The finding that models of the same manufacturer usually differ significantly from each other for two or more analytes (Table 2) initially was surprising to us. However, these model differences are likely due to continuing improvements in analyzer geometry, calibration and flushing techniques, temperature control, electronic signal modification, and other unknown factors. Manufacturers can be expected to continue to upgrade and introduce new instrument models, in order to decrease inaccuracy, decrease imprecision, decrease sample size, decrease instrument and technician errors, increase speed of analysis, and improve ease of quality control. Such improvements may have increased the quality and ease of blood gas and pH measurements, but currently significant differences between models and between manufacturers continue.
Clinical Importance of These Findings
How important are these model differences in research or clinical practice? The difference between models often exceeds twice the standard deviation (SD) of a single model, and especially the SD of a single instrument. (The average SDs for the 20 models at all participating laboratories at the three analyte levels are 2.5 mm Hg for Po2, 1.3 mm Hg for Pco2, and 0.008 for pH; historically SDs are even less for individual instruments.) In research, it is obviously unwise for an investigator to shift between models without first ascertaining their comparability. In clinical practice, the answer to relevance depends on whether one is distinguishing between normality and abnormality, defining the degree of abnormality, or making changes in therapy. Expressed on an absolute basis by referring to Table 4, numerically higher measures are most likely and middle measures are least likely to be affected by model differences; expressed on a percentage basis, both low Po2 and Pco2 values are most likely to be affected by model differences. It would be comforting, but unwise, to ignore these model differences. Differences in or unrecognized alterations in laboratory instrumentation could importantly influence clinical decisions regarding the following: the presence of degree of impaired oxygenation in disability evaluation; the necessity for oxygen supplementation; and the management of respiratory failure or severe respiratory alkalosis.
These manufacturer and model differences suggest that laboratory directors should consider the similarities and differences between models when reporting data and also when retiring older instruments and adding new instruments to their laboratories. When doing so, the laboratory director can find specific information on probable model differences by examining the complete reports of their recent proficiency testing surveys. Otherwise clinicians may be misled by the values they receive in the same institution from different instruments.