Patterns of Dissimilarities Among Instrument Models in Measuring Po2, Pco2, and pH in Blood Gas Laboratories: Materials and Methods

12 Apr

We tabulated the mean values for Po2, Pco2, and pH values from each of the 30 lots for each of the manufacturers with the largest number of participating instruments (AVL Scientific Corporation, [AVL], Roswell, Ga; Corning, Chiron Diagnostics Corp [COR], Medfield, Mass; Instrumentation Laboratory [IL], Lexington, Mass; and Radiometer America Inc [RAD], Westlake, Ohio) and calculated their positive or negative deviations from AIM values. To ascertain if the analyses differed among manufacturers, we prepared three graphs, one for each analyte, with AIMs of each lot on the abscissa and the means of the lot deviations from the AIMs for each of the four manufacturers on the ordinate. Three of the four manufacturers had several models represented by >15 instruments in the database: COR with 7 models; IL with 6; and RAD with 6. To determine whether analyses differed among models made by the same manufacturer, after calculating model deviations from AIM values, we prepared nine additional graphs, one for each analyte for each of these three manufacturers, with AIMs of each lot on the abscissa and lot deviations from the AIMs for each of their models on the ordinate. canadian neighbor pharmacy

Statistical and Graphic Analyses
We performed statistical analyses to seek differences among manufacturers and among models made by a specific manufacturer. Three analyses of variance (ANOVAs) were performed (one for each analyte) to determine whether there were systematic differences among manufacturers in their deviations from the AIM for each of the 30 lots. Nine additional ANOVAs were used (three analytes times three manufacturers) to compare the 30 lot values of six or seven models of each of the three manufacturers with their own manufacturer mean lot value. Scheffe tests were used to define the probability of significant differences between specific manufacturers or models. Manufacturers or models were graded as follows: V=“very dissimilar” if they differed with p<0.001; D=“dissimilar” if they differed with p<0.05 to >0.001; and N = “not dissimilar” if they differed with p>0.05.