BACKGROUND. Absolute neutrophil counts (ANCs) and absolute phagocyte counts

Transcription

1 2681 Utility of Automated Counting to Determine Absolute Neutrophil Counts and Absolute Phagocyte Counts for Pediatric Cancer Treatment Protocols Nobuko Hijiya, M.D. 1,2 Mihaela Onciu, M.D. 3 Scott C. Howard, M.D. 1,2 Zhe Zhang, M.S. 5 Cheng Cheng, Ph.D. 4,5 John T. Sandlund, M.D. 1,2 Emily P. Kyzer, M.S.N., R.N., C.P.N.P. 1 Fred G. Behm, M.D. 3,4 Ching-Hon Pui, M.D Department of Hematology-Oncology, St. Jude Children s Research Hospital, Memphis, Tennessee. 2 Department of Pediatrics, College of Medicine, University of Tennessee Health Science Center, Memphis, Tennessee. 3 Department of Pathology, St. Jude Children s Research Hospital, Memphis, Tennessee. 4 Department of Pathology, College of Medicine, University of Tennessee Health Science Center, Memphis, Tennessee. 5 Department of Biostatistics, St. Jude Children s Research Hospital, Memphis, Tennessee. Supported in part by a Cancer Center Support Grant (CA21765) from the National Cancer Institute and by the American Lebanese Syrian Associated Charities. The authors thank Julia Cay Jones, Ph.D., for her assistance with the scientific editing of the current article; Wei Liu, Ph.D., M.S., for her assistance with data analysis; and Imella Smith Herrington and Jeana Cromer for their assistance with article preparation. Dr. Pui is the American Cancer Society F.M. Kirby Clinical Research Professor. Address for reprints: Nobuko Hijiya, M.D., Department of Hematology-Oncology, St. Jude Children s Research Hospital, 332 North Lauderdale, Memphis, TN ; Fax: (901) ; E- mail: nobuko.hijiya@stjude.org Received June 25, 2004; revision received August 16, 2004; accepted August 18, BACKGROUND. Absolute neutrophil counts (ANCs) and absolute phagocyte counts (APCs) are used to guide cancer treatment. Although automated counting could replace manual counting, data showing correlations are lacking. By analyzing blood samples from children undergoing cancer treatment, the authors determined whether ANCs and APCs obtained by automated methods correlated positively with ANCs and APCs obtained manually. METHODS. The authors analyzed 3640 consecutive peripheral blood samples. Leukocyte counts determined by Beckman-Coulter Gen-S or HmX analyzers (Beckman-Coulter, Miami, FL) were used to calculate counts obtained by automated or manual methods. Automated differential counts were obtained by automated analyzers and manual differential counts were performed by medical technologists. Counts underwent linear regression analysis. The authors evaluated 5 cutoff values for ANCs and APCs commonly used in decision-making related to cancer treatment: 300/ L, 500/ L, 750/ L, 1000/ L, and 1500/ L. Manually determined ANCs and APCs served as standards to determine the sensitivity, specificity, positive and negative predictive values, and kappa coefficient for automated counting. RESULTS. R 2 values were 0.81 for ANCs determined by manual and automated methods and 0.84 for APCs determined by both methods. The specificity of the automated method was 90% for all ranges of ANCs and APCs, except one (APCs 300/ L). There was excellent agreement ( 0.9) between ANCs determined by manual and automated methods and APCs calculated by both methods. CONCLUSIONS. Automated methods of determining ANCs and APCs for children undergoing cancer treatment were reliable and can replace manual counting. Blood smear examination to validate ANCs and APCs determined by automated methods was needed only in selected cases. Cancer 2004;101: American Cancer Society. KEYWORDS: absolute neutrophil count, absolute phagocyte count, automated counts, manual counts, neutrophils, monocytes. The absolute neutrophil count (ANC) and absolute phagocyte count (APC) are often used to guide cancer treatments. 1,2 These values are obtained by multiplying the leukocyte count by the percentage of segmented neutrophils and bands (for the ANC) or by the percentage of segmented neutrophils, bands, and monocytes (for the APC). The leukocyte count is obtained by an automated counter. The percentages of neutrophils and monocytes can be obtained by either manual methods (the so-called manual ANC or manual APC) or by an automated counter (the so-called automated ANC or automated APC). Manual counting has been the gold standard in the accurate identification of cells in peripheral blood samples and in the generation of related numerical parameters used in patient care. However, 2004 American Cancer Society DOI /cncr Published online 29 October 2004 in Wiley InterScience (

2 2682 CANCER December 1, 2004 / Volume 101 / Number 11 manual counting is labor intensive and time-consuming and involves the examination of only a limited number of cells. Thus, the results of manual counting have varying degrees of reproducibility. 3 In addition, the low leukocyte counts that often occur after chemotherapy make it difficult to obtain a sufficient number of cells to render a meaningful manual differential count and to classify cells in a timely manner. These aspects can interfere with the timing of decisions required in patient care. The use of automated analyzers that employ various types of technology for identification and enumeration of nucleated cells in peripheral blood samples has alleviated the need for manual counting in the general practice setting. By their nature, these types of technology allow evaluation of a much larger volume of cells than is possible during the examination of a smear. Therefore, automated analyzers yield results that are highly accurate and highly reproducible. Analysis of a large quantity of cells is particularly important in patients with very low leukocyte counts. However, the reliability of this technology in the oncology setting has not been examined sufficiently. The most important factor that might interfere with the accuracy of automated differential counts in this context is the high frequency of abnormal and immature cells often encountered in patients with hematologic malignancies or with recovering bone marrow after chemotherapy or bone marrow transplantation. In addition, factors that interfere with the leukocyte count (e.g., the presence of numerous nucleated red blood cells [NRBCs], platelet aggregates and, rarely, cryoglobulins) may interfere with correct ANCs and APCs. 4 Previous studies 5,6 have proposed the use of ANCs obtained by automated means as a reliable parameter to replace manually obtained values, allowing for more timely availability of this information. However, data from these studies were limited to a small series of samples, and the applicability of their findings in a pediatric oncology setting was assessed insufficiently. In addition, there is no previous information regarding the use of automated analyzers to determine automated APCs. To determine whether ANCs and APCs obtained by automated methods can be used reliably for clinical decision-making in the setting of pediatric oncology, we correlated the automated counts with manual counts from a large number of peripheral blood samples obtained from children with cancer. MATERIALS AND METHODS The current study was approved by the institutional review board of St. Jude Children s Research Hospital in Memphis, Tennessee. During a 2.5-month period, 3643 consecutive samples of peripheral blood were collected from patients with cancer. All blood samples were processed according to standard laboratory procedures. These procedures included performing a complete blood count (CBC) with one of two hematology analyzers the Beckman- Coulter Gen-S or the Beckman-Coulter HmX (Beckman-Coulter, Miami, FL). For each sample, if a morphology flag was displayed by the analyzer, a smear was made on the integrated Beckman-Coulter Gen-S SM slide maker system and Wright stained on a HEMA-TEK 2000 slide stainer (Bayer Corporation, Tarrytown, NY). Smears were then examined by 1 of 26 experienced hematology technologists during 1 of 3 shifts. A 100-cell differential count was performed by the technologist, and all the cells available on the slide were recorded only if 100 cells were found. Smears were further examined by one of three hematopathologists if the technologist encountered cells that were difficult to classify or if the sample contained blasts that were not seen on the most recent differential count obtained from that patient. The data recorded for each sample included the main parameters of a CBC, the differential count obtained manually, and the differential count originally obtained by the analyzer. The automated ANC was calculated as the leukocyte count (percent neutrophils)/100. The manual ANC was calculated as the leukocyte count (segmented neutrophils bands)/(cells counted). The automated APC was calculated as the leukocyte count (percent neutrophils percent monocytes)/100 and the manual APC was calculated as the leukocyte count (segmented neutrophils bands monocytes)/(cells counted). The data were analyzed by standard linear regression first to evaluate the correlation between the automated ANC and the manual ANC and between the automated APC and the manual ANC. Next, we evaluated 5 cutoff values for ANCs and APCs commonly used to guide decision-making for children with cancer: 300/ L, 500/ L, 750/ L, 1000/ L, and 1500/ L. Agreement between automated and manual ANC and APC on above/below these cutoff points was then assessed by estimating the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), 7 and the kappa coefficient 8 using the manual ANCs and APCs as standards. Adjusting for agreement that might occur by chance, the kappa statistic measures the extent of exact agreement between automated and manual ANCs and between automated and manual APCs. The range of values for the kappa statistic is 0 (no agreement beyond that due to chance) to 1 (perfect agreement) indicates excellent

3 Correlation of Automated and Manual Counts/Hijiya et al FIGURE 1. Simple regression analysis of automated and manual (A) absolute neutrophil counts (R ) and (B) absolute phagocyte counts (R ) in the 3529 specimens with leukocyte counts 20,000/ L. FIGURE 2. Simple regression analysis of automated and manual (A) absolute neutrophil counts (R ) and (B) absolute phagocyte counts (R ) in the 720 specimens with leukocyte counts 1500/ L. agreement, indicates intermediate agreement, and 0.4 indicates poor agreement. 8 RESULTS Of the 3643 samples collected, 3 samples were excluded from our analysis because data pertaining to the percentage of monocytes were unavailable. Therefore, 3640 samples were available for analysis. Figures 1 and 2 illustrate the results of simple regression analyses of automated and manual ANCs and APCs when the specimens with a leukocyte count 20,000/ L (Figure 1) or 1500/ L (Figure 2) were included. The R 2 values were 0.93 for the ANCs and 0.95 for the APCs when the 3529 specimens with a leukocyte count 20,000/ L were examined. The R 2 values were 0.92 for the ANCs and 0.93 for the APCs when the 720 specimens with a leukocyte count 1500/ L were evaluated. When all samples (N 3640) were included, the R 2 values were 0.81 and 0.84 for the ANCs and APCs, respectively. Table 1 summarizes the results of validation and reliability tests. The sensitivity of automated counting of all examined ranges of ANCs and APCs was 96% and, except for the specificity of automated counting when the APCs were 300/ L, the specificity was 90%. Specificity tended to be lower when the ANCs

4 2684 CANCER December 1, 2004 / Volume 101 / Number 11 TABLE 1 Validity and Reliability of Automated Counting to Determine ANCs and APCs Cutoff value (cells per microliter) No. of samples a Validity b Reliability A B C D E F G H ANC APC Note. ANC: absolute neutrophil count; APC: absolute phagocyte count. a A: number of samples whose manual and automated ANCs and APCs are greater than the cutoff values; B: number of samples whose manual ANCs and APCs are greater than and whose automated ANCs and APCs are less than the cutoff values; C: number of samples whose manual ANCs and APCs are less than and whose automated ANCs and APCs are greater than the cutoff values; D: number of samples whose manual and automated ANCs and APCs are less than the cutoff values. b E: sensitivity of automated counting to measure ANCs and APCs (sensitivity is A/[A B]); F: specificity of automated counting to measure ANCs and APCs (specificity is D/[C D]); G: positive predictive value of automated counting in measuring ANCs and APCs (positive predictive value is A/[A C]); H: negative predictive value of automated counting in measuring ANCs and APCs (positive predictive value is D/[B D]). TABLE 2 False-Positive Results and Their Causes Cutoff value (cells per microliter) No. of samples (percentage 3640) with falsepositive results a Average difference between manual counts and cutoff values (%) No. of samples whose manual counts were > 30% lower than cutoff values (percentage 3640 specimens) b Blasts b Reasons for a discrepancy >30% Immature myeloid cells b Monocytes b ANC (1.2) (0.5) (0.7) (0.3) (1.2) (0.3) (1.3) (0.2) (1.6) (0.4) APC (1.4) (0.6) (1.4) (0.4) (1.6) (0.3) (1.8) (0.3) (2.0) (0.3) ANC: absolute neutrophil count; APC: absolute phagocyte count. a See column C of Table 1. b Some specimens had more than one cell type that caused a discrepancy Therefore, the total number of specimens with blasts, immature myeloid cells, or monocytes exceed cited as the cause of discrepancy may the total number of samples whose manual counts were 30% lower than the cutoff values. and APCs were lower. For all categories, the PPV of automated counting was 97% and the NPV was 94%. The kappa coefficient values for all analyzed ranges of ANCs and APCs were Table 2 summarizes the number of samples for which we obtained false-positive values (i.e., the automated count, but not the manual count, was higher than the cutoff value). Depending on the range of ANCs and APCs, approximately % of the 3640 specimens yielded false-positive values. Among those samples with false-positive values, the average difference between manual counts and the cutoff values was % of the appropriate cutoff value. The average difference was 30% for only % of specimens. Most of these specimens had a significant number of blasts, immature myeloid cells, immature

5 Correlation of Automated and Manual Counts/Hijiya et al monocytes, or very few cells available for manual counting. DISCUSSION In the current study, we examined the feasibility of applying the automated ANCs and APCs to therapeutic decision-making in a pediatric oncology setting. We used 2 Beckman-Coulter analyzers employing similar technologies to analyze 3640 consecutive blood samples. The results indicate excellent validity and overall reliability for the automated ANCs and APCs. Therefore, these automated counts are useful in clinical decision-making in cancer treatment and can replace manual counts in the majority of the cases. Regression analysis showed an overall positive correlation between automated and manual ANCs and between automated and manual APCs when all 3640 specimens were included. The correlation improved when only specimens with leukocyte counts 20,000/ L were included. This improvement is probably because counts above the normal range ( 20,000/ L) in our patient population are usually due to leukemic blasts, immature myeloid cells, or immature monocytes, each of which may not be identified accurately and which may be discriminated against by automated analyzers. 9,10 When the leukocyte count was 1500/ L, the cutoff value at which critical decision-making is usually done, there was excellent correlation. The results designated as false-positive results (i.e., the automated counts exceed the cutoff value but the manual counts do not) are associated with a risk of neutropenic patients receiving additional chemotherapy that results in more profound neutropenia. These false-positive results were limited to only % of 3640 specimens. For these specimens, the average difference between manual counts and cutoff values was only approximately 20%. Therefore, it is unlikely that a false-positive result would adversely affect a patient s treatment. Only % of samples had manual counts that differed from the cutoff values by 30%. Most of these specimens contained blasts, immature myeloid cells (myelocytes, metamyelocytes, and promyelocytes), or monocytes, all of which are not recognized accurately by automated counters. 9,10 The analyzer typically indicates the presence of these morphologically unusual cells by flags, and the manufacturers recommend manual verification when a morphology flag or a leukocyte flag is displayed by the instrument. The machines use flags to indicate various types of cell abnormalities, and flags were displayed during analysis of all samples in our study. Morphology flags are very common for specimens obtained after the completion of chemotherapy. Our results indicate that manual verification is not necessary for all samples, but it should be performed when flags indicate the presence of blasts, immature myeloid cells, or monocytes, each of which may interfere with neutrophil counting. It has also been reported that the presence of NRBCs, giant platelets, or platelet clumps interferes with the accuracy of overall leukocyte counts. 4 Validation should also be conducted when the analyzer flags suggest such findings. Although the overall sensitivity and specificity of automated counting were excellent, the specificity of the automated counting of APCs was lower than that of ANCs, especially when specimens contained few cells ( 500/ L). The lower specificity seems to be attributed to the relatively high false-positive rate associated with automated APCs (column C in Table 1). This scenario probably occurs most often when the bone marrow is recovering after chemotherapy and there are significant numbers of immature monocytes, which automated counters have difficulty in identifying. This discrepancy is in contrast with the findings involving granulocytes, whose high content of cytoplasmic granules makes consistent and accurate identification by automated analyzers more likely. Despite the limited ability of automated counters to recognize certain cell types, these devices have the advantage of analyzing hundreds to thousands of cells from each blood sample. Therefore, the results that they generate should have less statistical variation than manual measures, in which only 100 cells (or fewer in samples with a low leukocyte count) on a single blood smear are counted. Although we used manual counts as the standard in the current study, automated ANCs and APCs may be more accurate when the leukocyte count of a sample is low, as long as cells with unusual morphology are verified by inspection of blood smears. In addition, the use of automated counting avoids the duplicate errors 11 and intraobserver variation 12 that may occur with manual counting. In conclusion, automated counting is a reliable method of obtaining ANCs and APCs for use in clinical decision-making for pediatric patients undergoing cancer treatments, and automated counting can replace the more time-consuming manual counting. However, because of the limited ability of the automated analyzers to accurately classify abnormal cells that are often present in this setting, inspection of blood smears should remain an integral part of peripheral blood evaluation. Validation by a manual method should be done when the analyzer flags indicate the presence of blasts, immature myeloid cells, or monocytes (each of which may interfere with accurate neutrophil counting), or the presence of NRBCs or

6 2686 CANCER December 1, 2004 / Volume 101 / Number 11 platelet clumps (each of which may interfere with accurate leukocyte counting). REFERENCES 1. Aquino VM, Buchanan GR, Tkaczewski I, Mustafa MM. Safety of early hospital discharge of selected febrile children and adolescents with cancer with prolonged neutropenia. Med Pediatr Oncol. 1997;28: Rackoff WR, Gonin R, Robinson C, Kreissman SG, Breitfeld PB. Predicting the risk of bacteremia in children with fever and neutropenia. J Clin Oncol. 1996;14: Rümke CL. The statistically expected variability in differential leukocyte counting. In: Koepke JA, editor. Differential leukocyte counting. Skokie, IL: College of American Pathologists, 1979: Steele BW, Wu NC, Whitcomb C. White blood cell and platelet counting performance by hematology analyzers: a critical evaluation. Lab Hematol. 2001;7: Parham DM, Ready R, Stine K, Quiggins C, Becton D, North P. Comparison of manual and automated leukocyte counts for determination of the absolute neutrophil count: application to a pediatric oncology clinic. Med Pediatr Oncol. 2002;38: Warner BA, Reardon DM. A field evaluation of the Coulter STKS. Am J Clin Pathol. 1991;95: Campbell MJ, Machin D. Medical statistics: a commonsense approach (3rd edition). New York: John Wiley & Sons, Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33: Kapelushnik J, Kafka M, Moser A, Yermiahu T. Automated differential counts in acute promyelocytic leukemia patients may be misleading. Med Pediatr Oncol. 2002;38: Hübl W, Hauptlorenz S, Tlustos L, Jilch R, Fischer M, Bayer PM. ion and accuracy of monocyte counting. Comparison of two hematology analyzers, the manual differential and flow cytometry. Am J Clin Pathol. 1995;103: Rümke CL. The imprecision of the ratio of two percentages observed in differential white blood cell counts: a warning. Blood Cells. 1985;11: Koepke JA, Dotson MA, Shifman MA. A critical evaluation of the manual/visual differential leukocyte counting method. Blood Cells. 1985;11: