Metrology in medicine: from measurements to decision, with specific reference to anesthesia and intensive care

Abstract

Metrology is the science of measurements. Although of critical importance in medicine and especially in critical care, frequent confusion in terms and definitions impact either interphysician communications or understanding of manufacturers' and engineers' instructions and limitations when using devices. In this review, we first list the terms defined by the International Bureau of Weights and Measures regarding quantities and units, measurements, devices for measurement, properties of measuring devices, and measurement standards. The traditional tools for assessing the most important measurement quality criteria are also reviewed with clinical examples for diagnosis, alarm, and titration purposes, as well as for assessing the uncertainty of reference methods.

Figure 1.

Schematic representation of different types of systematic errors. Typically, these lines are these observed when there is an issue in calibration (blue line), in amplification (red line), or when insufficient sensing is corrected by adding a constant (orange line).

Figure 2.

Schematic representation of 2 types of random errors. Blue plots, small random error (2σ = 0.6) typically seen when an independent noise is properly filtered. In red plots, the variability in proportion to the measurement value indicating that the random noise is getting into the signal. In these examples, both data sets have correlation slopes on the identity line (black dotted line), indicating no systematic errors (x-axis and y-axis with arbitrary units).

Figure 3.

Schematic representation of the different types of measurement errors with indication of the formula by which it is derived and the corresponding quality criteria. m = mean; σ = standard deviation of replicate measurements on the same object. In this figure, the variability is expressed by ±2σ/m to fit with a confidence interval of 95%.

Figure 4.

Clark error grid for insulin dosing. Except for 2 measurements in zones C and D each, agreement between the 2 methods is acceptable for diabetes care in this example. (Reproduced from Ref. 20, with permission.)

Figure 5.

Bland-Altman plots. Mean difference = 17 ± 40 mg/dL. Limits of agreements (2σ) of the measurement error/ averaged value = 51%. (Reproduced from Ref. 20, with permission.)

Figure 6.

Modified error grid for the same data as in Figure 4. This error grid shows that the test device would have resulted in unacceptable, serious protocol violation, or even life-threatening therapy decisions in 10 cases, which may be considered unacceptable for clinical use in the intensive care unit. (Reproduced from Ref. 20, with permission.)

Figure 7.

Bland-Altman plot of mean arterial pressure measured by a noninvasive method (mBP) and the invasive method (mBPIA). Difference = −3.7 ± 6.2 mm Hg, ±10.3%.

Figure 8.

Same data as in Figure 6. Overlaid Bland-Altman plots of repeated measurements for each measurement technique (blue mBP; red = mBPIA). mBP: CR = 6.53, mBPIA: CR = 15.65