Even number bias in pulse oximetry measurements in hospitals in Denmark and the United Kingdom: an observational study

INTRODUCTION: It is not known if pulse oximetry oxygen saturation measurements (SpO 2) are subject to even-number bias and boundary effects which have been reported for other physiological measurements.

METHODS: We reviewed three large SpO 2 datasets from two countries for even-number bias: 3,351,981 continuous automated SpO 2 entries in Denmark (DK-continuous), 7,602,352 manual SpO 2 entries in Denmark (DK-manual), and 949,718 manual SpO 2 entries in the UK (UK-manual).

RESULTS: Even-numbered SpO 2 entries were commoner than odd numbers in all three datasets. Even numbers accounted for 50.6% of DK-continuous entries (95% CI: 50.6-50.7), 57.5% of DK-manual entries (95% CI: 57.4-57.6), and 59.7% of UK-manual entries (95% CI: 59.6-59.8): DK-manual vs. DK-continuous, relative risk [RR]: 1.14, 95% CI: 1.13-1.14, p < 0.0001; UK-manual vs. DK-continuous RR 1.18, 95% CI: 1.18-1.18, p < 0.0001. The even-number bias was more pronounced in UK medical patients for whom 64.9% of SpO 2 entries had even numbers (UK-manual medical entries vs. DK-continuous medical entries RR 1.29, 95% CI: 1.29-1.30, p < 0.0001). The proportion of even SpO 2 numbers was particularly high (67.0%; 95% CI: 66.5-67.5%) for UK medical patients who were using supplemental oxygen.

CONCLUSIONS: We identified even-number bias in routine pulse oximetry entries in the UK and Denmark. The bias was substantially greater for manually collected data than for automated continuous data. The high degree of even-number bias in patients who were receiving supplemental oxygen suggests that the SpO 2 might be rounded to meet target saturation range or Early Warning Score boundaries. This bias could influence patient management and clinical outcomes.