TABLE 3.
Results of the Adjusted Linear Mixed-Effects Models
| Model | Variables | Coefficients | p | N | Log-Lik |
|---|---|---|---|---|---|
| Association between race Spo2 bias (Supplemental Formula 5, http://links.lww.com/CCX/B379) | White | Baseline | 0.64 | 463 | –994.3 |
| Black | –0.23 | ||||
| Other | –0.31 | ||||
| Association between separate Skin Tone Scale and Spo2 bias (Supplemental Formula 6, http://links.lww.com/CCX/B379) | Fitzpatrick | –0.75 | 0.24 | 463 | –993.6 |
| Von Luschan | –1.10 | 0.16 | 463 | –993.3 | |
| Monk | –2.40 | 0.01 | 463 | –991.2 | |
| Delfin individual typology angle | –0.62 | 0.45 | 463 | –993.9 | |
| Delfin L* | 0.06 | 0.98 | 463 | –994.3 | |
| Konica Minolta L* | –0.77 | 0.38 | 424 | –914.0 | |
| Variable L* | –0.31 | 0.67 | 367 | –784.4 | |
| Delfin Melanin Index | 0.17 | 0.87 | 463 | –994.3 | |
| Association between six Skin Tone Scales and bias (Supplemental Formula 7, http://links.lww.com/CCX/B379) | Expected total effecta | –1.72 | 0.02 | 463 | –986.8 |
Expected total effect: the expected difference in estimated measurement bias of the darkest and lightest subject (assuming the normalized value of all skin tone measurements is 1 for the darkest subject and 0 for the lightest), computed as the sum of the separate coefficients.
Results of the four linear mixed-effects models with clinical variables (Sao2, pH heart rate, and mean arterial pressure) adjusted (Supplemental Formulas 5–8, http://links.lww.com/CCX/B379). Likelihood ratio tests (LRTs) are performed to demonstrate whether the null hypothesis should be rejected. Variables and coefficients are derived from the linear mixed-effects model with a negative value being a larger magnitude of bias, χ2 statistics, and p values are derived from LRT results. N is the sample size of each model. Red cells represent negative coefficient values, that is, the variable affects an overestimation of Sao2, and vice versa for green cells. Bold, underlined p values denote that the significance threshold was passed at 0.05 and the null hypothesis was rejected. The self-reported race alone (Supplemental Formula 5, http://links.lww.com/CCX/B379) presents coefficients in the expected direction (–0.23%; 95% CI, –0.76 to 0.30%; p = 0.64 for Black patients, compared with White patients), but the p value is not significant. When assessing the effect of a separate Skin Tone Scale on bias (Supplemental Formula 6, http://links.lww.com/CCX/B379), only the Monk Skin Tone Scale is shown to be significant (–2.40%; 95% CI, –4.32% to –0.48%; p = 0.01). The effect of all combined six Skin Tone Scales on the bias (the ones without missingness, Supplemental Formula 7, http://links.lww.com/CCX/B379) was found to be significant, with an expected total effect
of –1.72%, p = 0.02. Finally, when considering all eight Skin Tone Scale variables, this expected total effect remains in the expected direction (–3.80%), although the p value is not significant (p = 0.06).