Skin pigmentation

The effect of skin pigmentation on pulse oximeter accuracy remains poorly characterized. Pulse oximeters may be less accurate in patients with darker skin, especially at lower oxygen saturations. Of note, this inaccuracy can tend to overestimate oxygen saturation, which may lead to false reassurance during clinically real hypoxemia. The amount of inaccuracy varies significantly between different devices, manufacturers, and probes. The OpenOximetry.org Project is actively working to better understand the impact of pigment on oximeter accuracy and to develop new strategies to eliminate this source of error.

References: Okunlola et al, Resp Care, Nov 2021, Bickler et al, Anesthesiology 2005; Feiner et al, Anesth Analg 2007; Adler et al, Acad Emerg Med 1998; Bothma et al, S Afr Med J 1996; Zeballos et al, Am Rev Respir Dis 1991; Ries et al, Chest 1989; Emery J Perinatol 1987; Sjoding et al, N Engl J Med 2020

Keywords: skin, pigment, skin tone, accuracy

For oximeters that have been approved by the US FDA, accuracy over the SpO2 range of 70-100% is within a few SpO2 percent of the actual value when tested under laboratory validation conditions (for some oximeters, performance may be different when used in various clinical conditions). Most oximeters tend to be more accurate with higher SpO2 values than lower SpO2 values. 

The FDA has published guidelines for the recommended standards for pulse oximeter performance, which includes device accuracy. These guidelines require the devices to be tested with a minimum of 200 data points over an SaO2 range of 70% to 100%, and to be tested on people with different skin tones. Studies have shown that many low-cost pulse oximeters demonstrate highly inaccurate readings. However, some low-cost pulse oximeters have performed with similar accuracy to more expensive units when used in healthy subjects. However, it is important to recognize that many additional variables can make even good pulse oximeters have less accurate readings.

References: Lifebox Pulse Oximetry Learning Module; FDA Guidelines; Lipnick et al, Anesth Analg 2016

Keywords: accuracy, low-cost, guidelines, FDA

For FDA or ISO clearance, pulse oximeters must undergo testing in healthy human study subjects.

ISO 80601-2-61 defines the test requirements, and the FDA has published guidance that refers to this standard. The standard requires the devices to be tested with a minimum of 200 data points (paired observations: pulse oximeter, co-oximeter) evenly distributed over an SaO2 range of 70% to 100%. The standard requires that devices be tested on “10 or more healthy subjects that vary in age and gender” and that the study has subjects with “a range of skin pigmentations, including at least 2 darkly pigmented subjects or 15% of the subject pool, whichever is larger.”

Briefly, study subjects are semi-supine (30° head up) with a nose clip (to prevent breathing through the nose), breathing controlled mixtures of air-nitrogen-carbon dioxide via a mouthpiece from a partial rebreathing circuit with a voluntarily increased minute ventilation (sometimes coached with a metronome) and 10 to 20 L/min total fresh gas flow into the circuit. A 22-g radial artery catheter is placed to sample arterial blood for the measurement of SaO2. 

The gas mixture (e.g. nitrogen) of the circuit is manually adjusted to achieve a series of 10 to 12 stable SaO2 plateaus between 70% and 100% (approximately 70%, 73%, 76%, 80%, 83%, 86%, 89%, 92%, 95%, 98%, and 100%). Carbon dioxide is manually increased into the circuit to prevent hypocapnia. At each plateau (after 30-60 seconds of stability) an arterial sample is drawn and immediately “functional” arterial Sao2 (HbO2/[Hb+HbO2]) is determined by multi-wavelength oximetry (e.g. an ABL-90, OSM3 or similar device). After an additional 30 seconds of stability another sample is drawn. These SaO2 values are recorded and used to compare with recorded simultaneous SpO2 values from the device being tested.  During the tests, subjects’ extremities may be placed under a warmer to promote good perfusion. 

References: FDA Guidelines 

Keywords: FDA, guideline, clinical, accuracy, 510k, approval, ISO

The short answer is that there is no standardized approach.

According to FDA regulations, manufacturers of pulse oximeters should validate device performance on “a range of skin pigmentations, including at least 2 darkly pigmented subjects or 15% of the subject pool, whichever is larger.” While ‘darkly pigmented’ is not formally defined, it is routinely defined by testing labs as Fitzpatrick skin phototype groups V-VI. The Fitzpatrick scale is a widely used method for classifying skin pigment, from I (pale white) to VI (darkest brown). However, it was originally developed for skin photosensitivity typing, which is not the same as skin color, and should not be conflated with race or ethnicity. Studies have shown inaccuracies in self-reported values, especially in darker skin types. There are other similar visual skin color classification systems, but as with the Fitzpatrick scale, they are all subjective. 

In early 2021, in response to Sjoding et al’s findings of racial (note, not necessarily skin color) bias in pulse oximetry measurement, the FDA issued a statement encouraging further investigation on pulse oximeter accuracy in darker skin types. Given the limitations of the Fitzpatrick scale, there is a need to adopt a standardized objective method for measuring skin color that provides reliable, quantitative, and easily interpretable results. For more discussion on new ways to address racial inequity in oximetry, check out this paper (November 2021).

There are several commonly used methods for determining ‘skin color.’ Not all are used for the purposes of pulse oximeter validation cohort diversity. Below we discuss some of these methods including information on a few commonly encountered colorimeters on the market, including Mexameter, Colorimeter, and DermaCatch.

Below is a table summarizing several traditional classification systems for assessing skin color (not quantitative):

Reference: Ware et al, Racial Limitations of Fitzpatrick skin type, Cutis, 2020

Fitzpatrick Scale 

The Fitzpatrick skin type (FST) was developed in 1975 as a tool to assess the likelihood to burn during phototherapy treatments and is commonly used. The original scale was I-IV and it was only later that more diverse skin types were included. 

References: Image: CC BY 3.0 – D’Orazio et al, Int J Mol Sci, 2013; Eilers et al, Accuracy of Self-Report in assessing Fitzpatrick skin phototypes, JAMA Derm, 2013 ; Ware et al, Racial Limitations of Fitzpatrick skin type, Cutis, 2020 ; Wong et al, Analysis of discrepancies between pulse oximetry and arterial oxygen saturation measurements by race and ethnicity and association with organ dysfunction and mortality, JAMA, 2021

Von Luschan’s Chromatic Scale

The Von  Luschan’s Scale was developed as a tool to assess racial classifications according to skin color. It consists of 36 colored tiles which are compared to a person’s skin color.

This image is a reproduction of the Von Luschan scale which was originally created by Felix von Luschan and printed in Voelker, Rassen, Sprachen (1927). The original chart was scanned and the skin colors were copied using the paint program’s dropper tool.

Our team at Open Oximetry calculated the RBG values which are superimposed on the image of the scale. Notably, 13 and 14 were found to have the same RGB values.

References: Wikimedia Von Luschan’s Scale

Other ways to quantify skin pigment

Below we describe a few (of the many) products on the market to quantify skin pigmentation.

Mexameter® (MX 18) by Courage & Khazaka

• Narrow-band reflectance colorimeter (only targets melanin and hemoglobin)
• Output: melanin index and erythema index (arbitrary units 0-999)
• Melanin index can be falsely affected by erythema
• Requires base to be plugged in to display results (no computer needed)

Konica Minolta Chromameter

• The CM-700d Spectrophotometer is portable, wireless, and lightweight
• Allows for evaluation, reproduction, and control of pigment colors
• Aperture switches between 3mm and 8mm sizes to evaluate small to large samples
• Color spaces: L*a*b*, L*C*h, Hunter Lab, Yxy, XYZ, Munsell

For more information on studies conducted with the CM-700d Spectrophotometer, please view the following: 2022 study, 2020 study, and 2017 study.

Colorimeter® (CL 400) by Courage & Khazaka

• Full visible spectrum reflectance colorimeter
• Output: skin color RGB, L*a*b, xyz, ITA
• Requires USB connection to Windows PC to display results

SkinColorCatch® (previously DermaCatch) by Delfin Technologies

• Full visible spectrum reflectance colorimeter
• Output: skin color L*a*b, ITA, L*c*h, RGB, melanin and erythema indices
• Melanin and erythema indices insensitive to each other
• Portable and battery operated

Other Skin pigment scales:

Color Bar Tool

The color bar tool was developed by dermatologists to determine the amount of melanin pigment in various skin types and assess an individual’s potential to get a sunburn. This was done by comparing the color tool against the sun protected skin of the upper inner arm, and is best used when people are not flushed from exercise or heat to avoid any excess redness that might distort the visual perception of color.

Munsell Color Chart

The Munsell color chart (also known as the Munsell system of color notation) was created to specify an objective assessment of color based on three properties of color: hue, value, and chroma. Hue refers to the basic color which is described as red (R), yellow (Y), green (G), blue (B), and purple (P) across a range of values. Value refers to the lightness or darkness of a color, which ranges from 10 (pure white) to 0 (pure black). Chroma describes the color intensity, which ranges from light pastel colors to deeply saturated colors. The Munsell chart was originally designed as the official color system for soil colors and to be used in agriculture, but has been adopted for use to describe and assess human skin tones. Multiple sets of color tiles are available for purchase and can be compared against an individual’s skin to determine the closest color match. 

Various studies have used the Munsell system for skin color assessment. Many are related to dermatologic treatments and use the Munsell chart to describe the colors of capillary vascular malformations, melasma, hyperpigmentation, and other skin discolorations. The most common Munsell color charts used to determine skin pigments were the 5YR and 7.5YR, though 2.5YR and 10YR were also frequently used.

References: Reeder et al, CEBP, 2014, Wright et al, Skin Research & Technology, 2015, Ries et al, CHEST, 1989, Foglia et al, J Pediatr, 2017, Konishi et al, J Dermatol, 2007, Wright et al, Skin Res Technol, 2016, McCreath et al, J Adv Nurs, 2016

Massey-Martin Scale

The Massey-Martin Scale (also known as the New Immigrant Survey or NIS Skin Color Scale) is an 11-point scale, ranging from zero to 10, which was created to assess skin color. Zero represents albinism, or the absence of color, and 10 represents the darkest possible skin tone. The shades of skin color from 1 to 10 are depicted on the scale as a hand which differs in color for each number. The Massey-Martin scale has been used in various publications, especially topics related to skin color and the economic mobility of immigrants. The scale is intended to be essentially memorized by interviewers so that they can determine respondent’s skin tone without the respondent seeing the chart.

References: Hannon et al, Public Opin Q, 2016, Babool et al, Rheumatology, 2022, Massey NIS Skin Color Scale, 2003

Monk Skin Tone Scale

Additionally, some new color scales have been designed with the goal of improving skin tone evaluation in machine learning (ML) and artificial intelligence (AI), rather than for skin pigmentation quantification. The Skin Tone Research @ Google AI partnered with Harvard sociology professor Dr. Ellis Monk to open-source the Monk Skin Tone (MST) Scale, a continuous skin tone spectrum in the form of a 10-tone scale designed to represent a broader range of skin colors. Since AI can perpetuate unfair biases for people with darker skin pigmentation when it comes to camera technologies and image products, the MST is being utilized to evaluate datasets and ML models for better representation.

Interpreting quantitative results

Correlating quantitative measures to a visual color spectrum

Agache 2017 Measuring the Skin, Chapter 6 “The Measurement of Skin Color”:

• Skin color comprises of melanin content, oxy/deoxy-hemoglobin content, and endogenous/exogenous pigments such as bilirubin and carotene
• The CIE L*a*b color space is a device-independent reference for all the colors visible to the human eye and the most commonly used metric to quantify skin color
• Individual Typology Angle (ITA) is calculated from L and b and objectively categorizes skin color into 6 different groups, from very light to dark

Chau et al 2019 Cutaneous Colorimetry as Gold Standard for Skin Color Measurement:

• Explanation of CIE L*a*b color space
• Explanation of how skin colorimeters and spectrophotometers work

Visscher 2017 Skin color and pigmentation in ethnic skin:

• Correlates Fitzpatrick, ITA, Melanin Index

Wilkes et al 2015 Fitzpatrick Skin Type, Individual Typology Angle, and Melanin Index in an African Population: Steps Toward Universally Applicable Skin Photosensitivity Assessments:

• Correlates Fitzpatrick, ITA, Melanin index

Keywords: skin pigment, Fitzpatrick, melanin, skin color, Colorimeter

Additional References: Okunlola et al, Pulse oximeter performance, racial inequity, and the work ahead, Resp Care, Nov 2021; FDA guidance on pulse oximeters; February 19, 2021 FDA statement “Pulse Oximeter Accuracy and Limitations”: addresses Sjoding’s findings, research limitations, and the need for further study; Sjoding et al, Racial Bias in Pulse Oximetry Measurement, NEJM, Dec 2020

Comparison Table of Quantitative Methods for Assessing Skin Color:

 Table of Studies Comparing Quantitative Methods:

Systematic Review 2021

Note: DermaCatch (Colorix) is now SkinColorCatch (Delphin)

‘Differential bias’ (sometimes referred to as disparate bias) attempts to describe how much a pulse oximeter performs differently in people with light skin pigment as compared to dark skin pigment. To determine this, we compare the device’s bias (how much it over or under-reads SpO2) for people with very light skin and people with very dark skin*. 

Differential bias is calculated independently for two different SpO2 ranges (70-85% and 85-100%). Thresholds for allowable differential bias are not yet established though proposals have targeted 3.5-4% for the SpO2 70-85% range, and 1.5-2% for the SpO2 85-100% range. 

For example:

• Scenario 1: If a pulse oximeter on a person with light skin reads 1% higher than their true blood oxygen saturation (e.g., the oximeter shows 91% when the actual blood oxygen saturation is 90%), and on a person with dark skin it reads 2% higher (e.g., it shows 92% when the actual blood oxygen saturation is also 90%), the differential bias between the two is 1%.
• Scenario 2: If the pulse oximeter on a person with light skin reads 2% lower than their true blood oxygen saturation (e.g., the device shows 88% when the actual blood oxygen saturation is 90%), and on a person with dark skin it reads 1% lower (e.g., it shows 89% when the true value is 90%), the differential bias here is also 1%, but in the opposite direction.
• Scenario 3: If the pulse oximeter on a person with light skin shows no bias (e.g., reads exactly 90% when the actual blood oxygen saturation is 90%), but on a person with dark skin it reads 5% higher (e.g., the device shows 95% when the actual value is 90%), the differential bias is 5%.

These examples demonstrate how differential bias can manifest as either higher or lower readings depending on skin pigment.  

*Using modeling of skin pigment and SpO2 data, very light skin and very dark skin are defined as having an difference in ITA of 100. Read more about ITA here.