Measured vs Calculated Oxygen Saturation in a Population of Pediatric Intensive Care Patients

Calculated saturation values, obtained from blood gas analyzers, are commonly used to make clinical decisions about oxygenation status, despite the widespread belief that measured values are more precise. The tacit assumption is that the calculated value is close enough to the actual saturation for practical purposes. We sought to examine this assumption by determining the relation between calculated and measured oxygen saturation in whole blood obtained from a population of pediatric intensive care patients. METHODS: Samples of whole blood were obtained from 30 patients, separated into two portions, and simultaneously analyzed in a blood gas analyzer (Radiometer ABL 4) and an oximeter (Radiometer OSM-3). Data were evaluated for overall agreement using standard correlation and regression analysis with a paired t test for the difference between the means. The reliability of individual calculated saturation values in the clinical setting was evaluated by determining the limits of agreement (mean difference two standard deviations of the difference scores). A p value < 0.05 was required for significance. RESULTS: The mean difference between measured and calculated saturation was 0.7% ±2.8% ( x ¯ ± s , p = 0.18), with r 2 = 0.91 (p < 0.001). The limits of agreement ranged from -6.4% to +4.9% for all observations. When the data were stratified, r 2 for saturations below 95% was 0.8 (p<0.001) with limits of agreement ranging from -8.8 to +6.9%. For saturations > 95%, the limits of agreement ranged from 1.7% to +0.7%. CONCLUSION: The commonly used statistical analyses support the hypothesis that, on the average, there is no difference between measured and calculated saturations. However, clinical decisions are often based on individual rather than mean data values. An examination of the limits of agreement revealed that individual values for calculated saturation may be imprecise estimates of true (ie, measured) saturation and can introduce unacceptable errors if used to derive clinical indices.