Abstract Body (Do not enter title and authors here): Introduction
Recent retrospective studies have found overestimation of oxygen saturation by pulse oximetry in races with darker skin tone. Existing studies have used race as a poor surrogate for skin tone and have relied on paired pulse oximetry and blood gas measurements using timestamps, which may introduce significant error.

Hypothesis
We hypothesized that darker skin tone measured by spectrophotometry would lead to overestimation of oxygen saturation by pulse oximetry. We also hypothesized that multivariable modeling would partially correct this bias.

Methods
Consecutive patients less than 21 years of age undergoing cardiac catheterization were enrolled (N=461). Skin tone was measured at the dorsal distal phalanx using a spectrophotometer. Measurements were used to calculate the individual typology angle (ITA), which correlates with melanin content. Pulse oximetry (SpO2) values were recorded at the exact time of arterial blood sampling using two pulse oximeters (Nellcor and Masimo). Arterial saturation (SaO2), the reference value, was measured using an oximeter (Avoximeter). Multivariable linear regressions were used to quantify the effect of skin tone on pulse oximeter error. An initial cohort (n=319) was used to build a multivariable model, which was validated in a second cohort (n=142).

Results
A total of 156 (33.8%) patients identified as either part Hispanic or part non-white race (Table 1). Pulse oximetry bias, precision, and accuracy root mean square (ARMS) were worse among patients with the darkest skin tone (Fig 1; all p<0.001). Missed hypoxemia—SaO2 <88% and SpO2 >92%—was higher among the darkest patients for both devices (Nellcor 6.2% vs. 0.9%; Masimo 9.2% vs. 3.2%; p<0.05). ITA was associated with increasing error for both devices in the modeling cohort (Table 2, p<0.001). Using the model to correct SpO2 in the validation cohort decreased bias from 1.9% to 0.4% for the Nellcor device for ITA categories 5-6 and from 2.4% to 0.7% for the Masimo device. Occult hypoxemia also decreased from 1.9% to 0.0% for Nellcor, including from 4.0% to 0.0% in ITA 5-6, and from 3.7% to 1.2% for Masimo, including from 11.1% to 5.6% among ITA 5-6.

Conclusions
To our knowledge, this is the first prospective study in children measuring overestimation of saturation by pulse oximetry attributable to skin tone. These errors are partially correctable using modeling approaches with improved performance in a validation cohort.