LYDIA NELSON, M.S., MT(ASCP), SH, SAMUEL CHARACHE, M.D., SANDRA WINGRELD, MT(ASCP), AND EDWARD KEYSER, MT(ASCP)

Transcription

1 Laboratory Evaluation of Differential White Blood Cell Count nformation from the S-Plus V and H-1 in Patient Populations Requiring Rapid "Turnaround" Time LYDA NELSON, M.S., MT(ASCP), SH, SAMUEL CHARACHE, M.D., SANDRA WNGRELD, MT(ASCP), AND EDWARD KEYSER, MT(ASCP) Automated differential counts produced by the 8 S-Plus V (S + ) and the H-1 (H-1) were compared with routine and reference manual differentials with the use of samples from the adult emergency room (ER) and the neonatal intensive care unit (NCU), populations in which rapid reporting of laboratory results is considered important. Error rates for routine technologists were 9.% in the ER and 15.% in the NCU. Error rates for the two instruments were higher than those for technologists with samples from the ER but could be reduced to -5% if instrument flags and additional criteria were used to signal the need for a conventional differential. nstrument error rates were higher yet with NCU samples, and specificity was very low (1% for each device). There were small differences between the instruments in detection of immature neutrophils, butflagsfrom the H-1 were more specific (except for detection of nucleated red blood cells in samples from the ER). f either instrument were used in an adult ER and flags and additional criteria were used to signal the need for conventional differentials, 6-75% fewer manual counts would be performed with no decrease in accuracy and a considerable improvement in turnaround time. (Key words: Differential leukocyte count; Automation and flow cytometry) Am J Clin Pathol 1989;91: AUTOMATED HEMATOLOGC NSTRUMENTS such as the S-Plus V (S + ), H-1 (H-1), and others 281 " 1518 are capable of generating multiparameter blood counts and screening differentials in approximately 1 minute or less. Because of inherent analytic variability and cost of traditional eye-count differentials, many laboratories are relying on instrumentgenerated differential leukocyte counts to reduce operating costs and provide rapid turnaround time. Numerous studies have evaluated the reliability and efficiency of automated electronic and cytochemical leukocyte differentials in adult populations, ' - 6 ' 7-9 ' "' 516 but few studies have evaluated such differentials in children. 512 n this study we have investigated two patient populations in which rapid reporting of test results is considered critical: the adult emergency room (ER) and the neonatal intensive care unit (NCU). Our examination primarily addresses the question of whether automated differentials can be technically as accurate or contain the same infor- Received June 2, 1988; received revised manuscript and accepted for publication August 1, Address reprint requests to Dr. Charache: The Johns Hopkins Hospital, 6 North Wolfe Street, Meyer B121D, Baltimore, Maryland Departments of Laboratory Medicine and Medicine, The Johns Hopkins Medical nstitutions, Baltimore, Maryland mation as manual examination of the traditional blood film ("eye count"). f the latter is true, for the particular patient populations studied, we asked if white blood cell (WBC) "flags" associated with the automatic counters are reliable guides to the need for traditional differentials. We reassessed the method currently in use in our laboratory, which combines a manual scan of the film with the use of an automated microscope (see below). We also asked if automated differentials based upon histochemical staining and light scattering (five-part differentials performed by the H-1) are more useful than those based upon electrical impedence (three-part differentials performed by the S + ). Because automated instruments "flag" questionable results to indicate a need for further study, are the flags reliable enough to safely curtail the use of traditional differential methods, while providing the answers to questions considered important by clinicians? To assess the clinical utility of electronic and cytochemical differential information, for a period of three months, random blood samples from the ER and NCU were run on both the S + and the H-1. These samples were compared with microscopic differentials, for correlation of results, instrument sensitivity and specificity, as well as reliability and accuracy of instrument-generated WBC flags ( "R" flags and "NRBC," "blast," "left shift," "immature granulocyte," and "atypical lymphocyte" flags). Such a comparison of two hematology instruments, using different principles of operation, 61 provides insight not only into problems of instrument operation, but also problems of unique patient populations. Because the study addressed differential WBC counts, rather than examination of blood smears, assessment of red blood cell (RBC) morphologic characteristics by either technologists or automated devices was not examined. Determinations of erythrocyte size, shape, and hemoglobin content are an essential part of patient management, and omission of such data from this study should not be construed as denial of their importance. 56

2 56 NELSON ET AL. A.J.C.P. «May 1989 ER staff collects K EDTA Vacutainer or NCU staff collects 2 K 2 EDTA Microtainers Run sample on S + Run sample on H-1 Prepare blood film Mlnlprep. Stain slide. Perform 2-cell differential on Hematrak 6 ("routine differential" see text) Slide Review One of eight reference technologists does a -cell scanning review of reported differentials. Use 1-cell differential limits (1) as a guide and perform 1-cell eye count differential when discrepancies occur. Reference technologists resolve identification of questionable cells. Use reviewed differential as "truth" for instrument evaluation. FlG. 1. A flow chart of the steps involved in analyzing differential data. Blood Samples Materials and Methods Seven hundred seventy-six ER samples of venous blood were collected in K EDTA (Vacutainer, Becton Dickinson Corporation), whereas 176 neonatal blood samples were collected in K 2 EDTA (Microtainer, Becton-Dickinson). Over a three-month period, the samples were collected from adult and neonatal patients (gestational age, 2-1 weeks) with a wide variety of diagnostic problems. The adult samples were almost always collected by venipuncture, whereas pediatric samples were collected by heel stick or from umbilical venous catheters. All samples were collected only when requested by the physician caring for the patient. The time between sample collection and analysis was between minutes and four hours for most specimens. Sample Handling From the time of sample collection to testing, all samples were maintained at room temperature. Typically, two Microtainers were collected from each child: one was used for the S + count and the other for the H-1, with the blood film pulled at random from one of the Microtainers. Both instruments were maintained according to manufacturers' guidelines. The S + was calibrated against fresh blood, and the H-1 was calibrated against the S +. The S + calibration was checked every eight hours against samples that had previously been assayed by reference methods, whereas quality control of the H-1 was maintained with materials supplied by the manufacturer as well as carryover samples previously analyzed on the S + counter. n our laboratory, rather than routinely performing traditional (or "eye count") differentials on all blood films, differential leukocyte counts are routinely performed on Wright-stained blood films with a Hematrak 6 (Geometric Data). Before starting the Hematrak count, a technologist scans the blood film for quality of preparation and staining. f they are satisfactory, the instrument attempts to classify 2 WBCs. f the technologist's impression of the scanned film disagrees with the instrument count, or if abnormal cells are present, a 1 or 2-cell manual differential is done. t became evident early in the study that incomplete lysis of neonatal red blood cells in the peroxidase channel of the H-1 produced inaccurate WBC data in more than half of the samples (65 of 99). There were no problems with red blood cell (RBC) lysis in the basophil channel, and if initial WBC counts from the two channels disagreed, and there was enough blood remaining, the sample was diluted 1:1 with saline and reassayed. The diluted neonatal samples generally generated reliable WBC count and differential results. Data Analysis Reference Method. To provide a basis ("truth") for comparing electronic and cytochemical differentials with traditional differentials, all routine differentials were reviewed by one of eight experienced technologists. At least WBCs were scanned 17 ; in cases of leukopenia, blood smears were scanned for 5 minutes. f the reviewer's estimate for any cell line was outside the limits that might be expected from sampling errors, 17 a 1-cell eye-count differential was performed: in cases of questionable cell identification, a third technologist was used to settle disputes (Fig. 1). Errors on the original differential reports were tabulated and the frequency of laboratory errors calculated. The methods used to assess the automated differentials differ from those recommended by the National Committee on Clinical Laboratory Standards. 19 Those recommendations are not well-suited to conditions in our laboratory, and alternative techniques were devised to meet our physical andfinancialconstraints.

3 Vol.91 No. 5 EVALUATON OF DFFERENTAL WHTE BLOOD CELL COUNT Correlation and Regression Analysis. Classic correlation and linear regression analyses were performed on WBCs, RBCs, hemoglobin, and platelet counts as well as differential leukocyte data from all adult emergency room samples. The NCU data were not analyzed in this fashion because more than half the samples had to be diluted and almost all samples had abnormal WBC distributions (i.e., differential counts were flagged or not displayed). Overall Efficiency. Sensitivity, specificity, and efficiency for detection of abnormal samples were determined for the two automated systems. Reference differentials were considered normal if they did not violate any of the criteria in Table 1. S + data were judged abnormal if the instrument generated aflagor backlighted a parameter, and H-1 data were considered abnormal if a WBCflagor star* was generated. f no flags, backlighting, or stars were displayed, the differential was considered normal by instrument standards. Reliability of s. The final instrument evaluation involved a study of the reliability of individual WBC flags. Comparisons were made between the reference differential, impedence differential, and the cytochemical differential. Each instrument-generated WBCflag (or lack of one) was classified as true or false positive or true or false negative, based on the criteria listed in Table 2. Sample Handling Results Preliminary studies showed that blood counts from Vacutainers and carefully collected Microtainers were stable from minutes to six to eight hours. Conventional differentials performed on films prepared from two Microtainers collected one after the other from the same infant showed considerable differences, particularly in the proportion of cells classified as lymphocytes. Discrepancies between the pairedfilmswere greater with some reference technologists than with others. The consensus method used to establish "truth" minimized such differences but probably did not eliminate them. Correlation and Regression Analyses The two instruments showed good agreement (Table ) in measurement of hemoglobin concentration and WBC, RBC, and platelet counts. Granulocytes (or neu- * Within the context of this study, "backlighting" of a result on the S + report indicates failure of WBC distribution criteria in addition to those that evoke R-flags, or a mononuclear cell number in excess of 1,5/mm. "Starred" results on the H-l indicate disagreement in the WBC count between the peroxidase and basophil channels or failure of other instrument "self-check" analyses. Table 1. Criteria for the Evaluation of the Reference Differential The manual differential was classified as abnormal if one of the following criteria were met (all absolute numbers are cells/mm ): Any blasts are seen on the blood smear. Promyelocyte absolute number a 1 cells. Unclassified young cell absolute number a 1 cells. Myelocyte absolute number a: 1 cells. Plasma cell absolute number a 1 cells. Any combination of the above cells a 1 cells (i.e., 5 myelocytes plus 5 promyelocytes). Metamyelocytes a % or absolute number a 1,5 cells. Atypical lymphs a 5% or absolute number a 5 cells. Bands S; 1% or absolute number S; 1, cells. NRBCs ^ % or absolute number a 2 cells. Eos ^ 8% or absolute number a 7 cells. Basos i % or absolute number s 2 cells. 565 trophils) and lymphocytes enumerated by either counter showed excellent correlation with reference differential counts. H-l eosinophil counts showed a fairly good correlation (r =.811) with manual methods, but H-l and S + mononuclear or monocyte counts and H-l basophil counts did less well. These differences were undoubtedly exaggerated by the inaccuracies of our reference differentials, 17 but the calculations may overemphasize differences that are clinically not significant. Routine Differential Count Error Rate (Technologist Errors) When eight experienced reference technologists reviewed 776 routine emergency room differentials and 176 routine neonatal differentials, we found considerable differences in error rates. Samples from the ER had a 9.% (72 of 776) error rate, whereas those from the NCU had a 15.% (27 of 176) error rate. Both error rates are higher than our earlier laboratory error rate of 7%, 12 although none of the errors were of sufficient magnitude to have altered clinical management. The most common errors with samples from both sites were in classification of band and immature neutrophils and monocytes. Sensitivity and Specificity Studies Overall, 22 blood films from adults (29%) and 15 from neonates (77%) were considered abnormal, with 1.7 and 2. abnormalities per film, respectively. Among the ER samples, 276 were flagged by the S + and 192 by the H-1. Sensitivity and specificity data from samples from both sites (Table ) showed only slight differences between the two instruments, when their ability to detect abnormal samples (regardless of type of abnormality) was compared. For ER patients, the S + had a slightly better sensitivity rating than the H-l, whereas the H-l had a better predictive value for abnormal specimens and a slightly better efficiency. Data from newborns showed little or no dif-

4 566 NELSON ET AL. AJ.C.P. May 1989 Table 2. Criteria for the Evaluation of s for the S-Plus V and H-l* Blood Smear Criteria R-1 N False positive False negative Theflagis present and NRBCs are noted on the blood smear. Theflagis present and no NRBCs are noted on the blood smear. There is noflagand there are a% NRBCs or a cells. R-2 Blast Atyp G There is aflagand at least one blast is seen on the blood smear. There is aflagand atypical lymphs are i5% or &5 cells. There is aflagpresent and young cells, myelocytes, and promyelocytes are a 1 cells or metamyelocytes &% or 2 cells. R- L-shift There are a 15% bands or & 1,5 immature granulocytes. R-M Theflagis present and at least one or more of the R-1 through R- criteria are met. Theflagis present and any abnormal diff criteria are met. ' The criteria for false positive and false negative flags are similar to those for R-! or N flags. ference in overall performance between the two instruments. Studies by Payne and Pierre 1 suggested that instruments could vastly improve their ability to detect samples with abnormal differential counts by combining instrumentflagswith "check and review" limits. When all but two (scaled histograms and y-axis take off>2 mm) of the check and review limits of Payne and Pierre 1 were applied to the unflagged abnormal ER samples, we found the S + missed 9 samples or 5%. Missed abnormalities included increased band counts (2), eosinophils (8), atypical lymphocytes (), basophils (), and monocytes (1). The H-l missed (.%) of the abnormal samples. These abnormalities included increased band counts (27), atypical lymphocytes (), nucleated RBCs (2), and monocytes (2). Based on thesefigures,if we had used only instrument flags and check and review limits, the would have safely reduced manual differentials by 6% and the H-l would have reduced them by 75%. Reliability of White Blood Cell s The accuracy of individual WBC flags was evaluated with the use of manufacturer's criteria (Table 1). For example, an increased number of band neutrophils should generate an "R-" flag on the S + (or an RMflagif multiple abnormalities are present) and a "left shift" flag on the H-1; more immature neutrophils should generate "R-2" and "immature granulocyte"flags;and nucleated RBCs should generate "R-1" (S + ) or "N" (H-l) flags. n the adult ER population, the S + generated more true positive and fewer false positive flags than did the Table. Correlation Between Automated and Reference Differentials (specimens from adult emergency room) Measurement H-l versus S-Plus V WBC RBC HGB Pit H-l versus reference differential Neutrophils Lymphocytes Eosinophils Monocytes Basophils S-Plus V versus reference differential Granulocytes Lymphocytes Mononuclear cells r Table. A Comparison of Electronic and Cytochemical Screening Differentials for Adult and Pediatric Patients Test Method Sensitivity % true positive Specificity % true negatives Predictive value of normal Predictive value of abnormal Efficiency % of samples correctly classified ER NCU Note: These data reflect only the ability of flags to detect abnormal samples (i.e., those that require conventional differentials). They do not combine the use of "check and review limits" with instrument flags, as recommended by Payne and Pierre."

5 Vol.91 'No. 5 EVALUATON OF DFFERENTAL WHTE BLOOD CELL COUNT 567 Table 5. Reliability and Frequency of ndividual s as Detectors of Specific Abnormalities in the WBC Differentials of Adult ER Patients Emergency Room Population (N = 776) Hag R-1 11 (7%) (27%) 15 N 8 (1%) 9 (86%) R-2 15 (1%) 22 (59%) 7 12 G* ATL Blast 7 (78%) 21 (57%) (2%) 2 (22%) 16 (%) 1 (76%) R-* Bands ;> 15% 2 (7%) (6%) 5 57 Lt Shift* Bands 15% 25 (9%) 26 (51%) R-M* 5 (6%) 62 (6%) 97 8 (5%) (5%) 6 NA s Discrepancies between numbers of flags result from cross-flagging (i.e.. increased band counts may be flagged by "R" or "RM" or by "Left Shift" or "G"). = true positive: «false positive: <= false negative. H-1 (Table 5), whereas in the neonatal population there were more false positive and negativeflagswith the H-1. Overall, the reliability of the two instruments was similar with samples from the adult ER, with only 2% (Table 5) of the S + and 8% of the H-1 WBCflagscorresponding to abnormalities seen on the blood smear. n the NCU population, the S + WBCflagscorresponded to blood smear abnormalities 71% (Table 6) of the time, whereas Table 6. Reliability and Frequency of ndividual WBC s as Detectors of Specific Abnormalities in the WBC Differential NCU (N = 176) R-1 2 (77%) 6 (2%) 26 5 N 19 (86%) (1%) R-2 6 (6%) 7 (7%) 1 G* ATL Blast 19 (86%) 18 (56%) (%) (1%) 1 (%) 15 (1%) R-* Bands 15% or 1,5 11 (85%) 2 (15%) 1 8 Lt Shift* Bands 2: 15% or 1,5 (71%) 12 (29%) 2 1 R-M* 71 (76%) 2 (2%) (75%) (25%) 12 s ' Sec footnote to Tahlc 5. = true positive: = false positive; = false negative.

6 568 NELSON ET AL. A.J.C.P.-May 1989 the H-1 corresponded 2% of the time. f false positive blast flags were ignored, performance of the two instruments was very similar (71 vs. 7% "correct" flags). Discussion Electronic and cytochemical differential counters have made tremendous advances over the last ten years. By decreasing the number of conventional differentials, their use can reduce laboratory operating costs, decrease turnaround time, and yield a greater precision and accuracy than most manual differential counts. Used for that purpose, the questions of interest are the numbers of truly abnormal samples detected or missed and the number of false positive results. n the present study, "flags" generated by both instruments did not do particularly well in identifying abnormal samples (Tables -6). Atfirst,such data seem contrary to previously published reports, 912 but if a need for conventional differentials is based on both instrument-generatedflagsand check and review limits, 1 we found the S + missed only 5% and the H-1 only.% of the abnormal adult samples. Both instruments did better than our technologists, who had a 9.% error rate on adult samples. Such results clearly suggest that the combination of instrumentflagsand check and review limits, used by the operator or an online computer system, can yield results at least as accurate as conventional differentials when used for patients similar to those in our adult ER. A similar statement cannot be made for the performance of either instrument in the NCU. Both the S + and the H-1 generated some type of WBCflagfor almost all (97%) samples from the NCU. ncreasedflaggingwith NCU samples appeared to result largely from two factors. Many patients in the NCU were very sick, but, perhaps more important, both the S + and the H-1 have their WBCflaggingcriteria based on adult population norms. Neonates (especially premature infants) generally have larger and less mature appearing lymphocytes than adults. 1 The S + generated R-2flagsfor these larger cells (11 R-2 or R-M combinations/176 counts), and the H-1 generated 15 false positive blastflags (Table 6). The increased proportion of large unstained cells (H-1) that we observed (6.5% NCU mean, normal adult range, -%) may reflect a similar problem with age-related norms. For both the ER and the NCU patients, the reliability and frequency of individual WBCflagsprovides some additional insight into instrument capabilities (Tables 5 and 6). n both locations, physicians are particularly interested in accurate assessment of the proportion of immature neutrophils. Among ER samples, the H-1 was slightly better atflaggingleft shifts than the S + ("left shift" vs. "R"flags,9 vs. 7% true positive results). Among NCU samples, the H-1 performed well at flagging younger granulocytes (86% true positive results), whereas the S + was slightly better atflagginghigh band counts (85% true positive results). Overall, the H-1 missed about as many abnormal samples as did the S + ( vs. 27 false negative results of 176 NCU samples). Regardless of patient population, recognition of the presence of nucleated RBCs is important not only because large numbers of them can affect the WBC count, but because smaller numbers can have ominous significance in certain patient populations. Among ER samples, the S + outperformed the H-1 in this regard (7% true positive vs. 1% true positive results), but the two instruments were quite similar when analyzing samples from the NCU (77% vs. 86% true positive results). From a laboratory standpoint, the S + and H-1 have various advantages and disadvantages, depending on the patient population and laboratory needs, and both are useful in laboratory practice. WBCflagsserve as a useful adjunct, guiding the technologist to possible leukocyte abnormalities and the need for a conventional differential. f the initial instrument count were released to the physician as an interim count, before a conventional differential were performed, the H-1 report would be more advantageous. Knowledge of the eosinophil count from the H-1 could be quite valuable in some circumstances, and the H-1 could guide the physician to other specific WBC abnormalities (left shift, atypical lymphocytes, nucleated RBCs, blasts). The S + flags,on the other hand, are so general that the R-2, R-M, and...flagswould alert the physician to anything from increased eosinophils to blasts. From the point of view of a physician in an acute care situation, the important question may not be whether or not a patient's blood film is abnormal, but whether or not a "left shift" is present. n such a circumstance, if one interpreted flags literally (contrary to manufacturers' instructions), the H-1 would be a more useful instrument for rapid test results. Our study reflects instrument operations in actual clinical practice. None of our samples received special treatment, and none of our analyses were repeated. We believe such a test environment is a more useful evaluation of an instrument's capabilities than its use in either a basically normal population or an extremely abnormal population (leukemia patients). 7 Ourfindingsindicate that by combining WBCflagswith check and review limits, a differential generated by either instrument in an adult ER can be a safe and effective screening technique. t is unlikely that conventional differential counts can be completely replaced by electronic or cytochemical blood examinations, but the data presented suggested that they could be safely reduced by 6-75% in the adult ER. Our current turnaround time from the ER (including transit time) is about 18 minutes for conventional differentials, depending upon the time of day. f automated differentials

7 Vol.91 -No. 5 EVALUATON OF DFFERENTAL WHTE BLOOD CELL COUNT 569 were used, the reporting interval could be shortened to about 8 minutes for most samples. For those requiring manual differentials, accuracy could be increased if technicians spent more time at the microscopes with each slide, probably with little or no increase in overall turnaround time. Acknowledgments. The authors thank Dr. E. Simson and Dr. D. Saw for providing them with the opportunity to evaluate the H-l and for help with correlation and regression analysis. They also thank the staff of the Hematology Laboratory for their valuable assistance and the staff of the Adult Emergency Room and Neonatal ntensive Care Unit for their cooperation. References 1. Agress H, Downey H: The blood picture of human new-borns, with special reference to lymphocytes. Folia Haematol (Leipz) 196;55: Breakell ES, Marchand A, Simson E: Comparison of performance of leukocyte differential counting of the H6 system with a manual reference method using the NCCLS standard. Blood Cells 1985;11: Charache S, Nelson L, Keyser E, Metzger P: A clinical trial of three part electronic differential white blood cell counts. Arch ntern Med 1985;15: Cox CJ, Habermann TM, Payne B, KJee G, Pierre R: Evaluation of the Counter model S-Plus V. Am J Clin Pathol 1985;8: Cranendonk E, Van Gennip AH, Abeling M, Behrendt H: Reference values for automated cytochemical differential count of leukocytes in children -16 years old: comparison with manually obtained counts from Wright-stained smears. J Clin Chem Clin Biochem 1985;2: Decresce R: The H-1: a discrete fully automated complete blood count and differential analyzer. Laboratory Medicine 1986;17: Drewinko B: Utility of the H-l system in malignant disease. Proceedings of the H-1 Hematology Symposium 1985;1: Johnson OL: Leukocyte screening using the ELT-8/WS. Blood Cells 1985;11: Jones R, Hellman R, Twedt D: The leukocyte differential. Blood Cells 1985;11: National Committee for Clinical Laboratory Standards (NCCLS): The white cell differential. : A symposium. Blood Cells 1985;11: National Committee for Clinical Laboratory Standards (NCCLS): The white blood cell differential. : A symposium. Blood Cells 1985;11: Nelson L, Charache S, Keyser E, Metzger P: Laboratory evaluation of the three part differential. Am J Clin Pathol 1985,8: Operator's manual, Counter S Plus V. Hialeah, Florida: Electronics, Payne B, Pierre R: Using the three part differential: Part. nvestigating the possibilities. Laboratory Medicine 1986;17: Pierre R, Payne B, Lee WK, et al: Comparison of four differential methods with the National Committee for Clinical Laboratory Standards (NCCLS) reference method. Am J Clin Pathol 1987;2: Ross DW, Bentley S: Evaluation of an automated hematology system ( H-l). Arch Pathol Lab Med 1986;11: Rumke CL: The statistically expected variability in differential leukocyte counting. n: Koepke JA, ed. Differential leukocyte counting. Skokie, llinois: College of American Pathologists 1978: Salzman GC, Cromwell JM, Martin JC, et al: Cell classification by laser light scattering: identification and separation of unstained leukocytes. Acta Cytol 1975;19: Tentative standard for leukocyte differential counting H2-T. Villanova, Pennsylvania: National Committee for Clinical Laboratory Standards.