Accuracy

All measurements have errors. Error is usually determined by the comparison of a measurement to a "gold standard" of that measure. Unfortunately, all measurements, even the so-called gold standards, are subject to errors with respect to reproducibility. From a clinician's perspective, we can depend on a physiologic measurement only if it is accurate to the degree required for clinical decision making. For example, systemic blood pressure can be measured in several ways. We can listen to Korotkoff sounds by the use of a sphygmomanometer cuff and stethoscope; we can rely on an oscillometric automated noninvasive blood pressure device, or if we need continuous measurement, we may place an arterial cannula.^[1] Unfortunately, each of these techniques provides a slightly different arterial blood pressure value, and each has different sources of error. Our choice of method may be determined by accuracy or by our needs for the frequency of the data and the ease of retrieving these data. An automatic oscillometric device is usually chosen over manual auscultatory measurements for ease of acquisition and reproducibility. Two people taking auscultatory blood pressure measurements may hear the Korotkoff sounds at slightly different points and record different blood pressures. The required accuracy of a clinical monitor is determined by the smallest change in the measured variable that could affect a clinical decision. The requirements for "absolute accuracy" (is the measured value correct?) may be different from the requirements for "relative accuracy" (does the measured value follow trends?).

For example, a pulse oximeter estimates arterial hemoglobin oxygen saturation by measuring light absorbance. The pulse oximeter saturation estimate, SpO₂ , is compared with the hemoglobin saturation determined from analysis of an arterial blood sample by a laboratory cooximeter, SaO₂ . Some errors are associated with arterial blood sampling and in vitro analysis of the sample by the co-oximeter. Nevertheless, the co-oximeter is considered to be the gold standard in this comparison of methods.

The pulse oximeter's saturation value can be compared with that of the co-oximeter by the determination of bias and precision ( Fig. 30-2 ). Bias is the average difference between simultaneous values from the two methods, or the systematic error. If a pulse oximeter reads an average of 5% higher than the co-oximeter does, it has a bias of 5%, and we can adjust for that systematic error by recalibrating the device. Precision is the standard deviation of the difference between the two measurements, and it quantifies the random error, or "scatter." A higher value for precision indicates a larger random error. (This statistic may be more appropriately called "imprecision.") If the random error is too large, the device may not be clinically useful. We can adjust for systematic error (bias) by recalibration, but there is no way to adjust for random error.^[2]