- Email: [email protected]
Clinical usefulness of white blood cell count after cesarean delivery
Clinical Usefulness of White Blood Cell Count After Cesarean Delivery KATHERINE E. HARTMANN, MD, PhD, KATHERINE E. BARRETT, MD, VIRGIL C. REID, MD, MICHAEL J. MCMAHON, MD, MPH, AND WILLIAM C. MILLER, MD, PhD, MPH Objective: To examine changes in white blood cell (WBC) count after cesarean and estimate risk of postoperative infection. Methods: We measured complete blood cell counts at admission and on postoperative day 1 for 458 women who had cesareans. Information from charts was abstracted, and definitions of infectious outcomes and fever were applied by three physicians masked to laboratory results. We examined changes in absolute and relative WBC counts by labor status. Likelihood ratios for postoperative infection were calculated for statistically distinct categories of percentage changes. Results: We excluded 60 women with chorioamnionitis. Of the remainder, 34 (8.5%) developed endometritis and three (0.8%) pneumonia. Women who labored before cesarean (n ⴝ 198) had higher antepartum (P < .001) and postoperative day 1 (P < .001) WBC counts than those who did not (n ⴝ 200). However, change in WBC count after cesarean relative to antepartum was similar for both groups (P ⴝ .41), averaging a 22% increase. We grouped percentage changes into the following three levels: up to 24%, 25–99%, and at least 100%. The lowest level (n ⴝ 246) corresponded to a categoryspecific likelihood ratio for diagnosis of serious postpartum infection of 0.5 (95% confidence interval [CI] 0.3, 0.8), the midlevel (n ⴝ 141) to a category-specific likelihood ratio of 1.7 (95% CI 1.2, 2.3), and the highest level (n ⴝ 11) to a category-specific likelihood ratio of 5.8 (95% CI 1.8, 18.7). Conclusion: Labor influenced postcesarean WBC counts but did not obscure changes associated with infection. Information gained from changes in WBC counts can be used to assess risk of infection. (Obstet Gynecol 2000;96:295– 300. © 2000 by The American College of Obstetricians and Gynecologists.)
Complete blood cell counts are measured routinely the day after cesarean. This inexpensive test includes heFrom the Department of Epidemiology, School of Public Health, Department of Obstetrics and Gynecology, and the Department of Medicine, School of Medicine, University of North Carolina, Chapel Hill, North Carolina.
VOL. 96, NO. 2, AUGUST 2000
matocrit, hemoglobin concentration, and number of platelets and white blood cells (WBCs) per L. Although postoperative hematocrit or hemoglobin identifies serious anemia and blood loss, the postoperative platelet and WBC counts rarely are used for clinical decisions. In ordinary circumstances, WBC count is a first-line test for infection. However, it is believed to have little clinical use postpartum because of altered mobilization of WBCs in pregnancy and the puerperium. Pregnancy itself generates a small increase in the number of circulating WBCs.1,2 Pregnant women at term have WBC counts of roughly 1000 cells/mm3 higher than their own nonpregnant counts.3 Active labor is an inflammatory process. White blood cells infiltrate the cervix, releasing enzymes that break down the cervical stroma to allow dilation and effacement.4,5 To facilitate the process, WBCs demarginate from stores into active circulation. Stress and physical exertion during labor also are associated with increased circulating WBCs.6,7 In the absence of clinically apparent infection, labor is associated with a rapid increase in total WBC count of more than 10,000 cells/mm3 followed by a sharp decrease.8,9 Researchers have long recognized that uterine bacterial contamination exists in most laboring women, with the highest levels among those with membrane rupture.10,11 Bacteremia follows placental separation in 11% of cesareans12; however, labor is not a requirement for bacterial exposure. A study of glove contamination at cesarean found 36% of swabs obtained before delivery grew staphyloccoci and 9% of swabs after delivery grew organisms consistent with vaginal flora.13 Given the high prevalence of subclinical bacterial challenge, mobilization of WBCs indicates a deployment of defensive resources; thus, elevated levels might be better for documenting immunocompetence than for predicting risk of clinically important infections. The predominant
0029-7844/00/$20.00 PII S0029-7844(00)00911-X
295
view of postpartum WBC count is characterized by a prominent text: “ . . . in view of the physiologic leukocytosis of the early puerperium, these [WBC] findings are difficult to interpret.”14 Few studies focused on describing WBC counts around childbirth. Most categorize WBC counts as normal or abnormal based on dichotomous cut-points predictive of infection in other settings. We sought to characterize the absolute and relative changes in WBC counts associated with cesarean and to relate levels of change in WBC counts to risk of serious postoperative infection.
Methods Between May 1996 and September 1998, we enrolled 501 women who had cesarean deliveries at the University of North Carolina Women’s Hospital. Each participated in a randomized trial of vaginal preparation with 10% povidone iodine before cesarean. The study was open to all women admitted to labor and delivery who were mentally competent to consent. Exclusion criteria were allergy to povidone iodine, iodine, or shellfish; bleeding placenta previa; active genital herpes lesions; and highly emergent cesarean. Complete medical charts, required to assign diagnoses of infectious morbidity, were available for 498 participants. Analysis required that women have complete blood counts antepartum and on postoperative day 1 (the calendar day after the date of surgery). Nine subjects lacked complete blood counts, one lacked only an antepartum count, and 30 did not have them on postoperative day 1. Most of the latter group had indications, such as hemorrhage, that warranted complete blood counts before postoperative day 1 and did not have a subsequent one on postoperative day 1. Thus, 40 subjects were excluded for inadequate laboratory data, leaving 458 eligible for evaluation. Routine complete blood counts were processed in the clinical hematology laboratory. After discharge, one of four physicians reviewed each participant’s obstetric admission chart and abstracted characteristics necessary to define outcome, such as oral temperature and clinical examination results; and treatment for diagnosed or suspected infections. Demographic characteristics were collected from the Department of Obstetrics and Gynecology Perinatal Database. Two members of the same team of physicians systematically searched the participants’ entire charts a minimum of 3 months after discharge for evidence of complications, predominantly wound healing problems. A maternal-fetal medicine specialist and a generalist, masked to laboratory data, independently assigned diagnostic codes. Discrepant diagnoses,
296 Hartmann et al
White Blood Cell Count
in less than 3% of subjects, prompted joint chart reviews by those individuals and another obstetriciangynecologist. All discrepancies were resolved as oversights of information. To evaluate the importance of WBC counts, the primary outcome was serious postoperative infection that required intravenous broad-spectrum antibiotic administration. Wound separation was a secondary outcome. We established a priori definitions for uterine and pelvic infections, abdominopelvic abscess, wound healing complications, pulmonary morbidity, upper urinary tract infections, thrombophlebitis, and mastitis. Only a few of those conditions were seen. The following are operational definitions for conditions participants experienced: fever, oral temperature of at least 38.0C; chorioamnionitis, oral temperature of at least 37.7C before delivery, followed by broad-spectrum, intravenous antibiotics for treatment of intrauterine infection; endometritis, oral temperature at least 37.7C, with a physician’s note documenting uterine or low abdominal tenderness, followed by broad-spectrum, intravenous antibiotics for treatment of intrauterine infection; pneumonia, oral temperature at least 37.7C, with radiologic confirmation of an infiltrate for which the subject received antibiotic treatment; and wound separation, chart note documenting partial or complete disruption of abdominal incision that required wound care. Data analysis began with inspection of the overall distribution of antepartum and postoperative day 1 WBC counts for departures from normal distribution. All counts were approximately normal. We examined bivariate associations of WBC counts with participants’ demographic characteristics and histories before the current birth. We used t tests to examine the differences of means between groups, paired t tests for comparing antenatal and postoperative means, and calculated 95% confidence intervals (CIs). Median and intraquartile ranges were also calculated. We examined the absolute difference between postoperative day 1 WBC count and antepartum WBC count, and calculated the relative change in WBC count from antepartum to postoperative day 1 (using POD1 for postoperative day 1): Percent change in WBC count ⫽
共POD1 WBC count ⫺ antepartum WBC count兲 antepartum WBC count ⫻ 100.
We investigated WBC counts as they related to presence or absence of the primary outcome, serious postoperative infection, and the secondary outcome, wound separation. Receiver operating characteristic (ROC) curves were constructed for percentage changes in
Obstetrics & Gynecology
Table 1. Selected Characteristics of Participants Characteristic Maternal age (y)† White Black Other races Less than high school education High school education More than high school education Nulliparous Multiparous Multiparous subjects with prior cesarean
All (n ⫽ 398*)
Labor (n ⫽ 198)
No labor (n ⫽ 200)
27.6 ⫾ 6.2 199 (50.0) 112 (28.1) 87 (21.9) 192 (48.2)
27.3 ⫾ 6.1 99 (50.0) 57 (28.8) 42 (21.2) 89 (44.9)
27.9 ⫾ 6.2 100 (50.0) 55 (27.5) 45 (22.5) 103 (51.5)
102 (25.6) 104 (26.1)
50 (25.3) 59 (29.8)
52 (26.0) 45 (22.5)
160 (40.2) 238 (26.1) 171 (43.0)
107 (54.0) 91 (46.0) 57 (28.8)
53 (26.5) 147 (73.5) 114 (57.0)
* Excludes 60 women with chorioamnionitis. Mean ⫾ standard deviation. Except for maternal age, data are given as n (%).
†
WBC counts. Percentage changes were grouped by deciles of change from decreased WBC count through a 200% increase. We then calculated likelihood ratios for serious infection and wound separation. The CIs of the likelihood ratios and test of homogeneity were inspected in order to collapse categories of change in WBC counts that were similar to adjacent categories, to arrive at a parsimonious categorization of changes in WBC counts.15 To investigate the potential influence of misclassification bias, we compared results of analyses that required strictly defined fever for a diagnosis of serious postoperative infection (eg, a woman treated for endometritis who had a postoperative temperature under 38.0C was not considered a case of serious infection) with the more lenient clinical criteria of temperature of at least 37.7C.
Results Among 458 subjects, 253 actively labored before cesarean and 205 did not. Sixty subjects (13%) met criteria for chorioamnionitis, 55 who labored and five who did not. The mean (⫾ standard deviation [SD]) antepartum WBC count on admission to labor and delivery in those with chorioamnionitis was 11.3 ⫾ 4.6 cells/mm3 ⫻ 10⫺3; on postoperative day 1 it was 14.5 ⫾ 8.6 cells/mm3 ⫻ 10⫺3. Because the diagnosis was established before the postpartum period, women with chorioamnionitis were excluded from further consideration of using WBC counts for predicting the likelihood of serious postoperative infectious morbidity. There remained 398 subjects who were not diagnosed and treated for antepartum infections (Table 1). Thirtyfour women (8.5%) developed endometritis, two had pneumonia, and one had endometritis and pneumonia.
VOL. 96, NO. 2, AUGUST 2000
Among those 36 women, 23 were from the group that labored and 13 from the group that did not, for infectious complication rates of 11.5% and 6.5%, respectively. Eighty-four subjects (21%), including 50 who labored and 34 who did not, had fevers of at least 38.0C on postoperative day 1 or later. If strictly defined fever was required for the diagnosis of infectious complications, 31 subjects met criteria for endometritis, two for pneumonia, and one for both. Twenty-eight women (7.0%) had wound healing complications; four cases occurred among those who had endometritis. Seventeen of those 28 women had complete superficial wound disruption to the level of the fascia. There were no cases of fascial dehiscence. Records that described the characteristics of the incision at initial wound care for the separation indicated that six were attributed to hematomas, 12 to seromas, and seven to gross infections. There was no clear description of the appearance of the remainder. The mean antepartum WBC count was 1.5 cells/mm3 ⫻ 10⫺3 higher among those who labored compared with those who did not (Table 2). Both groups had a modest increase in WBC counts after cesarean, an average of 1.7 cells/mm3 ⫻ 10⫺3 among those who labored and 1.5 cells/mm3 ⫻ 10⫺3 among those who did not. Considered individually, 70% of those who labored and those who did not had some increase in WBC count. An average increase of 1.5 cells/mm3 ⫻ 10⫺3 remained for both groups when subjects with subsequent diagnoses of infection were excluded, suggesting that cesarean delivery independently contributed to WBC count increases regardless of labor status. The ranges of percentage changes in WBC counts were ⫺50.2% to ⫹258.5% for those who labored and ⫺31.6% to ⫹125.5% for those who did not. Average absolute changes and percentage changes in WBC counts were relatively insensitive to labor status and were not statistically distinct. Change in WBC was not influenced by antenatal steroid administration (P ⫽ .41 for absolute change; P ⫽ .39 for percentage change). For all subjects combined, measures of change were more informative than the absolute value of WBC counts on postoperative day 1. Traditional diagnostic test characteristics for predicting postcesarean infections are summarized in Table 3. Using dichotomous cut-points, change in WBC count lacked sensitivity, although low cut-points were specific and can be used to assess whether a patient is at low risk of infection. We examined ROC curves for percentage changes in WBC counts (not shown). The curves were relatively flat with no distinct shoulder that optimized sensitivity while minimizing false positives. The area under the curve using all data was 0.67. Those characteristics
Hartmann et al
White Blood Cell Count
297
Table 2. Antepartum and Postoperative Day 1 White Blood Cell Counts by Labor Status White blood cell count (cells/mm3 ⫻ 10⫺3) Labor status
Postoperative day 1
Absolute difference
Percentage difference
9.5 (9.1, 9.9) 9.2 7.3–11.1
11.0 (10.6, 11.5) 10.5 8.8 –12.6
1.5 (1.2, 1.9) 1.4 ⫺0.3–3.2
20.0 (15.6, 24.3) 16.7 ⫺3.7–39.0
11.0 (10.5, 11.6) 10.4 8.6 –12.7 ⬍.001
12.8 (12.2, 13.3) 12.0 10.3–14.2 ⬍.001
1.7 (1.2, 2.3) 1.7 ⫺0.3–3.5 .55
23.1 (17.1, 29.0) 15.5 ⫺2.5– 42.3 .41
Antepartum
No labor before cesarean (n ⫽ 200) Mean 95% CI Median Intraquartile range Labor before cesarean (n ⫽ 198) Mean 95% CI Median Intraquartile range P for no labor compared with labor CI ⫽ confidence interval.
indicated limited diagnostic usefulness based on a dichotomous cut-point. Category-specific likelihood ratios represent the change in odds of disease for a given test result. For clinical diagnosis of serious infection for women whose WBC counts double or more (at least 100%), the category-specific likelihood ratio was 5.8 (95% CI 1.8, 18.7) (Table 4); for those with increases from 25–99%, the category-specific likelihood ratio was 1.7 (95% CI 1.2, 2.3); and for those with up to 24% increases (including WBC count decreases), the category-specific likelihood ratio was 0.5 (95% CI 0.3, 0.8). Estimates were similar whether a fever of at least 38.0C was required for diagnosis of infection. Change in WBC count did not contribute substantially to identification of women at risk of wound healing complications. When women who also had endometritis were excluded (n ⫽ 4), the category-specific likelihood ratio of WBC count increase of at least 100% was 3.42 (95% CI 0.8, 15.0) and was 1.08 for lower increases (95% CI 0.8, 1.45). Category-specific likelihood ratios provide information that can be used to calculate the likelihood of serious postpartum infection across a range of test results and pretest probabilities (Figure 1). As the figure illustrates, changes of WBC exceeding 100% or under 25% could affect treatment decisions.
Discussion Lack of a meaningful standard is the central limitation of this study and others that evaluated risk and diagnosis of infection-related outcomes. Diagnoses such as chorioamnionitis and endometritis are often retrospective, based on physicians’ responses to patients’ symptoms. In the jargon of logic, the consequent, treatment for infection, affirms the antecedent, the physician’s judgment that the patient has an infection. Puerperal infections are diagnosed and treated before objective indicators are established, which often are not sought. Postpartum uterine cultures are not routine because they are costly, hard to obtain, produce results too late for clinical use, and rarely change treatment. Therefore, studying correlates of postpartum infection has built-in circularity: clinical diagnoses define the outcome, yet those diagnoses are based on the providers’ perceptions of risk and source of infection. White blood cell counts have properties that made it feasible to evaluate their importance. They were uniformly measured on postoperative day 1, so diagnostic bias associated with ordering of tests only for those patients who provoke clinical suspicion of disease was prevented. Postoperative day 1 WBC counts are not currently used in clinical decision-making. Even if phy-
Table 3. Test Characteristics of Dichotomous White Blood Cell Count Increases for Diagnosis of Infection After Cesarean Increase
Positive tests (%)
Sensitivity (%)
Specificity (%)
Positive predictive value (%)
Negative predictive value (%)
2.8 7.5 14.6 38.2
11 20 31 67
98 94 87 65
36 23 19 16
92 92 93 95
ⱖ100% ⱖ75% ⱖ50% ⱖ25%
298 Hartmann et al
White Blood Cell Count
Obstetrics & Gynecology
Table 4. Percentage Increase in White Blood Cell Count by Infection Status
Percentage increase in WBC count antepartum to postoperative day 1 ⱖ100 25–99 ⱕ24
Patients without infection n ⫽ 362 (n ⫽ 365)*
n
Proportion
n
Proportion
LR for clinical diagnosis of infection (95% CI)
4 (4) 20 (18) 12 (11)
0.11 0.55 0.33
7 121 234
0.02 0.33 0.65
5.8 (1.8, 18.7) 1.7 (1.2, 2.3) 0.5 (0.3, 0.8)
Patients with infection n ⫽ 36 (n ⫽ 33)*
LR for clinical diagnosis of infection and fever ⱖ38.0C (95% CI) 6.3 (1.9, 20.5) 1.6 (1.1, 2.3) 0.5 (0.3, 0.8)
WBC ⫽ white blood cell; LR ⫽ likelihood ratio; CI ⫽ confidence interval. * Definition requiring clinical diagnosis of infection and fever ⱖ 38.0C on or after postoperative day 1.
sicians noted the counts, the relative change in count would not likely be considered important. We are confident, based on our clinical experience, that WBC count did not influence decisions to initiate treatment and did not contribute to defining outcomes. By using operational definitions for infection-related outcomes and by masking the coding physicians to laboratory data, we ensured that uniform criteria were applied without influence from WBC count values. If the association between WBC count changes and likelihood of infection is to be clinically useful, the laboratory result must be available before treatment decisions are made. We examined the time sequence of onset of postpartum fever, WBC count, symptoms, and initiation of treatment. Among 36 subjects with postpartum infection, five had temperature elevations and uterine tenderness and initiation of antibiotics on the day of surgery, four were diagnosed and treated before noon on postoperative day 1, 12 were treated after noon on postoperative day 1, and 15 were treated after postoperative day 1. If noon is arbitrarily selected as the time by which results of morning blood collection complete
Figure 1. Probability of infection adjusted for change in white blood cell count. The clinician’s assessment (x-axis) provides the pretest probability, which is adjusted for WBC count using category-specific likelihood ratios (LR), to determine the posttest probability (y-axis) of serious infection.
VOL. 96, NO. 2, AUGUST 2000
blood counts would be available to clinicians, for 27 of 36 (75%) women, physicians had time to use WBC counts in clinical decision-making. In this cohort, using laboratory computer reporting records, we determined that laboratory results were available for 30 of 36 women (83%). After eliminating the five women for whom WBC count values could not contribute, the category-specific likelihood ratios for diagnosis of infection were 6.7 (95% CI 2.1, 22.6), 1.6 (95% CI 1.2, 2.3), and 0.5 (95% CI 0.3, 0.8) from the highest to lowest category of percentage changes. For women who did not have infection or receive antibiotics, the timing of laboratory results that might have confirmed clinical judgment cannot be determined because we lacked a point of reference to establish when physicians were confident that a participant’s postoperative course was uncomplicated. When the change in WBC count from preoperative to postoperative period is less than 25%, or is 100% (double) or greater, that information can be used to adjust the probability of infection. In our population, 64% of subjects were in one of those two categories. The decision to start antibiotics is based on the clinician’s estimate of the probability that a woman is infected and whether that probability exceeds a critical treatment threshold that is individualized and rarely explicit. For example, a physician with an aggressive antibiotic treatment philosophy might favor starting antibiotics whenever the probability of infection exceeds 25%. A physician with a conservative treatment philosophy might reserve treatment until the probability is 60% or greater. Medical costs are incurred from premature decisions to treat patients who might not need antibiotics, prolonging their stays and increasing costs, and from delayed initiation of antibiotics in those with infections, also prolonging their stays and increasing risk of complications. Consider a physician with a treatment threshold of 40%; with a probability of infection of 40%, the physician would initiate therapy. After evaluating a patient, that physician estimates the probability of in-
Hartmann et al
White Blood Cell Count
299
fection at 15–20%. If the patient has a change in WBC count that exceeds 100%, treatment would be warranted (Figure 1). Consider a second patient with an estimated probability of infection of 40%, representing a difficult decision for that clinician. If the woman’s WBC count increased only 12%, she would have a posttest probability of infection of only about 22% (Figure 1), indicating that treatment could be withheld with observation for change in status. Our findings lead to additional questions. Would characterization of the WBC subtypes that comprise the total WBC count distinguish normal from infectionrelated increases in WBC counts? Would similar patterns be observed among women who have vaginal births, or in a broader sample of those who have cesareans, or across many infectious outcomes? With careful attention to uniform definitions of outcomes, larger future studies could address those questions. Additional studies to derive likelihood ratios with greater precision and evaluate the sequence of symptoms, signs, diagnosis, and treatment will allow modeling of usefulness and cost implications of using those ratios in clinical decision-making.
8.
9. 10. 11. 12.
13. 14.
15.
Address reprint requests to:
References 1. Efrati P, Presentey B, Margalith M, Rozenszajn. Leukocytes of normal pregnant women. Obstet Gynecol 1963;23:429 –32. 2. Tysoe FW, Lowenstein L. Blood volume and hematologic studies in pregnancy and the puerperium. Am J Obstet Gynecol 1950;60: 1187–205. 3. Pitkin R, Witte D. Platelet and leukocyte counts in pregnancy. JAMA 1979;242:2696 – 8. 4. Uldbjerg N, Ekman G, Malmstrom A, Olsson K, Ulmsten U. Ripening of the human uterine cervix related to changes in collagen, glycosaminoglycans, and collagenolytic activity. Am J Obstet Gynecol 1983;147:662– 6. 5. Danforth DN, Veis A, Breen M, Weinstein HG, Buckingham JC, Manalo P. The effect of pregnancy and labor on the human cervix: Changes in collagen, glycoproteins, and glycosaminoglycans. Am J Obstet Gynecol 1974;120:641–51. 6. Landmann RM, Muller FB, Perini C, Wesp M, Erne P, Buhler FR.
300 Hartmann et al
7.
Changes of immunoregulatory cells induced by psychological and physical stress: Relationship to plasma catecholamines. Clin Exp Immunol 1984;58:127–35. Oshida Y, Yamanouchi K, Hayamizu S, Sato Y. Effect of acute physical exercise on lymphocyte subpopulations in trained and untrained subjects. Int J Sports Med 1988;9:137– 40. Delgado I, Neubert R, Dudenhausen JW. Changes in white blood cells during parturition in mothers and newborn. Gynecol Obstet Invest 1994;38:227–35. Gibson A. On leucocyte changes during labor and the puerperium. J Obstet Gynecol Br Empire 1937;44:500 –9. Harris JW, Brown H. Bacterial content of the uterus at cesarean section. Am J Obstet Gynecol 1927;13:133. Gilstrap LC, Cunningham FG. The bacterial pathogenesis of infection following cesarean section. Obstet Gynecol 1979;53:545–9. Boggess KA, Watts DH, Hillier SL, Krohn MA, Benedetti TJ. Bacteremia shortly after placental separation during cesarean delivery. Obstet Gynecol 1996;779 – 84. Yancey MK, Clark P, Duff P. The frequency of glove contamination during cesarean delivery. Obstet Gynecol 1994;83:538 – 42. Cunningham FG, MacDonald PC, Gant NF, Leveno KJ, Gilstrap LC, Hankins GDV, et al, eds. Williams obstetrics. 20th ed. Stamford, Connecticut: Appleton & Lange, 1997:551. Peirce JC, Cornell RG. Integrating stratum-specific likelihood ratios with the analysis of ROC curves. Medical Decis Making 1993;13:141–51.
White Blood Cell Count
Katherine E. Hartmann, MD, PhD University of North Carolina, Chapel Hill North Carolina Program for Women’s Health Research Cecil G. Sheps Center for Health Services Research 725 Airport Road, CB #7590 Chapel Hill, NC 27599-7590 E-mail: [email protected]
Received December 17, 1999. Received in revised form March 14, 2000. Accepted April 7, 2000.
Copyright © 2000 by The American College of Obstetricians and Gynecologists. Published by Elsevier Science Inc.
Obstetrics & Gynecology
Complete blood cell counts are measured routinely the day after cesarean. This inexpensive test includes heFrom the Department of Epidemiology, School of Public Health, Department of Obstetrics and Gynecology, and the Department of Medicine, School of Medicine, University of North Carolina, Chapel Hill, North Carolina.
VOL. 96, NO. 2, AUGUST 2000
matocrit, hemoglobin concentration, and number of platelets and white blood cells (WBCs) per L. Although postoperative hematocrit or hemoglobin identifies serious anemia and blood loss, the postoperative platelet and WBC counts rarely are used for clinical decisions. In ordinary circumstances, WBC count is a first-line test for infection. However, it is believed to have little clinical use postpartum because of altered mobilization of WBCs in pregnancy and the puerperium. Pregnancy itself generates a small increase in the number of circulating WBCs.1,2 Pregnant women at term have WBC counts of roughly 1000 cells/mm3 higher than their own nonpregnant counts.3 Active labor is an inflammatory process. White blood cells infiltrate the cervix, releasing enzymes that break down the cervical stroma to allow dilation and effacement.4,5 To facilitate the process, WBCs demarginate from stores into active circulation. Stress and physical exertion during labor also are associated with increased circulating WBCs.6,7 In the absence of clinically apparent infection, labor is associated with a rapid increase in total WBC count of more than 10,000 cells/mm3 followed by a sharp decrease.8,9 Researchers have long recognized that uterine bacterial contamination exists in most laboring women, with the highest levels among those with membrane rupture.10,11 Bacteremia follows placental separation in 11% of cesareans12; however, labor is not a requirement for bacterial exposure. A study of glove contamination at cesarean found 36% of swabs obtained before delivery grew staphyloccoci and 9% of swabs after delivery grew organisms consistent with vaginal flora.13 Given the high prevalence of subclinical bacterial challenge, mobilization of WBCs indicates a deployment of defensive resources; thus, elevated levels might be better for documenting immunocompetence than for predicting risk of clinically important infections. The predominant
0029-7844/00/$20.00 PII S0029-7844(00)00911-X
295
view of postpartum WBC count is characterized by a prominent text: “ . . . in view of the physiologic leukocytosis of the early puerperium, these [WBC] findings are difficult to interpret.”14 Few studies focused on describing WBC counts around childbirth. Most categorize WBC counts as normal or abnormal based on dichotomous cut-points predictive of infection in other settings. We sought to characterize the absolute and relative changes in WBC counts associated with cesarean and to relate levels of change in WBC counts to risk of serious postoperative infection.
Methods Between May 1996 and September 1998, we enrolled 501 women who had cesarean deliveries at the University of North Carolina Women’s Hospital. Each participated in a randomized trial of vaginal preparation with 10% povidone iodine before cesarean. The study was open to all women admitted to labor and delivery who were mentally competent to consent. Exclusion criteria were allergy to povidone iodine, iodine, or shellfish; bleeding placenta previa; active genital herpes lesions; and highly emergent cesarean. Complete medical charts, required to assign diagnoses of infectious morbidity, were available for 498 participants. Analysis required that women have complete blood counts antepartum and on postoperative day 1 (the calendar day after the date of surgery). Nine subjects lacked complete blood counts, one lacked only an antepartum count, and 30 did not have them on postoperative day 1. Most of the latter group had indications, such as hemorrhage, that warranted complete blood counts before postoperative day 1 and did not have a subsequent one on postoperative day 1. Thus, 40 subjects were excluded for inadequate laboratory data, leaving 458 eligible for evaluation. Routine complete blood counts were processed in the clinical hematology laboratory. After discharge, one of four physicians reviewed each participant’s obstetric admission chart and abstracted characteristics necessary to define outcome, such as oral temperature and clinical examination results; and treatment for diagnosed or suspected infections. Demographic characteristics were collected from the Department of Obstetrics and Gynecology Perinatal Database. Two members of the same team of physicians systematically searched the participants’ entire charts a minimum of 3 months after discharge for evidence of complications, predominantly wound healing problems. A maternal-fetal medicine specialist and a generalist, masked to laboratory data, independently assigned diagnostic codes. Discrepant diagnoses,
296 Hartmann et al
White Blood Cell Count
in less than 3% of subjects, prompted joint chart reviews by those individuals and another obstetriciangynecologist. All discrepancies were resolved as oversights of information. To evaluate the importance of WBC counts, the primary outcome was serious postoperative infection that required intravenous broad-spectrum antibiotic administration. Wound separation was a secondary outcome. We established a priori definitions for uterine and pelvic infections, abdominopelvic abscess, wound healing complications, pulmonary morbidity, upper urinary tract infections, thrombophlebitis, and mastitis. Only a few of those conditions were seen. The following are operational definitions for conditions participants experienced: fever, oral temperature of at least 38.0C; chorioamnionitis, oral temperature of at least 37.7C before delivery, followed by broad-spectrum, intravenous antibiotics for treatment of intrauterine infection; endometritis, oral temperature at least 37.7C, with a physician’s note documenting uterine or low abdominal tenderness, followed by broad-spectrum, intravenous antibiotics for treatment of intrauterine infection; pneumonia, oral temperature at least 37.7C, with radiologic confirmation of an infiltrate for which the subject received antibiotic treatment; and wound separation, chart note documenting partial or complete disruption of abdominal incision that required wound care. Data analysis began with inspection of the overall distribution of antepartum and postoperative day 1 WBC counts for departures from normal distribution. All counts were approximately normal. We examined bivariate associations of WBC counts with participants’ demographic characteristics and histories before the current birth. We used t tests to examine the differences of means between groups, paired t tests for comparing antenatal and postoperative means, and calculated 95% confidence intervals (CIs). Median and intraquartile ranges were also calculated. We examined the absolute difference between postoperative day 1 WBC count and antepartum WBC count, and calculated the relative change in WBC count from antepartum to postoperative day 1 (using POD1 for postoperative day 1): Percent change in WBC count ⫽
共POD1 WBC count ⫺ antepartum WBC count兲 antepartum WBC count ⫻ 100.
We investigated WBC counts as they related to presence or absence of the primary outcome, serious postoperative infection, and the secondary outcome, wound separation. Receiver operating characteristic (ROC) curves were constructed for percentage changes in
Obstetrics & Gynecology
Table 1. Selected Characteristics of Participants Characteristic Maternal age (y)† White Black Other races Less than high school education High school education More than high school education Nulliparous Multiparous Multiparous subjects with prior cesarean
All (n ⫽ 398*)
Labor (n ⫽ 198)
No labor (n ⫽ 200)
27.6 ⫾ 6.2 199 (50.0) 112 (28.1) 87 (21.9) 192 (48.2)
27.3 ⫾ 6.1 99 (50.0) 57 (28.8) 42 (21.2) 89 (44.9)
27.9 ⫾ 6.2 100 (50.0) 55 (27.5) 45 (22.5) 103 (51.5)
102 (25.6) 104 (26.1)
50 (25.3) 59 (29.8)
52 (26.0) 45 (22.5)
160 (40.2) 238 (26.1) 171 (43.0)
107 (54.0) 91 (46.0) 57 (28.8)
53 (26.5) 147 (73.5) 114 (57.0)
* Excludes 60 women with chorioamnionitis. Mean ⫾ standard deviation. Except for maternal age, data are given as n (%).
†
WBC counts. Percentage changes were grouped by deciles of change from decreased WBC count through a 200% increase. We then calculated likelihood ratios for serious infection and wound separation. The CIs of the likelihood ratios and test of homogeneity were inspected in order to collapse categories of change in WBC counts that were similar to adjacent categories, to arrive at a parsimonious categorization of changes in WBC counts.15 To investigate the potential influence of misclassification bias, we compared results of analyses that required strictly defined fever for a diagnosis of serious postoperative infection (eg, a woman treated for endometritis who had a postoperative temperature under 38.0C was not considered a case of serious infection) with the more lenient clinical criteria of temperature of at least 37.7C.
Results Among 458 subjects, 253 actively labored before cesarean and 205 did not. Sixty subjects (13%) met criteria for chorioamnionitis, 55 who labored and five who did not. The mean (⫾ standard deviation [SD]) antepartum WBC count on admission to labor and delivery in those with chorioamnionitis was 11.3 ⫾ 4.6 cells/mm3 ⫻ 10⫺3; on postoperative day 1 it was 14.5 ⫾ 8.6 cells/mm3 ⫻ 10⫺3. Because the diagnosis was established before the postpartum period, women with chorioamnionitis were excluded from further consideration of using WBC counts for predicting the likelihood of serious postoperative infectious morbidity. There remained 398 subjects who were not diagnosed and treated for antepartum infections (Table 1). Thirtyfour women (8.5%) developed endometritis, two had pneumonia, and one had endometritis and pneumonia.
VOL. 96, NO. 2, AUGUST 2000
Among those 36 women, 23 were from the group that labored and 13 from the group that did not, for infectious complication rates of 11.5% and 6.5%, respectively. Eighty-four subjects (21%), including 50 who labored and 34 who did not, had fevers of at least 38.0C on postoperative day 1 or later. If strictly defined fever was required for the diagnosis of infectious complications, 31 subjects met criteria for endometritis, two for pneumonia, and one for both. Twenty-eight women (7.0%) had wound healing complications; four cases occurred among those who had endometritis. Seventeen of those 28 women had complete superficial wound disruption to the level of the fascia. There were no cases of fascial dehiscence. Records that described the characteristics of the incision at initial wound care for the separation indicated that six were attributed to hematomas, 12 to seromas, and seven to gross infections. There was no clear description of the appearance of the remainder. The mean antepartum WBC count was 1.5 cells/mm3 ⫻ 10⫺3 higher among those who labored compared with those who did not (Table 2). Both groups had a modest increase in WBC counts after cesarean, an average of 1.7 cells/mm3 ⫻ 10⫺3 among those who labored and 1.5 cells/mm3 ⫻ 10⫺3 among those who did not. Considered individually, 70% of those who labored and those who did not had some increase in WBC count. An average increase of 1.5 cells/mm3 ⫻ 10⫺3 remained for both groups when subjects with subsequent diagnoses of infection were excluded, suggesting that cesarean delivery independently contributed to WBC count increases regardless of labor status. The ranges of percentage changes in WBC counts were ⫺50.2% to ⫹258.5% for those who labored and ⫺31.6% to ⫹125.5% for those who did not. Average absolute changes and percentage changes in WBC counts were relatively insensitive to labor status and were not statistically distinct. Change in WBC was not influenced by antenatal steroid administration (P ⫽ .41 for absolute change; P ⫽ .39 for percentage change). For all subjects combined, measures of change were more informative than the absolute value of WBC counts on postoperative day 1. Traditional diagnostic test characteristics for predicting postcesarean infections are summarized in Table 3. Using dichotomous cut-points, change in WBC count lacked sensitivity, although low cut-points were specific and can be used to assess whether a patient is at low risk of infection. We examined ROC curves for percentage changes in WBC counts (not shown). The curves were relatively flat with no distinct shoulder that optimized sensitivity while minimizing false positives. The area under the curve using all data was 0.67. Those characteristics
Hartmann et al
White Blood Cell Count
297
Table 2. Antepartum and Postoperative Day 1 White Blood Cell Counts by Labor Status White blood cell count (cells/mm3 ⫻ 10⫺3) Labor status
Postoperative day 1
Absolute difference
Percentage difference
9.5 (9.1, 9.9) 9.2 7.3–11.1
11.0 (10.6, 11.5) 10.5 8.8 –12.6
1.5 (1.2, 1.9) 1.4 ⫺0.3–3.2
20.0 (15.6, 24.3) 16.7 ⫺3.7–39.0
11.0 (10.5, 11.6) 10.4 8.6 –12.7 ⬍.001
12.8 (12.2, 13.3) 12.0 10.3–14.2 ⬍.001
1.7 (1.2, 2.3) 1.7 ⫺0.3–3.5 .55
23.1 (17.1, 29.0) 15.5 ⫺2.5– 42.3 .41
Antepartum
No labor before cesarean (n ⫽ 200) Mean 95% CI Median Intraquartile range Labor before cesarean (n ⫽ 198) Mean 95% CI Median Intraquartile range P for no labor compared with labor CI ⫽ confidence interval.
indicated limited diagnostic usefulness based on a dichotomous cut-point. Category-specific likelihood ratios represent the change in odds of disease for a given test result. For clinical diagnosis of serious infection for women whose WBC counts double or more (at least 100%), the category-specific likelihood ratio was 5.8 (95% CI 1.8, 18.7) (Table 4); for those with increases from 25–99%, the category-specific likelihood ratio was 1.7 (95% CI 1.2, 2.3); and for those with up to 24% increases (including WBC count decreases), the category-specific likelihood ratio was 0.5 (95% CI 0.3, 0.8). Estimates were similar whether a fever of at least 38.0C was required for diagnosis of infection. Change in WBC count did not contribute substantially to identification of women at risk of wound healing complications. When women who also had endometritis were excluded (n ⫽ 4), the category-specific likelihood ratio of WBC count increase of at least 100% was 3.42 (95% CI 0.8, 15.0) and was 1.08 for lower increases (95% CI 0.8, 1.45). Category-specific likelihood ratios provide information that can be used to calculate the likelihood of serious postpartum infection across a range of test results and pretest probabilities (Figure 1). As the figure illustrates, changes of WBC exceeding 100% or under 25% could affect treatment decisions.
Discussion Lack of a meaningful standard is the central limitation of this study and others that evaluated risk and diagnosis of infection-related outcomes. Diagnoses such as chorioamnionitis and endometritis are often retrospective, based on physicians’ responses to patients’ symptoms. In the jargon of logic, the consequent, treatment for infection, affirms the antecedent, the physician’s judgment that the patient has an infection. Puerperal infections are diagnosed and treated before objective indicators are established, which often are not sought. Postpartum uterine cultures are not routine because they are costly, hard to obtain, produce results too late for clinical use, and rarely change treatment. Therefore, studying correlates of postpartum infection has built-in circularity: clinical diagnoses define the outcome, yet those diagnoses are based on the providers’ perceptions of risk and source of infection. White blood cell counts have properties that made it feasible to evaluate their importance. They were uniformly measured on postoperative day 1, so diagnostic bias associated with ordering of tests only for those patients who provoke clinical suspicion of disease was prevented. Postoperative day 1 WBC counts are not currently used in clinical decision-making. Even if phy-
Table 3. Test Characteristics of Dichotomous White Blood Cell Count Increases for Diagnosis of Infection After Cesarean Increase
Positive tests (%)
Sensitivity (%)
Specificity (%)
Positive predictive value (%)
Negative predictive value (%)
2.8 7.5 14.6 38.2
11 20 31 67
98 94 87 65
36 23 19 16
92 92 93 95
ⱖ100% ⱖ75% ⱖ50% ⱖ25%
298 Hartmann et al
White Blood Cell Count
Obstetrics & Gynecology
Table 4. Percentage Increase in White Blood Cell Count by Infection Status
Percentage increase in WBC count antepartum to postoperative day 1 ⱖ100 25–99 ⱕ24
Patients without infection n ⫽ 362 (n ⫽ 365)*
n
Proportion
n
Proportion
LR for clinical diagnosis of infection (95% CI)
4 (4) 20 (18) 12 (11)
0.11 0.55 0.33
7 121 234
0.02 0.33 0.65
5.8 (1.8, 18.7) 1.7 (1.2, 2.3) 0.5 (0.3, 0.8)
Patients with infection n ⫽ 36 (n ⫽ 33)*
LR for clinical diagnosis of infection and fever ⱖ38.0C (95% CI) 6.3 (1.9, 20.5) 1.6 (1.1, 2.3) 0.5 (0.3, 0.8)
WBC ⫽ white blood cell; LR ⫽ likelihood ratio; CI ⫽ confidence interval. * Definition requiring clinical diagnosis of infection and fever ⱖ 38.0C on or after postoperative day 1.
sicians noted the counts, the relative change in count would not likely be considered important. We are confident, based on our clinical experience, that WBC count did not influence decisions to initiate treatment and did not contribute to defining outcomes. By using operational definitions for infection-related outcomes and by masking the coding physicians to laboratory data, we ensured that uniform criteria were applied without influence from WBC count values. If the association between WBC count changes and likelihood of infection is to be clinically useful, the laboratory result must be available before treatment decisions are made. We examined the time sequence of onset of postpartum fever, WBC count, symptoms, and initiation of treatment. Among 36 subjects with postpartum infection, five had temperature elevations and uterine tenderness and initiation of antibiotics on the day of surgery, four were diagnosed and treated before noon on postoperative day 1, 12 were treated after noon on postoperative day 1, and 15 were treated after postoperative day 1. If noon is arbitrarily selected as the time by which results of morning blood collection complete
Figure 1. Probability of infection adjusted for change in white blood cell count. The clinician’s assessment (x-axis) provides the pretest probability, which is adjusted for WBC count using category-specific likelihood ratios (LR), to determine the posttest probability (y-axis) of serious infection.
VOL. 96, NO. 2, AUGUST 2000
blood counts would be available to clinicians, for 27 of 36 (75%) women, physicians had time to use WBC counts in clinical decision-making. In this cohort, using laboratory computer reporting records, we determined that laboratory results were available for 30 of 36 women (83%). After eliminating the five women for whom WBC count values could not contribute, the category-specific likelihood ratios for diagnosis of infection were 6.7 (95% CI 2.1, 22.6), 1.6 (95% CI 1.2, 2.3), and 0.5 (95% CI 0.3, 0.8) from the highest to lowest category of percentage changes. For women who did not have infection or receive antibiotics, the timing of laboratory results that might have confirmed clinical judgment cannot be determined because we lacked a point of reference to establish when physicians were confident that a participant’s postoperative course was uncomplicated. When the change in WBC count from preoperative to postoperative period is less than 25%, or is 100% (double) or greater, that information can be used to adjust the probability of infection. In our population, 64% of subjects were in one of those two categories. The decision to start antibiotics is based on the clinician’s estimate of the probability that a woman is infected and whether that probability exceeds a critical treatment threshold that is individualized and rarely explicit. For example, a physician with an aggressive antibiotic treatment philosophy might favor starting antibiotics whenever the probability of infection exceeds 25%. A physician with a conservative treatment philosophy might reserve treatment until the probability is 60% or greater. Medical costs are incurred from premature decisions to treat patients who might not need antibiotics, prolonging their stays and increasing costs, and from delayed initiation of antibiotics in those with infections, also prolonging their stays and increasing risk of complications. Consider a physician with a treatment threshold of 40%; with a probability of infection of 40%, the physician would initiate therapy. After evaluating a patient, that physician estimates the probability of in-
Hartmann et al
White Blood Cell Count
299
fection at 15–20%. If the patient has a change in WBC count that exceeds 100%, treatment would be warranted (Figure 1). Consider a second patient with an estimated probability of infection of 40%, representing a difficult decision for that clinician. If the woman’s WBC count increased only 12%, she would have a posttest probability of infection of only about 22% (Figure 1), indicating that treatment could be withheld with observation for change in status. Our findings lead to additional questions. Would characterization of the WBC subtypes that comprise the total WBC count distinguish normal from infectionrelated increases in WBC counts? Would similar patterns be observed among women who have vaginal births, or in a broader sample of those who have cesareans, or across many infectious outcomes? With careful attention to uniform definitions of outcomes, larger future studies could address those questions. Additional studies to derive likelihood ratios with greater precision and evaluate the sequence of symptoms, signs, diagnosis, and treatment will allow modeling of usefulness and cost implications of using those ratios in clinical decision-making.
8.
9. 10. 11. 12.
13. 14.
15.
Address reprint requests to:
References 1. Efrati P, Presentey B, Margalith M, Rozenszajn. Leukocytes of normal pregnant women. Obstet Gynecol 1963;23:429 –32. 2. Tysoe FW, Lowenstein L. Blood volume and hematologic studies in pregnancy and the puerperium. Am J Obstet Gynecol 1950;60: 1187–205. 3. Pitkin R, Witte D. Platelet and leukocyte counts in pregnancy. JAMA 1979;242:2696 – 8. 4. Uldbjerg N, Ekman G, Malmstrom A, Olsson K, Ulmsten U. Ripening of the human uterine cervix related to changes in collagen, glycosaminoglycans, and collagenolytic activity. Am J Obstet Gynecol 1983;147:662– 6. 5. Danforth DN, Veis A, Breen M, Weinstein HG, Buckingham JC, Manalo P. The effect of pregnancy and labor on the human cervix: Changes in collagen, glycoproteins, and glycosaminoglycans. Am J Obstet Gynecol 1974;120:641–51. 6. Landmann RM, Muller FB, Perini C, Wesp M, Erne P, Buhler FR.
300 Hartmann et al
7.
Changes of immunoregulatory cells induced by psychological and physical stress: Relationship to plasma catecholamines. Clin Exp Immunol 1984;58:127–35. Oshida Y, Yamanouchi K, Hayamizu S, Sato Y. Effect of acute physical exercise on lymphocyte subpopulations in trained and untrained subjects. Int J Sports Med 1988;9:137– 40. Delgado I, Neubert R, Dudenhausen JW. Changes in white blood cells during parturition in mothers and newborn. Gynecol Obstet Invest 1994;38:227–35. Gibson A. On leucocyte changes during labor and the puerperium. J Obstet Gynecol Br Empire 1937;44:500 –9. Harris JW, Brown H. Bacterial content of the uterus at cesarean section. Am J Obstet Gynecol 1927;13:133. Gilstrap LC, Cunningham FG. The bacterial pathogenesis of infection following cesarean section. Obstet Gynecol 1979;53:545–9. Boggess KA, Watts DH, Hillier SL, Krohn MA, Benedetti TJ. Bacteremia shortly after placental separation during cesarean delivery. Obstet Gynecol 1996;779 – 84. Yancey MK, Clark P, Duff P. The frequency of glove contamination during cesarean delivery. Obstet Gynecol 1994;83:538 – 42. Cunningham FG, MacDonald PC, Gant NF, Leveno KJ, Gilstrap LC, Hankins GDV, et al, eds. Williams obstetrics. 20th ed. Stamford, Connecticut: Appleton & Lange, 1997:551. Peirce JC, Cornell RG. Integrating stratum-specific likelihood ratios with the analysis of ROC curves. Medical Decis Making 1993;13:141–51.
White Blood Cell Count
Katherine E. Hartmann, MD, PhD University of North Carolina, Chapel Hill North Carolina Program for Women’s Health Research Cecil G. Sheps Center for Health Services Research 725 Airport Road, CB #7590 Chapel Hill, NC 27599-7590 E-mail: [email protected]
Received December 17, 1999. Received in revised form March 14, 2000. Accepted April 7, 2000.
Copyright © 2000 by The American College of Obstetricians and Gynecologists. Published by Elsevier Science Inc.
Obstetrics & Gynecology