INTRODUCTION

Top Abstract Introduction Materials and Methods Results and Discussion References

Automated identification and susceptibility testing methods have become the mainstay of clinical microbiology laboratories and have provided laboratory staff with an efficient means of organism classification, thereby giving physicians the availability of rapid and generally accurate results that is necessary to guide therapy. Errors do occur, however, and as competing resources put more demands on shrinking laboratory staffs, expertise in recognizing errors and evaluating questionable results from these instruments is diminishing. This places a greater demand on the manufacturer to provide an instrument and a database that are as free from error as possible. However, the laboratory is no less accountable for undetected errors that may or may not result in an adverse outcome for patients. As new versions of databases become available, it is critical that appropriate evaluations be conducted in order to provide laboratory staff with unbiased data from which to make purchasing decisions. The update that resulted in the MicroScan Rapid Neg ID3 (RNID3) panel was designed to improve accuracy in the identification of common human clinical taxa by modifying the biochemical tests in the panel by replacing 10 tests on the RNID2 panel with new substrates. The new panel also eliminated the need for a mineral oil overlay on the decarboxylase tests and increased the shelf life from 6 months to 1 year when stored at 2 to 8°C. In this study, the MicroScan RNID3 panel was evaluated using members of the Enterobacteriaceae and some common gram-negative nonfermenters. (Part of this research was presented at the 99th General Meeting of the American Society for Microbiology, Chicago, Ill., 30 May to 3 June 1999.)

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results and Discussion References

Bacterial strains. Table 1 lists the 475 biochemically typical and atypical Enterobacteriaceae and Table 2 lists the 36 commonly isolated biochemically typical glucose nonfermenting gram-negative bacilli that were used in this study. All strains were of human origin. This set also contained larger numbers of genera and species not likely to be found in routine work but which were contained in the database, thereby presenting a true challenge to the system.

Media and biochemical tests. Biochemical tests for identification were performed using conventional media and the methods described by Edwards and Ewing (2), with some modifications by Hickman and Farmer et al. (3, 4). Commercial media were used wherever possible.

Identification panel. The RNID3 panel is manufactured by Dade Behring, Inc., MicroScan Inc., West Sacramento, Calif., and is designed for use with the Walk/Away automated instrument. The 36 substrates contained on the panel include arabitol, adonitol, lysine, and ornithine with the decarboxylase base, glucuronic acid, arabinose, urea, inositol, mannitol, mannose, melibiose, raffinose, salicin, sorbitol, sucrose, glucose, and tryptophan for indole production; 4-methylumbelliferyl (MeU)--D-xyloside, MeU--D-galactopyranoside, MeU--D-glucopyranoside, MeU-N-acetyl--D-glucosaminide, MeU--D-glucopyranoside, MeU--D-mannopyranoside, MeU--D-glucuronide, MeU--D-N,N'-diacetylchitobioside, MeU--D-galactopyranoside, MeU-phosphate (both acid and alkaline pH); L-arginine-7-amido-4-methylcoumarin (AMC), -L-glutamate-AMC, N-glutaryl-glycyl-arginine-AMC, -L-glutamic acid-AMC, L-pyroglutamic acid-AMC, L-proline-AMC, and L-tyrosine-AMC. These substrates work by one of the following mechanisms: hydrolysis of fluorogenic substrates, pH changes following substrate utilization, production of specific metabolic byproducts, or the rate of production of specific metabolic byproducts after 2.5 h of incubation in the Walk/Away. While this newer panel does require the automated addition of a rapid indole reagent to the tryptophan well, it no longer requires mineral oil overlays on the decarboxylases.

Definitions. The category of "correct" implies that the organism identification was correct to the genus and species levels at the end of the 2.5-h incubation time. The category of "correct with additional tests" implies that the organism identification was correct to the genus and species levels after the performance of additional off-line biochemical tests that are suggested by the manufacturer. These tests are suggested when the probability of accuracy of identifications of the first organisms listed is below 85% (low probability). There are 48 possible additional tests that would extend the time of final identification to 24 h. The category of "correct to genus" indicates that the genus identification was correct but the species was incorrect when the database indicated that the organism could be identified to the species level. The category of "error" implies that both the genus and species identifications are incorrect, and "no ID" indicates that no identification was given by the system or that the system gave a choice of multiple answers that were not resolved by additional testing.

Statistical analysis. Results were evaluated by chi-square analysis with Yates' corrected coefficient to arrive at a P value.

	RESULTS AND DISCUSSION

Top Abstract Introduction Materials and Methods Results and Discussion References

Table 1 shows the results of testing 475 strains of Enterobacteriaceae in the RNID3. Table 2 shows the accuracy of identification of five glucose-fermenting and nonfermenting organisms that are commonly isolated in clinical laboratories. At 24 h, after additional testing, the RNID3 accurately identified 88.8% of 511 strains evaluated in this study.

Table 3 lists the errors that occurred in identification. Twenty-six (5.1%) strains were misidentified. Of these 26 errors, 21 were at probability levels of 88% or greater; 14 were at probability levels of 95% or higher. Of the five Salmonella misidentifications, one was at a probability level of 76%, two were at a probability level of 90%, one was at a probability level of 95%, and one was at a probability level of 99%, suggesting that an inaccurate result might be submitted to the physician. Both of the Shigella misidentifications were at probability levels of 98% or greater. Because the system is based on fluorogenics, there are no conventional tests readily available with which to compare possibly incorrect reactions. Testing on nine strains yielded a result of "no identification" (very rare biotype).

Table 4 shows the results of testing 130 strains that more closely resemble the assortment of strains that might be routinely found in local area hospital laboratories. After additional testing was completed, 91.5% of the strains were correctly identified.

Bascomb et al. reported 97.1% accuracy with 405 fresh clinical isolates and 90.7% accuracy with 247 stock isolates (1). Schreckenberger et al. reported an accuracy of 96.2% after additional testing was completed and an error rate of 3.8% using 317 fresh clinical isolates (P. C. Schreckenberger, S. Connell, J. Skinner, B. L. Zimmer, D. Glenn, D. A. Bruckner, J. M. Janda, and S. L. Abbott, 98th Gen. Meet. Am. Soc. Microbiol. 1998, poster C-153, 1998). One might expect that the accuracy of the system when testing fresh clinical isolates might be higher than that observed with stock strains.

In an effort to determine the cause of the errors in identification, the strains for which incorrect answers were given were retested in duplicate. One panel was processed without oil overlays on the decarboxylases; a matching panel had the decarboxylases overlaid with mineral, as was the procedure with the RNID2 panels. One strain each of Escherichia coli and Escherichia hermannii gave correct identifications on the panels that were overlaid with oil, while the answers from the panels with no oil overlays remained incorrect.

In this study, the RNID3 panel performed as well for the identification of bacteria routinely isolated in hospital laboratories as it does for the identification of a challenge set of strains (P > 0.05). In addition, the accuracy of the new panel in this study was not significantly different (P > 0.05 [Yates' correction]) from the RNID2 results previously published (5), for which the same study set was used.

An in-depth study of the accuracy of this panel to correctly identify less commonly encountered glucose nonfermenting gram-negative organisms will be the subject of a future publication.