2014

**Comparisons Between Manufacturers**

The three graphs comparing manufacturers with each other for each analyte (Figs 1-3) visually demonstrate the differences between manufacturers along the complete range of intensities. Table 1 indicates the degree of the statistical dissimilarity between manufacturers for each analyte using the three traditional ANOVAs. For the 18 paired analyte comparisons, only the AVL and RAD mean Po2 values are not graded V. In Figure 1, however, even for this comparison, their visual dissimilarity can be seen, because a “crossover” pattern occurs with AVL deviations consistently higher than RAD values at Po2 intensities below 80 mm Hg, and consistently lower than RAD deviations at Po2 intensities above 110 mm Hg. Seeing this pattern, we performed an unscheduled ANOVA using the 10 lots with Po2 values below 79 mm Hg. Scheffe tests showed that all four manufacturers produced V results for these 10 lots, including the RAD and AVL comparison.

For Pco2 comparisons (Fig 2), AVL deviations are consistently positive, RAD deviations are consistently negative, and the COR and IL patterns cross. For pH comparisons (Fig 3), the AVL and IL deviations are consistently negative and different while the COR values are consistently slightly more positive than the RAD values. Thus, differences between each manufacturer for all analytes are evident both graphically and statistically. flovent inhaler

The nine graphs from three manufacturers were studied carefully. To demonstrate the importance of graph analyses, three simplified graphs (with only four instruments per graph for clarity) were selected (Figs 4-6). They illustrate dissimilarities among the models of each manufacturer for each of these three analytes.

*Figure 1. Po2 measures of FCE by blood gas analyzers of four manufacturers are compared. Mean values are derived from approximately 75 AVL analyzers, 360 COR analyzers, 300 IL analyzers, and 180 RAD analyzers for each of 30 lots. Deviations from each lot AIM values are on the ordinate. Each pattern is different by ANOVA. See the first paragraph of the “Results” section regarding the statistical confirmation of these differences.*

*Figure 2. Pco2 measures of FCE by blood gas analyzers of four manufacturers are compared. Values are derived from the same lots and analyzers as in Figure 1. Differences between manufacturers range from *2.5 mm Hg at low Pco2 intensities to 5.5 mm Hg at high Pco2 intensities. Each pattern is different visually and statistically.

*Figure 3. pH measures of FCE by blood gas analyzers of four manufacturers are compared. Values are derived from the same lots and analyzers as in Figure 1. The range of differences between manufacturers is approximately 0.03 U over the entire pH range. Each manufacturers mean values are quite consistently related to each other. Each pattern is different visually and statistically, even for the COR and RAD models. Contrast this consistency with the COR 238 and 278 model patterns shown in Figure 6.*

*Figure 4. Four RAD models compared for Po2. Fifteen to 41 analyzers of each model determine each deviation from the AIM (representing >900 instruments) values for 30 lots of FCE proficiency testing materials for Po2 values from 43 to 175 mm Hg. The 520 model diverges from the relatively similar 3 and 30 models and dissimilar 300 model as Po2 intensity increases above 90 mm Hg.*

*Figure 5. Four IL models compared for Pco2. Some models demonstrate a “crossover” pattern. Thirty to 100 analyzers of each model determine each deviation from the AIM (representing >900 instruments) values for 30 lots of FCE for Pco2 range of 22 to 72 mm Hg. The 1306 and 1312 models are visually similar, but the 1420 model is clearly different. However, this difference is not statistically significant using data from the full range of Pco2 values for the ANOVA, because the mean deviations for the 1306 and 1312 models are only 0.2 to 0.4 mm Hg, respectively, different from the 1420 models. When the ANOVA is restricted to the upper half of Pco2 intensities, it confirms that the 1420 model is V from the 1306 and 1312 models, with mean deviations of 3.1 and 3.3 mm Hg, respectively, from the 1420. The 1620 model is visually different and V from the other three models.*

*Figure 6. Four COR models compared for pH. Eighteen to 170 analyzers of each model determine each deviation from the AIM (representing >900 instruments) values for 30 lots of FCE for pH range of 7.132 to 7.616. The 278 model values are visually lower than the other three models by <0.01 U on average, but are V from them. The 178 and 840 models are minimally dissimilar. The 238 model values visually “crossover” the 178 and 840 values. Using the full range of values for ANOVA, the 238 does not statistically differ from the 178 and 840 models. However, using pH values above 7.40 for ANOVA, the 238 model is V from the 178 model.*

**Table 1—Dissimilarity Between Blood Gas Analyzers by Manufacturer**

COR | IL | RAD | |

AVL | WV | WV | NW |

COR | wv | WV | |

IL | wv |