precision, and accuracy root mean square error (ARMS)—were calculated as in previous NIRS studies.¹⁷ Individual-level bias was calculated as the difference between the NIRS measurement and either Svo₂ or the weighted average of Svo₂ and Sao₂. Bias is reported as the average of the individual-level bias for a given category. Precision was calculated as the SD of the individual-level bias. ARMS was calculated as the square root of the average squared difference between the NIRS measurement and Svo₂ or the weighted average of Svo₂ and Sao₂. An overall test (analysis of variance for means and Bartlett’s for SD) was first performed for categorical comparisons. If significant, pairwise comparisons were then conducted, and t tests were used for pairwise comparisons of means and were adjusted for multiple comparisons using the Bonferroni correction. Overall, Bartlett’s tests were not significant, so pairwise comparisons were not conducted. Normality was evaluated both by visual inspection of histograms and D’Agostino’s K-squared test. ARMS were compared by calculating 95% CI by bootstrapping with 10,000 repetitions and then comparing estimates using the Wald test adjusted for multiple comparisons. Bland–Altman plots were used to visualize the distribution of bias. ITA was divided into previously described categories of very light (greater than or equal to 55), light (greater than or equal to 41 to less than 55), intermediate (greater than or equal to 28 to less than 41), tan (greater than or equal to 10 to less than 28), brown (greater than or equal to −30 to less than 10), and dark (less than −30).^30–32 For categorical comparisons, patients were grouped as ITA 1 and 2 (very light or light), ITA 3 and 4 (intermediate or tan), and ITA 5 and 6 (brown or dark). Bias, precision, and ARMS were compared across these categories.

Multivariable linear regressions were used to determine the association of ITA and bias, conditional on patient covariates. Regressions were adjusted for mixed venous saturation, pH, heart rate, mean arterial pressure, vasopressor use, age, temperature, and hemoglobin. Profile plots were used to visualize the output of these regressions. All analyses were performed using Stata version 14.2 (StataCorp LP, USA). A statistical significance criterion of P < 0.05 was used for all analyses. A data analysis and statistical plan was written and filed with the institutional review board before data were collected.

Results

Demographics

The median age was 3.4 yr (range, 0.0 to 19.1 yr), and a slight majority of patients were male (n = 57; 51.8%; table 1). The median ITA was 41.4, and the median melanin index was 0.8. The most common cardiac lesions were hypoplastic left heart syndrome (12;, 10.9%), patent ductus arteriosus (10; 9.1%), and pulmonary atresia (10; 9.1%). There were 25 (22.7%) patients who had undergone cardiac repair, 29 (26.4%) who had undergone palliation, 10 (10.1%) who had a transplant, and 46 (41.8%) who were unrepaired.

Table 1.

Demographics and Skin Tone Measurements

Characteristic	ITA 1–2 (Lightest; n = 56)	ITA 3–4 (n = 17)	ITA 5–6 (Darkest; n = 17)	Overall
Age, yr	3.4 (1.5, 7.7)	4.7 (0.2, 12.5)	3.1 (0.5, 11.5)	3.4 (0.5, 11.2)
Height, cm	93.5 (74.5, 114.4)	102.0 (58.5, 146.0)	92.5 (61.7, 151.0)	94.8 (63.0, 138.0)
Weight, kg	13.5 (9.3, 22.8)	17.1 (4.9, 38.2)	10.9 (6.4, 40.0)	14.4 (6.7, 32.4)
Sex
Female	21 (37.5%)	19 (51.4%)	13 (76.5%)	53 (48.2%)
Male	35 (62.5%)	18 (48.7%)	4 (23.5%)	57 (51.8%)
Ethnicity
Hispanic	5 (8.9%)	9 (24.3%)	5 (29.4%)	19 (17.3%)
Not Hispanic	51 (91.1%)	28 (75.7%)	12 (70.6%)	91 (82.7%)
Race*
American Indian/Alaska Native	1 (1.8%)	1 (2.7%)	0 (0.0%)	2 (1.8%)
Asian	1 (1.8%)	3 (8.1%)	0 (0.0%)	4 (3.6%)
Black	1 (1.8%)	8 (21.6%)	13 (76.5%)	22 (20.0%)
White	54 (96.4%)	33 (89.2%)	4 (23.5%)	91 (82.7%)
Quantitative skin tone
ITA	53.3 (47.1, 58.4)	28.5 (19.9, 34.6)	-22.7 (-32.6, -13.5)	41.4 (22.6, 53.4)
Melanin index	0.6 (0.5, 0.8)	1.1 (0.9, 1.3)	1.9 (1.5, 2.1)	0.8 (0.6, 1.2)
Saturation, %
Arterial saturation	95 (90, 97)	92 (87, 96)	94 (92, 96)	94 (89, 97)
Mixed venous saturation	67 (62, 74)	69 (65, 74)	70 (67, 73)	68 (63, 74)
75:25 arterial:mixed venous average	73.0 (69.6, 79.1)	74.8 (70.0, 79.8)	76.3 (72.0, 78.5)	73.8 (70.0, 79.8)
NIRS	69 (60, 74)	63 (59, 73)	56 (48, 63)	65 (56, 73)
Clinical variables
Heart rate, beats/min	101 (90, 119)	101 (82, 131)	118 (101, 139)	101 (90, 126)
Hemoglobin, g/dl	12.0 (10.6, 13.6)	13.6 (11.4, 16.0)	12.5 (10.4, 13.5)	12.3 (10.8, 14.1)
Mean arterial pressure, mmHg	55 (51, 62)	58 (53, 65)	59 (54, 68)	57 (52, 63)
pH	7.42 (7.37, 7.45)	7.41 (7.38, 7.43)	7.41 (7.39, 7.44)	7.41 (7.38, 7.44)
Temperature, °C	36.6 (36.5, 36.9)	36.6 (36.2, 36.8)	36.5 (35.8, 37.0)	36.6 (36.2, 36.9)
Vasopressor use	6 (10.7%)	3 (8.1%)	1 (5.9%)	10 (9.1%)
Carboxyhemoglobin (n = 100), %	1.5 (1.1, 1.7)	1.5 (1.2, 1.7)	1.5 (1.2, 1.7)	1.5 (1.1, 1.7)
Methemoglobin (n = 101), %	0.8 (0.7, 1.1)	1.0 (0.8, 1.1)	0.9 (0.8, 1.0)	0.8 (0.7, 1.1)

^*Respondents could select more than one.

ITA, individual typology angle; NIRS, near-infrared spectroscopy.

Bias, Precision, and ARMS

The median Svo₂ was 68% (IQR, 63 to 74%), and the median Sao₂ was 94% (IQR, 89 to 97%). The median 75:25 weighted average of Svo₂ and Sao₂ was 73.8% (IQR, 70.0 to 79.8%). The median time difference between Sao₂ and Svo₂ measurement was 7 (IQR, 4 to 9) minutes. The median difference in NIRS between the Sao₂ and Svo₂ measurements was 1 min (IQR, 0 to 2).

Comparing NIRS to Svo₂, the overall mean bias was −2.7% (table 2). The mean bias for the darkest patients in ITA categories 5 and 6 was −12.8%, which was significantly larger than ITA 3 and 4 at −2.5% with a difference of 10.3% (95% CI, 4.4 to 16.3; P < 0.001) and ITA 1 and 2 at 0.3% with a difference of 13.1% (95% CI, 7.5 to 18.7; P < 0.001). There was no significant difference in precision represented by the SD of bias. The overall ARMS was 9.72. The ARMS for ITA 5 and 6 was 15.23, which was higher than ITA 3 and 4 at 9.08 with a difference of 6.15 (95% CI, 2.32 to 9.97; P = 0.005) and ITA 1 and 2 at 7.80 with a difference of 7.43 (95% CI, 3.71 to 11.15; P < 0.001).

Table 2.

Bias, Precision, and ARMS

Skin Tone	NIRS versus Svo2			NIRS versus Weighted Average
	Mean Bias	Precision (SD)	ARMS	Mean Bias	Precision (SD)	ARMS
ITA 1–2	0.30	7.87	7.80	−6.21	7.52	9.70
ITA 3–4	−2.49	8.86	9.08	−8.28	8.54	11.81
ITA 5–6	−12.82	8.47	15.23	−19.00	8.27	20.63
Overall	−2.66	9.40	9.72	−8.88	9.08	12.67

ARMS, accuracy root mean square; ITA, individual typology angle; NIRS, near-infrared spectroscopy; Svo₂, mixed venous oxygen saturation.

Findings were similar when comparing NIRS to the weighted average of Svo₂ and Sao₂ but with generally larger biases and ARMS. The overall mean bias was −8.9%. The mean bias for the darkest patients with ITA 5 and 6 was −19.0%, which was again larger than ITA 3 and 4 at −8.3% with a difference of 10.7% (95% CI, 5.0 to 16.4; P < 0.001) and ITA 1 and 2 at −6.2% with a difference of 12.8% (95% CI, 7.4 to 18.2; P < 0.001). There was no significant difference in precision. The overall ARMS was 12.67. ARMS for ITA 5 and 6 was 20.63, which was higher than ITA 3 and 4 at 11.81 with a difference of 8.82 (95% CI, 4.60 to 13.03; P < 0.001) and ITA 1 and 2 at 9.70 with a difference of 10.93 (95% CI, 7.06 to 14.80; P < 0.001).

Bland–Altman plots were used to visualize the distribution of bias (fig. 1). The mean bias of NIRS compared to Svo₂ was −2.7%, and the 95% limit of agreement was (−21.1, 15.8). The mean bias when compared to the weighted average of Svo₂ and Sao₂ was −8.9% with 95% limit of agreement of (−26.7, 8.9).

Fig. 1.

Bland–Altman plots. (A) Near-infrared spectroscopy (NIRS) compared to mixed venous oxygen saturation (Svo₂) with a mean difference of −2.7% and a 95% limit of agreement of (−21.1, 15.8). (B) NIRS compared to the weighted average of Svo₂ and arterial oxygen saturation (Sao₂) with a mean difference −8.9% and a 95% limit of agreement of (−26.7, 8.9).

Regression Analyses

Multivariable linear regressions were adjusted for the mixed venous saturation, pH, heart rate, mean arterial pressure, vasopressor use, age, temperature, and hemoglobin level. ITA was significantly associated with increasing error when NIRS was compared both to Svo₂ (0.173; 95% CI, 0.115 to 0.231; P < 0.001) and the weighted average of Sao₂ and Svo₂ (0.169; 95% CI, 0.111 to 0.226; P < 0.001; fig. 2). Findings were similar when using melanin index in place of ITA (NIRS vs. Svo₂: −11.496; 95% CI, −14.966 to −8.026; P < 0.001; and NIRS vs. weighted average: −11.295; 95% CI, −14.730 to −7.859; P < 0.001). Model characteristics can be found in supplemental tables E1 to E4 (Supplemental Digital Content, https://links.lww.com/ALN/E220).

Fig. 2.

Predicted error by individual typology angle (ITA). The panels show profile plots of multivariable linear regressions for near-infrared spectroscopy (NIRS) error based on ITA when NIRS was compared to mixed venous oxygen saturation (Svo₂; A) and the weighted average of Svo₂ and arterial oxygen saturation (Sao₂; B). ITA was significantly associated with error using both comparisons. Regressions are adjusted for mixed venous saturation, pH, heart rate, mean arterial pressure, vasopressor use, age, temperature, and hemoglobin level.

Discussion

We report a real-world study in children of the effect of skin tone on NIRS performance. Bias and ARMS were larger than those reported for pulse oximetry, and both were larger when the weighted average of Sao₂ and Svo₂ was used for comparison. NIRS bias was larger in those with darker skin tone in unadjusted and adjusted analyses. Underestimation of Svo₂ by NIRS may lead to unnecessary interventions and may contribute to inequities in morbidity and mortality. Our findings suggest that current validation methods are inadequate in ensuring equitable accuracy across diverse skin tones.

A recent systematic review²² identified just two studies investigating cerebral NIRS and race.^19,21 Our study improves on key limitations of these previous studies. We quantitatively measured skin tone in place of using race as a proxy variable. We also directly measured Sao₂ and Svo₂, allowing for quantification of bias. Similar to Sun et al.²¹ finding an effect of race on NIRS, we identified a significant effect of skin tone on NIRS. Both bias and ARMS increased as skin tone became darker, suggesting that melanin interferes with NIRS performance. Interestingly, we did not find significant differences in precision represented by the SD of the bias. There was substantial imprecision across skin tones, and this does not worsen as skin becomes darker.

Our study is in contrast to that of Stannard et al.,¹⁹ who did not find an association between race and NIRS. This may be due to differences in devices. Our study used the INVOS 5100C, which is the same device used by Sun et al.²¹ In contrast, Stannard et al.¹⁹ studied the FORE-SIGHT Elite tissue oximeter (Edwards Lifesciences, USA). The FORE-SIGHT device uses five wavelengths in place of the two used by the INVOS device, which may help correct for additional chromophores, such as melanin.^18,19 Further research is needed to determine whether this discrepancy between studies is in fact due to differential device performance or because previous studies have not directly measured skin tone.

Our overall bias, precision, and ARMS were similar to a previous study of the INVOS device.¹⁷ Of note, this study used the weighted average of Sao₂ and Svo₂. This study also found increasing (more negative) bias with darker skin pigmentation measured categorically, although this association did not reach significance. They found an average bias of −10.1% in the darkest patients, which was similar but somewhat smaller than our findings. This difference likely reflects differences in skin tone categorization.

Interestingly, we found larger bias and ARMS when comparing NIRS to the weighted average of Sao₂ and Svo₂. This was true both for the overall population and for varying skin tones. This is surprising because the INVOS system was originally validated against the weighted average and not Svo₂ alone.²⁷ This occurred because most NIRS values were less than the mixed venous saturation, so the addition of the higher arterial saturation increased the difference between the two values. This supports the clinical use of NIRS as a surrogate measure of Svo₂. However, it is difficult to justify using NIRS as a surrogate measure of Svo₂ physiologically as regional saturation should represent a mixture of venous and arterial blood in the tissue. Further studies in larger cohorts are necessary to clarify this relationship.

Our study builds on the pulse oximetry literature that skin pigmentation can interfere with optical medical devices and make performance worse. The ARMS and bias reported here are substantially higher than those reported in pulse oximetry studies, which is consistent with previous literature.¹⁷ However, the bias and ARMS for the darkest patients in our study were substantially higher than overall bias and ARMS reported in previous NIRS studies. The observed errors were larger than what is expected from the inherent variability in NIRS and is therefore likely to lead to differences in clinical management. Additionally, the effect of skin tone on pulse oximetry leads to overestimation as skin becomes darker, while with NIRS it leads to underestimation.

These devices estimate oxygen saturation using the ratio of light transmitted through or reflected by the tissue at about 660 nm (transmitted more by oxygenated hemoglobin) and about 900 nm (transmitted more by deoxygenated hemoglobin). The spectral transmission of melanin increases over this range, effectively filtering light that reaches the blood. Because of the shift created by the use of polychromatic light sources rather than monochromatic ones, the ratio of ratios (sometimes referred to as R) used in the calculation of oxygen saturation varies with melanin concentration rather than staying constant across the range of wavelengths utilized as is assumed.³³ R is systematically higher in individuals with higher melanin content. This may be the cause of inaccurate values when other chromophores absorb light at similar wavelengths to hemoglobin. The fact that most pulse oximeters are transmittance devices (light going through measured area) and NIRS is a reflectance device (light reflected back) may explain why the effects of melanin are different. The exact mechanism of melanin-based bias in these devices is still being investigated.

Limitations

The generalizability of our study is limited by the use of a single NIRS device, and the measured biases may not be applicable to other devices. Future studies will investigate additional devices, especially those like the FORE-SIGHT that use additional wavelengths and may account for additional chromophores more effectively. We also recognize that many clinicians use the trend in NIRS as opposed to the specific value. As saturations in the catheterization laboratory are obtained at a fixed timepoint, we were not able to examine the effect of skin tone on the trend in NIRS. However, previous studies have found low baseline NIRS to be a poor prognostic factor for cardiac surgical outcomes,^19,21 and the raw NIRS value is used both for prognostication and acute bedside assessment if trends are not available. Additionally, the NIRS value is often used at the bedside as a substitute for Svo₂ to calculate the ratio of pulmonary and systemic blood flow. In all of these settings, children with darker skin may receive different care due to poorer NIRS performance. Our measurement of Svo₂ and Sao₂ was not simultaneous, and this may affect calculations using the weighted average of Sao₂ and Svo₂. We believe the effect of this on our findings to be minimal given the relatively short time between measurements and very little change in NIRS between the two timepoints for each patient. Finally, the oximeter used in this study measures fractional saturation, which contains carboxyhemoglobin and methemoglobin in the denominator, as opposed to functional saturation typically used in NIRS and pulse oximetry validation studies. However, these values are similar in settings where carboxyhemoglobin and methemoglobin are low as in our study population.

Conclusions

Darker skin tone is associated with worse NIRS performance, leading to lower NIRS values compared to mixed venous saturation using the INVOS 5100C system. This may lead to differences in care and contribute to inequities in outcomes in patients with darker skin tone. Methods of either correcting these inaccurate measurements or enhancing measurement accuracy should be undertaken to improve equity. Moreover, more stringent validation guidelines should be implemented to ensure equity in performance across varying skin tones for future optical devices.

Research Support

Supported by grant No. T32 HS026122 from the Agency for Healthcare Research and Quality (Rockville, Maryland; to Dr. Starnes). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. This work was also supported in part by Vanderbilt Clinical and Translational Science Award (CTSA) grant No. UL1TR002243 from the National Center for Advancing Translational Sciences/National Institutes of Health (Bethesda, Maryland) and by the Jordan Hackett Foundation (to Dr. Starnes).

Competing Interests

An INVOS 5100C NIRS device was provided by Medtronic (Minneapolis, Minnesota) for the performance of this study. Medtronic had no role in the collection, analysis, or interpretation of the data. They further had no role in the writing of the report or the decision to submit for publication. Dr. Soslow also reports consulting for Capricor (San Diego, California) and Sarepta (Cambridge, Massachusetts) regarding Duchenne muscular dystrophy therapeutics and teaching for NS Pharma (Paramus, New Jersey). The other authors declare no competing interests.

Supplemental Digital Content

Figure E1 and Tables E1 to E4, https://links.lww.com/ALN/E220