Detecting Errors in Blood-Gas Measurement by Analysiswith Two Instruments

CLIN. CHEM. 33/4, 512-517 (1987) Detectng Errors n Blood-Gas Measurement by Analysswth Two Instruments LouIs F. Metzger, Wllam B. Stauffer, Ann V. Kruplnskl, Rchard P. MIIlman,3 George S. Cembrowskl,2 and Allan I. Pack1 We performed a two-stage prospectve evaluaton of the error detecton capabltes of duplcate analyss of blood-gas specmens. In the frst stage we analyzed 161 specmens wth a Cornng Model 175 blood-gas analyzer as the test nstrument and a Cornng Model 178 analyzer as the reference nstrument, and n the second stage we analyzed 1544 specmens wth two Model 178 analyzers. In each stage the desgnated reference nstrument underwent troubleshootng whenever an analytcal error was detected; the test nstrument underwent troubleshootng only when error condtons were ndcated by means other than duplcate analyss. An error was consdered to have occurred t the dfference between the duplcate analyses exceeded.2 (for ph),.53 kpa,.e., 4 mmhg (pc2), or 7% (Po2). The number of specmens for whch errors were detected was 97 (6.1%) n the frst stage, 23 (1.5%) n the second. For each analyte more errors were detected wth the Model 175 analyzer (of older desgn) than wth the newer Model 178 analyzer. Furthermore, n certan perods assocated wth the use of partcular electrodes there were very hgh error rates for ndvdual analytes: 8% for 18% for p2. We conclude that duplcate analyss should be consdered as a possble requred standard for error detecton. Addtonal Keyphrases: qualty control. ph - analytcal-error assessment economcsof laboratoryoperaton The utlty of duplcate analyss of blood-gas specmens for the detecton of analytcal errors s a subject of current debate. On the one hand, some consder t an essental part of qualty control (1, 2), whle others tend to de-emphasze ts usefulness (3, 4). Part of the reason for ths dversty of vews s the lack of defntve data on the subject (5). The value of dual-nstrument analyss les n tscapablty to detect ntermttent ( random ) errors. Contnuous ( systematc ) errors should be detected by the use of control specmens, whch typcally are analyzed every 8 h. The purpose of the study was to characterze the type (contnuous vs ntermttent), frequency, and magntude of errors encountered n blood-gas analyses, and to determne whether duplcate analyss could be useful for detectng these errors. Because the results could depend on the specfc nstrument beng tested, we performed prospectvestudes of the type and frequency of errors wth two dfferent bloodgas analyzers, one of a recent desgn and one older. Methods Samples All blood specmens of at least 1.5 ml obtaned from patents wthout contagous dsease were accepted nto the study and were smultaneously measured on two blood-gas Cardovascular Pulmonary Dvson, and Department of Pathology and Laboratory Medcne, Hosptal of the Unversty of Pennsylvana,34 SpruceSt., Phladelpha, PA 1914. 3Current address: Rhode Island Hosptal, Pulmonary Dvson, Provdence,RI 292. Receved October 23, 1986; acceptedjanuary 14, 1987. analyzers. The volume requrement, whch allows the four analyses enumerated below, precluded the use of specmens contaned n 1-mL syrnges and capllary tubes, specmens that exhbt greater varablty than do the larger samples. Specmens from patents wth contagous dseases were excluded to avod unnecessary exposure of laboratory personnel. All technologsts partcpatng n ths study had receved tranng n theoretcal and practcal aspects of blood-gas analyss and had performed well n ths laboratory s annual profcency tests. Qualty Control For qualty control of the analyzers we used Confrm qualty-control reagent (Cba-Cornng Dagnostcs Corp., Medfeld, MA) and performed routne preventve mantenance, wth mmedate correctve mantenance n response to nstrument-generated error sgnals. Confrm s a tonometered aqueous buffer avalable n Normal, Alkaloss, and Acdoss formulatons for yenf\jng nstrument performance throughout a range of ph, P2, and Pco2 values. We used the mult-rule control protocol proposed for a three-level system by Westgard et al. (6). All three formulatons of Confrm were analyzed at least once on each 8-h shft. If the mult-rule control procedure was volated, the technologst checked the system, the frst step beng the re-analyss of the Confrm sample ndcatng out of control data; f the result was unchanged, the technologst nvestgated further, e.g., by removng bubbles or changng membranes or electrodes. Study Protocol The study conssted of two consecutve parts, each lastng approxmately 1 weeks. In Part A, we analyzed each of 161 specmens once wth a Model 175 and once wth a Model 178 Blood Gas Analyzer (both from Cba-Cornng), the former desgnated the test nstrument and the latter the reference nstrument. Specmens were ntroduced nto the analyzers wthout regard to order of njecton. Both nstruments were controlled as descrbed above, but n addton, we verfed the proper operaton of the reference nstrument f the duplcate (.e., two-nstrument) analyss protocol ndcated a malfuncton of that nstrument (see below for full detals). In contrast, f duplcate analyss ndcated a malfuncton of the test nstrument, we dd not ntate nvestgaton of the test nstrument untl that malfuncton was sgnalled by ether the bult-n error checks of the nstrument or results for the control materal. In a prelmnary study we checked for the presence of bas between the two models of analyzers. Only Pco, assays show meanngful bas. Ths dfference, whch s expected (Cba- Cornng, personal communcaton), was elmnated durng the study by applyng a correctng equaton to all Model 175 values for Pco, before calculatng the between-nstrument dfferences. Ths equaton was obtaned from lnear regresson analyss of data for 15 blood specmens analyzed n both the 175 and 178 analyzers: corrected pco2n kpa(model 175) = 1.128 x measured Pco2 n kpa(model 175) -.55. 512 CLINICALCHEMISTRY, Vol. 33, No. 4, 1987

In Part B of the protocol we analyzed 1544 specmens wth two Model 178 blood-gas analyzers, the one desgnated the reference nstrwnent beng the same as the reference nstrument n Part A of ths study. We compared the values obtaned for each varable (ph, Pco2, po2) wth the test and reference nstruments for each blood specmen as follows: ph = test ph - ref. ph Pco2 (kpa) = test pco2(kpa) - ref. pco2(kpa) Po2 (%) = [(test Po2 - ref. po2)/ref. Po2l X 1 If the dfference between these nstruments exceeded our chosen ranges of acceptablty, based on those of Elser et al. (2)-pH ±.2, Pco2 ±.53 kpa,.e., ± 4 mmllg, or po2 wthn 7% of the Po2 as measured by the reference nstrument-we consdered the results to be n error. f such an error occurred, we followed the protocol shown n Fgure 1. The frst step of ths protocol was to analyze the specmen agan, once wth each analyzer. If repeatng the analyss on both nstruments gave results wthn acceptable lmts, we assumed that an ntermttent error had occurred and took no further analytcal steps. An error was defned as ntermttent f, on re-analyss, the between-nstrument dfference was acceptable or f, on analyss of control materal, no error was found. The ntermttent error was assgned to the nstrument that n the repeat analyss showed the greatest absolute change toward the average of the repeat results. If both nstruments showed equal changes, we consdered the stuaton ambguous and dd not assgn an ntermttent error to ether nstrument. The subsequent steps were as llustrated n the Fgure. When the error was assgned only to the test nstrument, troubleshootng was not performed, but rather, the study contnued untl (a) the error condton spontaneously dsappeared; (b) the test nstrument, through ts own bult-n error-detecton system, sgnalled an error condton; or (c) a routne scheduled run of Confrm detected an out-of-control condton. When the Confrm data were out of lmts on both nstruments, the error was taken to be contnuous (as opposed to ntermttent) and was assgned to both nstruments; n ths event, the reference nstrument was brought nto control, but the test nstrument was allowed to reman out of control untl the error condton was sgnalled by other than dscrepant between-nstrument duplcate measurements. CRITERIA FOR 5ENLL IPISTRUREOT ERROR EXCEllED 1OCT1 NCAIO ON TEST XXII REFERENCE INSTNVNEXTS XIS RESULTS ACCEflANLE? VIE I) IATENAIVFRNIT ERROR ASSIGNEE TO INSTNIOT SANdING I, CREATES? AIXAIVIE VALUE ANALYZE CONTSOL MATERIAL OS TOWARD AVERAGE OF NEW VAlUES. 1111 LIMITS? I.. AM 515 2) - INTERXITTEFF TO ID ERROR IRSTRUJUIITS. 55ICEFO RE YOWl.IIOAOIRC. MO AN TI., - I) COOTIS 131 ASSIGNED TO TEST. NO TAOUELE$IIOOTIIC. IV ON RI.. a, T.. A) COOTINUGJV REFURONCE. ERROR ASSICIER TROURLEST TO - RN ROth 5) CNITIIOJRJS ERROR ASSIGNEE - TO ROTS TEST AXE REFERENCE. TROURLEST REFENEECI. Fg. 1. Protocol flow chartfordetermnatonoftypeoferror (contnuous or ntermttent) encountered and the nstrument responsble for the error (test or reference) Data Analyss We kept complete records of all analyses, and recorded the seral numbers of the electrodes used throughout the study. For statstcal analyses of these data, we used the BMDP statstcal software programs (7). Wthn-nstrument dfferences. We calculated the wthnnstrument precson for blood specmens from the betweenrun dfferences obtaned when duplcate analyses were done wth the same nstrument. We used only those data for whch there was not a between-nstrument error.the SD of the dfference was calculated accordng to the equaton: SD = V d2/2n (4). The pared t-testwas used to assesswhether for each analyte for each analyzer the mean dfference between runs dffered sgnfcantly from zero (p <.5). Between-nstrument dfferences. We calculated the between-nstrument precson of analyss of blood, to evaluate whether our crtera for between-nstrument errors were reasonable. For ths we used the mean and SD of the dfference between the nstruments, after frst elmnatng outlers that exceeded three SDs from the ntal mean. The pared t-testwas used to assesswhether dfferences between the two analyzers were sgnfcantly dfferent from zero (p <.5). We calculated the overall frequency of errorsn specmens by dvdng the number of specmens havng errors n one or more analytes by the total number of specmens. To determne the frequency of errors n results for a gven analyte, we dvded the number of errors observed for that analyte by the total number of specmens. Usng ch-square analyss (4), we compared both the total number of specmens wth errors and the number of errors for each analyte n Part A wth that n Part B. We also examned the relaton between the frequency of errors for each analyte and the specmen accessonnumber, to see f errors were clustered nto groups or dspersed randomly throughout the course of the study. For ths, we dvded our study nto batches of 1 specmens and calculated the number of errors n each batch. We notced clusters of errors that appeared to be related to specfc electrodes, and therefore we used a ch-square analyss to compare the error rate for each electrode wth that for all other electrodes consdered together. Clncal-correlaton studes. Whenever we assgned an error to the test nstrument, the results from the test and reference nstruments were nterpreted separately. We classfed the acd-base status of each accordng to a modfed verson of the Goldberg nomogram (8). We called an error clncally sgnfcant f there was a dfference n these acdbase classfcatons. For example, a Pco2 error that shfted the acd-base classfcaton from metabolc alkaloss to acute respratory alkaloss was called sgnfcant. For P2, one of us (R.P.M.) determned whether the dfferent Po2 values would have resulted n a change n patent management, such as a change n the prescrbed concentraton of nspred oxygen or a request for another specmen for bloodgas assay. We called the error clncally sgnfcant f such a change would have occurred. Results Wthn-nstrument dfferences. We compared the means and SDs for Confrm for each of the analytes (ph, Pco2 P) n each of the three formulatons wth the values obtaned from all laboratores usng the same lot of materal and same model of analyzer, the latterdata beng provded by the manufacturer (Confrmaton Statstcal Analyss, Cba- CLINICALCHEMISTRY, Vol. 33, No. 4, 1987 513

Cornng). We found our mean values to be wthn 2 SDs of the means of all laboratores and our SDs to be less than twce thers. The wthn-nstrument precson for blood was smlar for all nstruments used n ths study (see Table 1). Snce the same Model 178 was used as the reference nstrument n both parts A and B, the data obtaned wth ths nstrument were pooled. A small but statstcally sgnfcant bas between the frst and second runs was found for Pco2 on the test nstrument of Part A and for P2 on all nstruments. Between-nstrument dfferences. Sgnfcant between-nstrument dfferences n non-erroneous measurements ndcate the presence of between-nstrument bases that are extremely small (see Table 2). The proporton of specmens for whch duplcate analyses fell outsde of the acceptable range was hgher n Part A (6.1%) than n Part B (1.5%). Ths overall dfference n the number of specmens wth errors was sgnfcant (p <.1) by the ch-square test. The number of errors assgned to the fve categores ndcated n Fgure 1 was for Part A: 1) 126, 2) 18,3) 14,4), and 5). For Part B the number of errors n each category was: 1) 35, 2), 3) 1, 4), and 5). For each of the analytes consdered separately, there were sgnfcantly more errors n Part A than n Part B (p <.1) by the ch-square test. These errors are shown n more detal n Fgure 2a. We found 16 ph errors n Part A. Of the 1 assgnedto the test nstrument (error rate =.6%), eght were ntermttent and two contnuous. In Part B of the study no ph errors were found. For Pco2, 76 errors were found n Part A (Fgure2b). Of these, 42 were assgned to the test nstrument (error rate = 2.6%); 4 ntermttent, 2 contnuous. A total of 25 Pco, errors were found n Part B. Of these, 14 were assgned to the test nstrument (error rate =.9%): 13 ntermttent, and one contnuous. When P2 was examned, there were 84 errors n Part A (Fgure2c).Ffty-onewere assgned to the test nstrument (error rate = 3.2%): 41 ntermttent, 1 contnuous. A total of 11 po2 errors were found n Part B. Nne of them, all ntermttent, were assgnedto the test nstrument (error rate =.6%). When errorfrequences were relatedto specmen-accesson number there were no peaks n the error rate for ph n ether part of the study. For Pco, (Fgure 3, Panel I) there was an error peak of 8% between specmens number 21 and 3 n Part A. No error rates of 5% or over were observed n Part B. In a smlar manner, the dstrbuton of p2errors relatve to the specmen-accesson number s shown n Fgure 3, Panel II. In Part A there was an nterval between specmens number 21 and 3 when there was a partcularly large cluster of errors, resultng n an error rate of 18% for that nterval. Furthermore, n the 121 to 14 range there s another peak, reachng a 5% error rate. In Part B, there was an error rate of 6% between specmens 1 and 1. Clncal-correlaton studes. The number of clncally sgnfcant errors n Part A were as follows: three of 1 ph errors, 26 of 42 Pco2 errors, and nne of the 51 Po, errors. In all, 39 of the 13 errors (37%) attrbuted to the test nstrument were clncally sgnfcant. In terms of the number of specmens for whch there was an error for at least one analyte, 39 of the 97 specmens (4%) contaned clncally sgnfcant errors. In Part B, fve of the 14 errors n Pco2 and none of the nne errors n Po, were clncally sgnfcant. That s, fve of the 23 errors (22%) n Part B were clncally sgnfcant. For none of the specmens n Part B were there errors n results for more than one analyte, so the error frequency remans at 22% when consdered n terms of specmens wth one or more errors. Correlaton between electrode use and error prevalence. Two electrodes had a sgnfcantly greater number of errors than ther cohorts. One was a po2 electrode from the test nstrument n Part A (p <.1) that was n place durng the hgh Po, error perods of Part A (specmens 21 to 3 and 121 to 14). The other was a Po2 electrode from the test nstrument n Part B (p <.5) used durng the hgh P2 error perod at the begnnng of Part B (specmens 1 to 1). Three of four perods wth error rates of 5% or greater were assocated wth the use of specfc electrodes. Note, however, that the frequency of errors produced by these electrodes was varable,.e., they were an ntermttent source of error; they performed satsfactorly after the perods of hgh error prevalence. Table 1. Wthn-Instrument Dfferences: Blood-Gas Analyses Part A, Test Model (175) Part B, Test Model (178) Parts A & B, Ref. Model (178) No. of No. of No. of Mean SD specmens Mean SD specmens Mean SD specmens ph.1.21 129.8.34 34.1.22 143 kpa _.58.13 69 -.3.7 9.2.14 82 2 % l.44 1.76 73 O.8S.82 22 1.4812 1.7 98 a,b values for Run I vs Run 2 by same nstrumentdffered sgnfcantlyfrom zero ( p <.1, 3p <.1). The numberof specmensanalyzedwth the referencenstruments not equalto the sumof those analyzedon thetwo test nstrumentsbecauseoutlers >3 SD from the mean were excluded. See Methods. Table 2. Between-Instrument Dfferences: Blood-Gas Analyses PartA No. No. Mean SD specmens Mean SD specmens ph.43 153.12a.31 pco2 kpa 2b.16 1518.12 %.488 2.33 1541 -.2 1.49 NbMean dfference between the analyseson the test and the reference nstrumentwas sgnfcantlydfferent from zero (bp <.1, 8p <.1). The numberof specmens for the dfferent analytes n each part varesbecause outlers>3 SD from the mean were excluded. See Methods. PartB 1465 1453 1437 514 CLINICALCHEMISTRY, Vol. 33, No. 4, 1987

D I C w I E &-2.4 A I I II -.4 7. 7.2 7.3 7.4 7.5 7.6 2, mlu REFERENCE ph (ph unts) A II qu I a b PANEL I NUMBER 1 OF PCO2 NUMBER OF PCO2 5 PANEL U I I I I I o 4 9 12 16 E-41 o 4 9 12 15 SPECIMEN ACCESSION NUMBER A B 2 A L.I 1 U a- 2 B Is NUMBER OF P2 1 S n[-,rf[111 4 5 12 16 z 122 I22 I.- w 1212 5.4-2 p p 2 4 6 8 1 REFERENCE Pco2 (mmhg) 4 2 J. I PAR1A I II I 8 1 I II I I 4 12 2 2 14 12 112 2 3 4 12 12 REFERENCEP2(mmH) Fg. 2. Dstrbutonof errorsn ph,, and z that weredetectedby duplcateanalyssand thatwereassgned to the testnstrument Intermttent errors are desgnated by sod lnes, contnuous errors by brsken lnes.the heght of eachlnendcatesthemagntudeof onespecfcerror Dscusson In ths study we addressed the queston of how many errors n the measurement of blood-gas tensons and ph are detected by dual-nstrument analyss that would not be detected wth a sngle-nstrument analyss. The results ndcate that, even n a well-controlled laboratory, there are perods when errors may be rather frequent, and durng these perods many of the errors may be clncally sgnfcant. We also found that errors occurred more frequently wth an older blood-gas analyzer than wth one of newer desgn. In consderng the results of our study we frst dscuss c 1 NUMBER OF P2 h o 4 9 12 16 SPECIMEN ACCESSION NUMBER Fg. 3. Tme dstrbuton of 2 and D2 errors assgned to the test nstrumentthroughout the study thoseaspects related to study desgn. Our laboratory proce. dure s typcal of many laboratores n that we use three formulatons of commercally prepared aqueous buffer as the qualty-control materal. We took partcular care to desgnate one nstrument the reference nstrument and the other the test nstrument. We dd sobecause we ntended to dentfy the true ncdence of errors that would have occurred f duplcate-nstrument analyss had not been performed. Thus troubleshootng was not performed on the test nstrument, even f duplcate analyss suggested a malfuncton, because ths would have falsely decreased the number of errors detected by two-nstrument analyss. Rather, we dd troubleshootng only when a malfuncton was sgnalled by other means, e.g., f the nstrument was seen to be out of control when control materal was analyzed or by the bultn error checks of the nstrument. In the absence of duplcate analyss these are the only ways that would allow detecton of nstrument malfuncton. In contrast, we felt t mportant to mantan optmal functonng of the reference nstrument, Therefore ths analyzer had troubleshootng performed whenever errors were detected, ncludng those detected by duplcate-nstrument analyss. Furthermore, we used the same Model 178 blood-gas analyzer as a reference n both Parts A and B, sothat results n the two parts of the study could be compared. B CLINICALCHEMISTRY, Vol. 33, No.4, 1987 515

We also had to decde, n plannng the study, what would consttute an unacceptable dfference between the duplcate analyses. Our crtera-±.2 ph unts, ±.52 kpa,.e., ± 4 mmhg for Pco2, ± 7% for p2-were based on prevously suggested values (7). Comparson of the crtera establshed before the study wth the wthn-nstrument standard devatons that were actually found (Table 1) ndcates that our lmts were n fact very conservatve. A large analytcal error, exceedng three to four SDs, s requred to exceed our lmts; therefore t s unlkely that a result was called an error when n fact t was a vald measurement. Wth ths desgn we detected a relatvely low overall prevalence of errors for each analyte, the hghest error rates beng for Pco, (2.6%) and po2 (3.2%), both n Part A of the study. Furthermore for all analytes, errors were more common wth the older (Model 175) analyzer. Most errors were ntermttent (89% of the total). Furthermore, the vast majorty of these ntermttent errors fell nto category 1 of Fgure 1,.e., the error was not reproduced n the second analyss. Ths dstrbuton emphaszes the fleetng nature of the ntermttent error. Such errors wll not be detected by means other than analyss of samples wth two nstruments. Although the overall prevalence of errors was low there were, as mentoned, tmes durng the study when the error rates were mugh hgher. Error rates durng these ntervals can be dsturbngly hgh, and ther frequency may well mpact on patent care. In consderng ths, however, t s mportant to consder not only the frequency of errors but also ther magntude. Thus one must defne what magntude of error to call clncally sgnfcant. Ths depends on the purpose of the analyss, e.g., montorng changes n blood-gas tensons vs populaton screenng. The former, whch concdes wth our purpose, requres greateraccuracy and precson than does the latter (9). Rather than desgnatng specfc numercal lmts to defne the term clncally sgnfcant, we chose a functonal defnton: f an errorn ph or Pco2 resulted n a change n nterpretaton or f an error n measurement of po2 would have resulted n a change n patent management, the error was called clncally sgnfcant. Thus, durng perods of hgh error prevalence there may well be a hgh proporton of mstakes that adversely affect patent care. We cannotdefntely dentfy the cause of these perods of hgh error rate. We consder t lkely, however, that they reflectmalfuncton of partcular electrodes, because some electrodes had a sgnfcantly hgher error rate than others (see Results). However, ths does not represent defntve proof, because a hgh error rate owng to any cause would necessarly be attrbuted to some electrode. Moreover, these same electrodes were used durng other parts of the study wthout producng unusually hgh error rates. Although our study does not reveal the cause of the ntermttent errors, t does ndcate that even wth modern analyzers and current qualty-control technques such errors do occur, as prevously shown by Leary et al. (1). These authors reported that erroneous results were detected n 3.9% of specmens analyzed. However, our study desgn dffered from thers n two mportant respects. Frst, we desgnated one analyzer the test nstrument so that a true prevalence of errors could be determned. More mportantly, we analyzed not only the overall rate of errors but also how the rate vared durng the study. Thus we dentfed perods of much hgher error rate (up to 18%). Ths adds weght to the concluson reached by Leary et al. that duplcate analyss s ther prncpal mechansm of random error control. These data lead naturally to the queston of whether concurrent analyss wth two separate nstruments should be made part of the routne qualty-control procedure n the blood-gas laboratory. Before consderng ths t s prudent to examne whether our laboratory could be consdered less well controlled than the typcal laboratory. If ths were true, then our experence mght overestmate the error prevalence of others. We beleve ths not to be the case, for the followng reasons. The technologsts nvolved had all receved n-house tranng and had passed annual wrtten and practcal examnatons. Full-tme technologsts staff the day, evenng, and nght shfts, wth part-tme staff fllng n for weekends, etc. Ths latter group was also extensvely traned and met our examnaton requrements. Our analyzers are relatvely modern; the Cornng 175 beng fve years old and the 178 two years old. Durng the course of the study our analyss of qualty-control materal was smlar to that for other smlar nstruments. The means of our analyses were wthn 2 SDs of the mean of all lke nstruments. Furthermore, our SDs were slghtly smaller than the group devatons for all values for analyte, except for the hgh pco2, for whch our standard devaton was slghtly larger. Thus the frequency of errors we experenced s not lkely to be nflated n relaton to other laboratores usng smlar equpment and smlar control materal. In fact, t may be that we experenced fewer errors than mght occur n other laboratores, because we used a mult-rule qualty-control program, whch s potentally more senstve n detectng errors than are other protocols. Furthermore, the frequency of errors noted n ths study s probably lower than f we had ncluded less-deal samples, those wth volumes less than 1.5 nl. Because such duplcate analyss s the only procedure that wll detect ntermttent errors, t s a potentally mportant component of an optmal qualty-control strategy. Obvously t wll add to drect laboratory cost, but costs can be revewed n a broader sense. In partcular, what s the hdden cost of reportng an erroneous result? Wll a decson negatvely affectng patent care be made? Wll a second specmen be drawn and analyzed? Nonetheless, n ths tme of concern about cost t seems napproprate to smply add duplcate analyss to exstng qualty-control strateges. Rather, there s a need to consder the two major strateges-duplcate analyss and the use of qualty-control materal on a regular bass-and to develop an optmal combnaton of these that wll lead to suffcent error detecton at acceptable cost. Our study provdes nformaton as to the number of errors detected by dual-nstrument analyss. We dd not study the effcacy of duplcate analyss wth a sngle nstrument. However, Lunetsky et al. (1) evaluated ths by computer smulaton and showed unsatsfactory error-detecton capabltes. A study smlar to the present one s needed, to evaluate the utlty of runnng qualty-control materal once per 8-h perod, n a settng where duplcate analyss s and s not routnely performed. Then the results of the two studes could be used to objectvely assess the effcacy of both types of qualty-control strateges, and then to desgn an optmal qualty-control protocol. In concluson, n the present study we have demonstrated that even n a laboratory wth hghly traned staff and wth analyzers meetng qualty-control standards, ntermttent errors of clncal sgnfcance occur, at tmes frequently. Only duplcate-nstrument analyss can detect ntermttent errors, so t should be consdered as a requred routne 516 CLINICAL CHEMISTRY, Vol. 33, No. 4, 1987

procedure. However, there s a need to combne ths n some fashon wth other strateges to provde the optmal lowest cost approach to effectve qualty control n the blood-gas laboratory. Many of our laboratory staff worked on ths study. For ths we thank all of the technologsts nvolvedn data collecton and L Peters for the multple retypng of ths manuscrpt. Also, we are grateful to Robert F. Moran (Cba-Cornng) for many helpful suggestons. References 1. Leary ET, Delaney CJ, Kenny MA. Use ofequlbrated blood for nternal blood-gasqualty control. ClnChem 1977;23:493-53. 2. Elser RC, Stler J, Garver C. A flexble and versatle program for blood-gasqualty control. Am J COnPathol 1982;78:471-8. 3. Ross JW. Blood gas nternal qualty control. Pathologst 198;34:377-9. 4 Barnett RN. Clncal laboratorystatstcs. 2nd ed.boston: Lttle, Brown and Co., 1979. 5. Mohler JG, Coller CR., Brandt W, Abramson J, Verkak G, Yates S. Blood gases.in: Clausen JL, ed. Pulmonary functon testng gudelnes and controverses New York: Academc Press, 1982:223-57. 6. Westgard JO, Moran RF, Groth T. Performance valdaton of blood gas nstruments usngthree levelsofcontrol for ph, p and p: procedure and assessmentby nteractve computer smulaton. In: Sggaard-Andersen, ed. Proc. 5th meetng, Internatonal Federaton of Cln Chen Expert Panel on ph and Blood Gases. Copenhagen: Prvatepress,1981:115-37. 7. Dxon WJ, ed. BMDP statstcal software, 1983 prntng wth addtons.berkeley-los Angeles-London: Unversty of Calforna Press,1983. 8. Goldberg M, Green SB, Moss ML, Marbach CB, Garfnkel D. Computer-basednstructons and dagnossofacd-basedsorders. J Am Med Assoc 1973223:269-75. 9. Harrs EK. Statstcal prncples underlyng analytc goal-settng n clncal chemstry. Am J Cl Pathol 1979;72:374-82. 1. Lunetsky ES, Cembrowak CS, Metzger LF. Evaluaton of patent duplcatesfor the qualty control of blood gasanalyzers:a computer smulaton approach. Submtted for publcaton. CLINICALCHEMISTRY, Vol. 33, No. 4, 1987 517