Accuracy and Utility of Differential White Blood Cell Count in the Neonatal Intensive Care Unit

Transcription

1 HEMATOPATHOLOGY Accuracy and Utility of Differential White Blood Cell Count in the Neonatal Intensive Care Unit SAMUEL CHARACHE, M.D., 1 LYDIA NELSON, M.S., M.T.(ASCP)SH,' DAISY SAW, M.D., 2 EDWARD KEYSER, M.T.(ASCP), 1 AND SANDRA WINGFIELD, M.T.(ASCP) 1 The clinical utility of the complete blood cell count (including the differential white blood cell count) as a means to follow the course of infants in a neonatal intensive care unit was assessed. Utility was judged for three purposes: (1) predicting the onset of clinically unrecognized disease, (2) assessing the severity of current disease, and (3) following a trend during treatment. Neither conventional nor automated differential counts were useful for surveillance (predicting the onset of clinically unrecognized disease). The white blood cell count, the platelet count, and the absolute immature neutrophil count and immature/total neutrophil ratio were useful to assess the severity of current clinical events. The white blood cell count was superior to the differential count for following trends in patients' conditions. Information regarding nuclear immaturity derived from automated counts and the cost and slowness of the manual differential count are good indications for decreased use of conventional counts, increased use of certain features of the automated differential, or both, in neonatal intensive care units. (Key Words: Complete blood counts; Differential white counts; Automation; Clinical laboratory correlation; Decision making) Am J Clin Pathol 1992;97: The differential white blood cell count, and particularly the proportion of immature neutrophils, has been used as an index of inflammation or infection. 1 ' 2 In acute situations, neonatologists use neutrophilia or neutropenia, 3 " 10 an increased proportion of immature neutrophils, 1 ' 3 ' 5 '" -13 ential count can be produced by a variety of noninfectious complications in the newborn period. 14 Performance of conventional differential counts is subject to a variety of errors, 215 " 18 particularly in newborns, whose cells are hard to classify. 19 Accepting such limitations, the presence of toxic granulation, 13 and thrombocytopenia and recognizing imprecise correlations between 1 ' 13 to assess the severity of a recognized clinical illness. These indices also are used to assess both the response to therapy as an illness evolves and to detect serious illness (particularly sepsis) before it becomes clinically evident. Differential counts are used for those purposes despite recognition that abnormalities in the differlaboratory measurements and the conditions of their patients, clinicians use blood cell counts to support or refute, rather than prove, clinical impressions. Trends over time are considered more useful than isolated measurements to gauge a patient's improvement or deterioration, and several counts per day can be requested for extended periods of time. Unfortunately, the differential count is labor intensive From the 'Departments of Laboratory Medicine and Medicine, Johns and hospital laboratories are under tremendous pressure Hopkins Medical Institutions, Baltimore, Maryland, 2 Departmenl to of reduce Pathology, New York University, New York, and Department of Clinical differentials,' performed automatically by combinations labor costs. Recent development of "screening Chemistry, Memorial-Sloan Kettering Cancer Center, New York, New York. Received April 24, 1991; received revised manuscript and accepted for publication June 24, Presented in part at the October 1989 meeting of the American Society of Clinical Pathologists. Daisy Saw was a pathologist for Technicon Instrument Corporation until November Address reprint requests to Dr. Charache: The Johns Hopkins Hospital, 600 North Wolfe Street, Meyer B 12ID, Baltimore, Maryland of electrical impedance, 20 light scattering, 21 and histochemical stains, 22 offers a possible solution. Automated differential counts provide precise measurement of the major classes of leukocytes. They do not directly assess nuclear immaturity and cannot recognize such features as toxic granulation, but they do provide flags that alert the technologist that such abnormalities may be present. The flags are used by the laboratory only to signal the 338

2 CHARACHE BT AL. 339 Accuracy and Utility ofdifj r erential WBC in the NICU need for a conventional differential. Reliability and clinical utility of automated differential counts have been evaluated in adults, 20 ' 21 ' 23 ' 24 and to a lesser degree in older children. 25 If they could be shown to yield results that perform as well as conventional differential counts in neonates, the small sample size, short analysis time, and rapid reporting characteristics of contemporary instruments might solve a difficult problem in laboratory practice. The present study compared the accuracy and utility of conventional and automated differential counts in an unselected group of blood samples from the neonatal intensive care unit (NICU). Neither type of differential was accurate in this group of patients and neither was a reliable guide to clinical status. The conventional differential was superior to the automated count, but the difference between them was not large, suggesting certain changes in clinical practice. Blood Samples MATERIALS AND METHODS One hundred seventy-six samples were collected in Na 2 EDTA Microtainers (Becton-Dickinson Corp., Rutherford, NJ), during a 2.5-month period, from 63 babies. Only samples that were collected in duplicate (i.e., two Microtainers) were used in the study. Gestational ages varied from 23 to 41 weeks (mean, 28.7 weeks) and birth weights from 510 to 3930 g (mean, 1512 g). The specimens were collected by heel stick or from umbilical venous catheters and were collected only when requested by the house officer caring for the child. The time at room temperature between sample collection and sample analysis was generally 1-4 hours, but a few samples were slightly older. Sample Handling In addition to the routine conventional differential, all available samples also were run on the Coulter S Plus IV (S + 4) (Coulter, Hialeah, FL) and the Technicon H-l (H-l) (Technicon, Tarrytown, NY), both instruments being maintained according to manufacturers guidelines. Conventional differential counts were performed on Wright-stained blood smears using a Hematrak 360 automated microscope (Geometric Data, Wayne, PA; no longer manufactured) when possible: technologists scanned each smear before the instrument count and opted for an eyecount differential if more than a few cells could not be identified by the Hematrak or if its count was perceived as incorrect. All conventional differential counts were reviewed carefully by a group of eight experienced technologists to estimate the technologist's error rate; only corrected counts ("reference counts") were used in calculations. Details of the review process and a discussion of the accuracy of our "routine" count are published elsewhere. 15 Differential counts were considered to differ if discrepancies were greater than those expected from sampling errors. 18 The S + 4 and the H-1 provided routine 8-part blood counts and screening differential counts. The S + 4 differential was based on measurements of cell volume after partial lyses of cell membranes under precisely controlled conditions, with differentiation of white cells into three classes (lymphocytes, mononuclear cells, and granulocytes). The H-l uses light scattering and cytochemical stains to differentiate between white cells, lymphocytes, monocytes, neutrophils, eosinophils, and basophils. With the S + 4, an "R3" or "RM" flag was displayed if there was a "left shift" among the neutrophils. The H-l generated "left shift," "immature granulocyte," and "blast" flags. Cultures Retrospectively, central laboratory files were reviewed for results of cultures and antigen or antibody detection tests. Thirty-two cultures were performed on the same day as one of the study blood counts. Excluded from analysis were two negative-antigen detection tests and several tests for cytomegalovirus antibody, all of which yielded negative results or were due to immunoglobulin G, presumably of maternal origin. Degrees of Laboratory Abnormality (Scores) The director of the NICU was asked to provide a set of criteria for degree of abnormality of the white blood cell and differential count. These criteria of abnormality (0, 2+, or 4+) were based on days after delivery (day 1, 2, and 3 or after). The list was circulated among the attending staff of the NICU, modified slightly, and then used to categorize each blood count for each child (Table 1). For brevity, data derived in that fashion are called scored variables. In many respects, this system is similar to clinical and laboratory scoring schemes developed by other investigators Instrument Flags If all flags were considered, virtually every differential count calculated by either instrument should have been rejected in favor of a conventional count. Only the R3 and RM flags from the Coulter Counter and the left shift and immature granulocyte flags from the Technicon Counter were considered because they might reflect the presence of a "left shift" in neutrophils. Recognizing that the process had not been approved for clinical use, we Vol. 97 No. 3

3 340 HEMATOPATHOLOGY TABLE 1. DEGREE OF LABORATORY ABNORMALITY FOR INFANTS IN THE NICU Age (0-48 hrs) Age (49-72 hrs) Age after (72 hrs) WBC Absolute neutrophil Absolute juvenile or younger Immature neut Total neutrophil Absolute lymphocyte Platelet Lobularity index <6 <4 <6 >30 >33 >30 <3.5 <2.5 <2 >15 >20 >11 >1.4 >2.0 >1.0 >16 >30 >15 <.8 <.5 <.8 >10 <150 <100 >10 <150 All ages: 0, >2.4; ; 4+ < 1.6 <4 >33 <1.5 >16 >1.5 >27 <.5 <100 <6 >30 <1.5 >9 >1.0 >14 <1.5 <150 <4 >33 <1.0 >1.5 >33 <1.0 >15 <100 All units are in ]0 9 ceils/l except for the immature neutrophil/total neutrophil ratio and the lobularity index. analyzed flagged counts as if the flags were correct, eliminating only those reports in which the instrument gave no results at all and those H-1 counts that were unreliable because of incomplete lysis of red cells (a problem solved in many cases by prior dilution of the sample 15 ). The R3 flag from the S + 4 was scored 2+ and the RM flag was scored 3+. The H-l "left shift" flags (LS1-LS3) were scored 1-3+ and the immature granulocyte flag was scored 2+, with a maximum total score of 4+. Clinical Correlations After completion of laboratory studies, attempts were made to obtain hospital charts for each of the studied patients. Sixty-one charts were obtained and reviewed by one of us (S.C.). Charts were reviewed to ascertain the broad pattern of the child's condition during his or her entire NICU stay, with particular attention to events and blood counts 48 hours before, during, and 24 hours after the day of sample collection. The child's condition at the time of the count was scored from O (progressing normally) to 4+ (critically ill) on the basis of the attending neonatologist's notes. The trend of events during the preceding 48 hours was noted (worse, same, or better) and the emergence of previously unrecognized disease during the subsequent 24 hours was recorded. Among the 61 evaluable patients, 170 blood counts were analyzed. Twenty-one of the 61 patients had uncomplicated courses during the period of study and another 8 had complicating conditions that might not be expected to alter their blood counts (i.e., congenital heart disease, hydrocephalus, and so on). One hundred fifteen of the total of 170 counts were calculated in the remaining 32 infants. All data were analyzed together (the complete set), and those from the 32 sick infants were analyzed separately (the sick set) (Table 2). Values of r 2 (see below) were similar or equal for the two analyses, and only data for the complete set are listed. Regression Analyses DATA ANALYSIS The relationship between current clinical status and each of 10 laboratory data (listed in Table 3), singly and in combinations, was studied by a stepwise regression technique using the SAS statistical program (SAS Institute, Cary, NC). 1 Measurements that were significantly abnormal if high or low (white blood cell count, platelet and absolute neutrophil or granulocyte counts) were analyzed only as scored variables. Measurements that might be expected to vary monotonically with clinical status (i.e., absolute bands, the ratio bands/neutrophils, platelets) were analyzed in terms of both absolute value and score in separate analyses. In addition to scores based on the individual components of a differential count, overall abnormality for the type of count based on the presence of any abnormal finding also was analyzed using the highest degree of abnormality for any component. For example, one sample might have had increased bands but a normal granulocyte count, and another might have had normal bands but an increased granulocyte count. However, both were considered abnormal in the overall analyses. TABLE 2. CLINICAL DIAGNOSES OF BABIES IN THE "SICK SET" Diagnosis Respiratory insufficiency Necrotizing enterocolitis Possible sepsis Pneumatosis of bowel Diarrhea Chorioamnionitis Cerebial edemia Twin-turn transfusion Meningitis Number A.J.C.P. March 1992

4 CHARACHE ET AL. 341 Accuracy and Utility of Differential WBC in the NICU TABLE 3. GAMMA STATISTICS FOR SCORED MEASUREMENTS ASSOCIATION BETWEEN DEGREE OF LABORATORY AND CLINICAL ABNORMALITY Gamma Probability Value Automated counters* Conventional differential count Coulter differential count H-l differential count WBC Platelet Absolute neutrophil Absolute immature neutrophil Immature/total neutrophil Conventional differential any abnormality Absolute granulocyte R3-RM flag Coulter counter any abnormality Absolute neutrophil Immature neutrophil Lobularity index HI-any abnormality > See text for scoring of flags. * The white blood cell and platelet counts were virtually identical on the two automated counters. Gamma Statistic Regression techniques usually are used for comparison of continuous variables, and those studied here are classified or categorized (0, 1-4+). The gamma statistic was proposed by Goodman and Kruskal 2930 as a measure of association between sets of classified variables when both classifications have intrinsic order (i.e., 1-4+). Gamma varies between +1 and 1, much as the regression coefficient does. Calculation of gamma for pairs of variables (clinical status versus each of the scored laboratory measurements) were made using Lotus 2.01 software (Lotus Development Corp., Cambridge, MA) and a personal computer. Gamma statistics were used to calculate "Z" statistics, and these in turn were used to calculate the probability that associations were not due to chance. A BC " HK J K L M N LINICAL WBC PLATELET Trend Analysis Because physicians use blood counts to assess trends in a patient's conditions in response to therapy, each count was compared with preceding counts (less than 48 hours), and direction of change in the current count was expressed as worse (-), same (0), or better (+). Changes were considered to have occurred only if the datum moved in or out of categories listed in Table 1. Clinical change (-, 0, or +) was determined from the attending physician's notes in the patient's charts. Laboratory and clinical change were compared for concordance for all available measurements; S + 4 differential counts were available for comparison in about one half of the cases, and H-1 differential counts were available in about one quarter of them because preceding counts were not always calculated on both instruments. Data were arranged in a 3 X 3 table (-, 0, + versus -, 0, +) and analyzed by means of the gamma statistic. 8/23 27 SI 9/4»/l /1 -CONVENTIONAL DIFF COULTER DIFF FIG. 1. Example of a patient's course (see text). Vol. 97 No. 3

5 342 HEMATOPATHOLOGY Case Summary RESULTS An example of analysis of a particularly complex patient's course is depicted in Figure 1. Scores for the child's clinical course are plotted on the top line; letters refer to events during the period of observation. Serial scores for various components of the blood count are plotted below. Each line goes from a minimum of 0 to a maximum of 4+, but actual values are not plotted on the ordinate (degree of abnormality) because of overlap of scales. Platelet transfusions are indicated by arrows. The time line breaks between 11 and 27 days of life. The child was born (A) after premature rupture of membranes. She did well on ampicillin (B,C,D), and feedings were advanced (E). On the 27th day of life (F), she developed bloody diarrhea. When her condition stabilized (G), a hemicolectomy was performed. After operation she required pressors; at (H) the dopamine dosage was reduced. At (I), Candida were reported in samples of urine and pleural fluid and amphotericin therapy was started. The remainder of her colon was removed (J) and she maintained slight improvement (K). Her condition remained stable, despite bleeding from the colostomy (L). She improved somewhat (M), and by the 47th day (N) was described by the attending neonatologist as "still critically ill but stable." In this patient scored clinical abnormalities were most closely paralleled by white blood count scores (Fig. 1, second line from top) and scores from the Technicon H-l flags (bottom line). Scores from the conventional differential did not reflect eventual improvement, whereas scores from the Coulter flags returned to normal while the patient was still quite sick. The statistical analyses described below compare laboratory scores as predictors of the clinical score at that time for each blood count on each patient. ERRORS IN DIFFERENTIAL COUNTS A detailed analysis of the accuracy of the differential counts is given elsewhere. 15 Virtually every electronic or cytochemical differential was flagged as abnormal (168/ 176 Coulter counts and 169/176 H-l counts), precluding any utility of the flags as indicators for performance of a conventional differential. When five experienced reference technologists reviewed the 176 conventional differential counts, they found 27 (15.3%) in which at least one measurement was significantly different from the reported value. That error rate is considerably higher than our 7% error rate for specimens from adults. 20 The most common errors were in estimating the proportion of band neutrophils and in distinguishing lymphocytes from monocytes. Corrected differential counts, which had been through the review process, were used for subsequent comparisons. CULTURES Cultures from only four infants yielded unequivocally positive results: two blood cultures, one peritoneal fluid, and one urine culture. All four infants were considered independently to show 4+ clinical abnormalities on those days. No type of differential detected all 4 patients, but the immature/total neutrophil ratio and the H-l immature neutrophil flag detected 3 of the 4. DIFFERENTIAL COUNTS VERSUS CLINICAL STATUS Surveillance or Prediction In no instance did an abnormal differential count of any type predict the onset of a previously unrecognized clinical illness. Assessment of Current Status Linear Regression. When all components of the conventional blood count were considered, only platelet count, immature/total neutrophil ratio, and absolute total TABLE 4. PREDICTION OF CLINICAL STATUS FROM COMBINATIONS SCORED LABORATORY MEASUREMENTS Measurement Platelet Absolute Neutrophils or Granulocyte Absolute Immature Neutrophils Immature/ Total Neutrophils Overall Abnormal* r 2 Conventional blood count Coulter differential H-l differential N/A N/A t :): Values listed in the left portion of the table are probabilities that the T statistic is different from zero. N/A = not available on automated counter. Highest degree of abnormality for any component of the differential count. t R3 and RM flags. J Left shift and immature granulocyte flags. A.J.CP. March 1992

6 CHARACHE ET AL. 343 Accuracy and Utility o] rential WBC in the NICU TABLE 5. TRENDS IN SCORED BLOOD COUNTS AS PREDICTORS OF CHANGES IN NEONATES' CLINICAL COURSE Scored Type of Blood Count White blood cell count Platelet count Conventional differential Coulter differential* H-l differential* * Includes "flags." Number of Patient Episodes Gamma P and immature neutrophil counts were significantly related to clinical status (Table 4). Omission of the absolute immature neutrophil count made only a small difference (r 2 = versus 0.321). Analogs of those combinations ("R3 and "RM" or "left shift" and "immature granulocyte" from the S + 4 or the H-l) did less well, but if overall abnormality of the differential was considered as a variable, the conventional differential was the best predictor, followed by the H-1 and then the S + 4. If only measures of neutrophil immaturity from the three types of differential were analyzed, the conventional differential was best, and the Coulter worst (P = , , and 0.45, respectively). The immature/total neutrophil ratio was better than the absolute immature neutrophil count (P = versus 0.031) but there was little difference between the ratio and the immature neutrophil flags from the H-l (P = versus ). Gamma Statistic. Scored white blood cell and platelet counts were related more closely to clinical status than any of the other scored measurements (Table 4). The ratio of immature to total lymphocytes from the conventional differential, the combined immature granulocyte, and left shift flags from the H-l and the absolute granulocyte count from the S + 4 had somewhat lower significance. Trend Analyses. Trends in clinical course were related most closely to trends in the scored white cell count (P < 0.003, Table 5). In contrast to results obtained with individual counts, trends in the platelet count were not closely related to trends in clinical status, probably because platelet transfusions were administered as needed. Trends in the conventional differential count were associated with trends in clinical status (P = 0.028). USE OF AUTOMATED DIFFERENTIAL COUNTS FOR SCREENING If the technologist relies solely on the manufacturers' instructions, neither the S + 4 three-part differential nor the H-1 cytochemical differential has any role to play in the NICU: 97% of the samples would have required conventional differential counts as contrasted to 15% of samples in an adult population studied with the S If blast flags were ignored, and flags from the H-1 had been used as a screening device, 108 rather than 170 tests would have been reported, a saving of more than one third of the examinations. DISCUSSION Three conclusions can be reached from our data. First, errors in differential counts are more common in neonates than in adults. Second, performance of differential counts in apparently healthy neonates is not an effective means of surveillance of unrecognized illness. Finally, carefully reviewed and corrected conventional differential counts are better than their automated analogues for assessing acute illness and for following trends during therapy for a recognized illness, but the differences between the two kinds of differential counts are small and neither type of count is closely related to clinical status. It is difficult for technologists to classify the white cells of neonates. Our error rate was approximately twice as high on such samples as on samples from older patients. If the original uncorrected differential counts had been used for analysis, automated differential counts might have fared even better than they did here. If routine differential counts were only done by reference technologists, those errors might be decreased, but in the current era of cost-containment, possibility of such a change seems a remote. 32 No component of the blood count or differential was of use in predicting the onset of clinically unsuspected illness. Surveillance counts would not, therefore, appear to be cost-effective. When individual counts were analyzed, the platelet count, the white count, and derivatives of the differential count could be used to predict clinical status. The ratio of immature to total neutrophils derived from the conventional differential count appeared to be more useful than absolute neutrophil or absolute immature neutrophil counts, a finding supported by earlier work. 5 ' 2 In many respects, those findings are similar to those in an earlier publication, 26 which found a low platelet count, an abnormal white cell count, and degenerative changes in neutrophils to have specificities of more than 90% for diagnosis of neonatal sepsis. Similar results were obtained using the gamma statistic, although the assumptions of the test (ordered categorized variables) are different from those in regression analysis. Only about one third of the change in current clinical status (r 2 ) could be predicted from the conventional blood count. The H-l was almost as good as the conventional count (0.34 compared to 0.30) if flags were considered valid indicators of a left shift. A search for trends is one of the most common reasons given for serial (performed daily or more frequently) differential counts, primarily to support clinical impressions Vol. 97 No. 3

7 344 HEMATOPATHOLOGY of improvement or deterioration. 33 Trends in the scored white blood count were related most closely to trends in clinical status, but trends in the conventional differential count were only marginally better than automated measurements with the H-1. Questions about the cost-to-efficiency ratio of differential blood counts are being forced on clinical hematology laboratories and will be forced on neonatologists because automated microscopes are no longer being manufactured in the United States and there is a shortage of competent technologists that is as serious as the more publicized shortage of nurses. More widespread (but not exclusive) use of automated electronic differential counts and their flags might permit a cost-effective solution to a perceived need for differential counts in the NICU. Reporting of a left shift or immature granulocyte flag, in that scenario, would have essentially the same clinical implications as a present report of 15% band neutrophils. If clinicians were unwilling to relinquish conventional differential counts, and particularly enumeration of band neutrophils, use of automated differential counts as a screening technique also might decrease use of conventional differential counts by one third or more. Acknowledgments. The authors thank Susan Eckert, Joann Wilson, M.T.(ASCP), and Sandra Humbertson, M.T.(ASCP) for their help in preparing this study. They also thank the staff of the JHH Hematology Laboratory for their help in collection of data, Dr. Douglas Jones and the NICU staff for their cooperation, Dr. Alan M. Gittelsohn and Dr. Allyn W. Kimball, Jr. for statistical advice, the Technicon Corporation for the opportunity to use and understand the H-1, and Dr. Elkin Simson and Cindy Kessler for their helpful comments. REFERENCES 1. Zipursky A, Palko J, Milner R, Akenzua GI. The hematology of bacterial infections in premature infants. Pediatrics 1976;57: Mathy KA, Koepke JA. The clinical usefulness of segmented vs. stab neutrophil criteria for differential leukocyte counts. Am J Clin Pathol 1974;61: Xanthou M. Leucocyte blood picture in ill newborn babies. Arch Dis Child 1972;47: Engle WD, Rosenfeld CR. Neutropenia in high-risk neonates. J Pediatr 1984;105: Manroe BL, Rosenfeld CR, Weinberg AG, Browne R. The differential leukocyte count in the assessment of outcome of early-onset neonatal group B streptococcal disease. J Pediatr 1977;91: Philip AGS. White blood cells and acute phase reactants in neonatal sepsis. Pediatrie 1984;34: Gregor J, Hey E. Blood neutrophil response to bacterial infection in thefirstmonth of life. Arch Dis Child 1972;47: Boyle RJ, Chandler BD, Stonestreet BS, Oh W. Early identification of sepsis in infants with respiratory distress. Pediatrics 1978; 62: Benuck I, David RJ. Sensitivity of published neutrophil indexes in identifying newborn infants with sepsis. J Pediatr 1983; 103: Manroe BL, Weinberg AG, Rosenfeld CR, Browne R. The neonatal blood count in health and disease. I. Reference values for neutrophilic cells. J Pediatr 1979;95: Sen S, Agarwhal KN. Leucocytic alterations in diagnosis of neonatal sepsis. Indian J Med Res 1975;63: Akenzua GI, Hui YT, Milner R, Zipursky A. Neutrophil and band counts in the diagnosis of neonatal infections. Pediatrics 1974; 54: Kuchler H, Flicker H, Gugler E. La formule sanguine dans le diagnostic precoce de la septicemie du nouveau-ne. Helv paediat Acta 1976;31: Christensen RD, Rothstein G. Pitfalls in the interpretation of leukocyte counts in newborn infants. Am J Clin Pathol 1979; 72: Nelson L, Charache S, Wingfield S, Keyser E. Laboratory evaluation of differential white count information from the Coulter S Plus IV and Technicon H-l in patient populations requiring rapid "turn-around" time. Am J Clin Pathol 1989;91: Keitges PW, Koepke JA. Report on hematology photomicrograph transparencies Am J Clin Pathol 1971;55: Winkel P, Statland BE, Saunders AM, Osborn H, Kupperman H. Within-day physiologic variation of leukocyte types in healthy subjects as assayed by two automated leukocyte differential analyzers. Am J Clin Pathol 1981;75: Rumke CL. Variability of results in differential counts on blood smears. Triangle 1959; 4: Agress H, Downey H. The blood picture of human new-borns, with special reference to lymphocytes. Folia Haemat 1936;55: Charache S, Nelson L, Keyser E, Metzger P. A clinical trial of threepart electronic differential white blood cell counts. Arch Intern Med 1985;145: Salzman GC, Crowell JM, Martin JC, et al. Cell classification by laser light scattering: Identification and separation of unstained leukocytes. Acta Cytol 1975; 19: Ross DW, Bardwell A. Automated cytochemistry and the white cell differential in leukemia. In: Ross DW, Brecher G, Bessis M. Automation in Hematology. New York: Springer Verlag, 1981, pp Duncan KL, Gottfried EL. Utility of the three-part leukocyte differential count. Am J Clin Pathol 1987;88: Ross DW, Bentley SA. Evaluation of an automated hematology system (Technicon H-l). Arch Pathol Lab Med 1986; 110: Cranendonk E, van Gennip AH, Abeling NGGM, Behrendt H. Reference values for automated cytochemical differential count of leukocytes in children 0-16 years old: Comparison with manually obtained counts from Wright-stained smears. J Clin Chem Clin Biochem 1985;23: Rodwell RL, Leslie AL, Tudehope DI. Early diagnosis of neonatal sepsis using a hematologic scoring system. J Pediatr 1988; 112: Spector SA, Ticknor W, Grossman M. Study of the usefulness of clinical and hematologic findings in the diagnosis of neonatal bacterial infections. Clin Pediatr 1981;20: Tollner U. Early diagnosis of septicemia in the newborn. Eur J Pediatr 1982;138: Goodman LA, Kruskal WH. Measures of association for cross classifications. J Am Statist Assn 1954;49: Goodman LA, Kruskal WH. Measures of association for cross classifications. III. Approximate sampling theory. J Am Statist Assn 1963;58: Bull BS, Korpman RA. Intralaboratory quality control using patient's data. In Cavill I. Quality Control. Edinburgh: Churchill Livingston, 1982, pp Bentley SA, Pegram MD, Ross DW. Diagnosis of infective and inflammatory disorders byflowcytometric analysis of blood neutrophils. Am J Clin Pathol 1987;88: Kovatch AL, Wald ER. Evaluation of the febrile neonate. Semin Perinatology 1985;9: A.J.C.P. March 1992