Pulse oximetry values from 33,080 participants in the Apple Heart & Movement Study

Wearable devices that include pulse oximetry (SpO(2)) sensing afford the opportunity to capture oxygen saturation measurements from large cohorts under naturalistic conditions. We report here a cross-sectional analysis of 72 million SpO(2) values collected from 33,080 individual participants in the Apple Heart and Movement Study, stratified by age, sex, body mass index (BMI), home altitude, and other demographic variables. Measurements aggregated by hour of day into 24-h SpO(2) profiles exhibit similar circadian patterns for all demographic groups, being approximately sinusoidal with nadir near midnight local time, zenith near noon local time, and mean 0.8% lower saturation during overnight hours. Using SpO(2) measurements averaged for each subject into mean nocturnal and daytime SpO(2) values, we employ multivariate ordinary least squares regression to quantify population-level trends according to demographic factors. For the full cohort, regression coefficients obtained from models fit to daytime SpO(2) are in close quantitative agreement with the corresponding values from published reference models for awake arterial oxygen saturation measured under controlled laboratory conditions. Regression models stratified by sex reveal significantly different age- and BMI-dependent SpO(2) trends for females compared with males, although constant terms and regression coefficients for altitude do not differ between sexes. Incorporating categorical variables encoding self-reported race/ethnicity into the full-cohort regression models identifies small but statistically significant differences in daytime SpO(2) (largest coefficient corresponding to 0.13% lower SpO(2), for Hispanic study participants compared to White participants), but no significant differences between groups for nocturnal SpO(2). Additional stratified analysis comparing regression models fit independently to subjects in each race/ethnicity group is suggestive of small differences in age- and sex-dependent trends, but indicates no significant difference in constant terms between any race/ethnicity groups for either daytime or nocturnal SpO(2). The large diverse study population and study design employing automated background SpO(2) measurements spanning the full 24-h circadian cycle enables the establishment of healthy population reference trends outside of clinical settings.