Objective. The evaluation of an infant for suspected sepsis often includes obtaining blood for laboratory tests. The shortcomings of the current practice are that the infant has to appear clinically ill for the diagnosis to be entertained, and the conventional laboratory tests are invasive. We have found that the clinical diagnosis of neonatal sepsis is preceded by abnormal heart rate characteristics (HRC) of reduced variability and transient decelerations, and we have devised a predictive HRC monitoring strategy based on multivariable logistic regression analysis that was developed at one tertiary care NICU and validated at another. We hypothesized that HRC monitoring, which is continuous and noninvasive, might be an adjunct to conventional laboratory tests in the diagnosis of neonatal sepsis. The objective of this study was to test the hypothesis that HRC monitoring adds information to conventional laboratory tests in diagnosing neonatal sepsis.
Methods. We prospectively collected heart rate data in 678 consecutive infants who stayed >7 days in the University of Virginia NICU from July 1999 to July 2003. We prospectively measured HRC and noted 149 episodes of sepsis with positive blood cultures for which data were available in 137. We obtained all laboratory test results for ratio of immature to total neutrophil count, white blood cell count, glucose, platelet count, HCO3, arterial partial pressure of carbon dioxide, and pH. We tested hypotheses using multivariable logistic regression modeling adjusted for repeated measures.
Results. We found that the HRC index, which was available 92% of the time, was highly significantly associated with sepsis (receiver-operating characteristic [ROC] area: 0.73). The ratio of immature to total neutrophil count, white blood cell count (available 4%–8% of the time, usually around the time of suspected sepsis), and blood glucose and pH (available 28% and 38% of the time) were also significantly associated with sepsis (ROC area: 0.75). HRC and laboratory values added independent information to each other, and a predictive model using all significant variables had ROC area of 0.82.
Conclusions. HRC monitoring adds independent information to laboratory tests in the diagnosis of culture-positive neonatal sepsis.
Infants in the NICU are at risk for sepsis, and those at highest risk are those who are born with very low birth weight (VLBW; <1500 g). Approximately 56000 VLBW infants are born in the United States each year.1 Survival of this group has improved with advances in neonatal intensive care, but late-onset sepsis continues to be a major cause of morbidity and mortality. In a large study, the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network found a 2.5-fold increase in mortality and >30% increase in hospital stay in the 21% of VLBW infants with culture-proven sepsis.2 It was concluded that strategies to reduce the incidence and severity of neonatal sepsis are “needed urgently.” One such strategy might be continuous noninvasive physiologic monitoring of NICU patients optimized to detect early signs of illness.
Diagnosis of neonatal sepsis is difficult because the clinical signs are subtle and nonspecific and laboratory tests including the “gold standard” blood culture are not always reliable.3–6 For example, Fanaroff et al.7 in the NICHD Neonatal Research Network found a variety of associated clinical and laboratory findings. Many were nonspecific, and none had high predictive accuracy. This can be attributed to the complexity and the variability of the host response to infection, which has been called the systemic inflammatory response syndrome (SIRS).8,9
Before clinical signs of sepsis, neonates have reduced heart rate variability (HRV) and transient decelerations similar to the findings in distressed fetuses.10 Heart rate (HR) measures that have been optimized to detect these abnormal heart rate characteristics (HRC) have been used to develop a predictive model for sepsis that added information to gestational age, birth weight, and days of age.11 The model was developed at one NICU and validated at another.11 The predictive model results, which we call the HRC index, also identify infants who are at higher risk for in-hospital death.12 Previously, we found that infants with clinical findings of SIRS and either positive or negative blood cultures had identical abnormal HRC.10 Certainly blood cultures are an imperfect gold standard for sepsis, and there are other causes of SIRS.
Our goal in this work was to test the hypothesis that the HRC index adds statistically significant information in the diagnosis of neonatal sepsis to the clinical laboratory values that are already available to physicians. Specifically, we asked whether there was statistically significant improvement in the accuracy of the regression analysis when HRC index was combined with laboratory values as indicators of sepsis.
We studied all admissions to the University of Virginia NICU from July 1999 to July 2003 that were 7 or more days of age. The clinical research protocol was approved by the Human Investigations Committee of the University of Virginia. Health care personnel were not aware of the result of the HRC monitoring before September 2003.
Each hospital course was divided into 6-hour blocks beginning at midnight, a total of 71478 for all of the patients. For each block, we recorded (1) blood cultures obtained for the clinical suspicion of sepsis, (2) laboratory test results, and (3) the HRC index.
We defined sepsis to be present when a physician suspected the diagnosis, obtained a blood culture that grew bacteria not ordinarily considered to be a contaminant, and initiated antibiotic therapy of 5 or more days' duration. This definition is consistent with the diagnosis of “proven sepsis” offered by the Centers for Disease Control and Prevention.13 For infections with coagulase-negative Staphylococcus, our definition is consistent with the NICHD Neonatal Research Network definition of “possible” coagulase-negative Staphylococcus sepsis.2
Laboratory Test Results
We obtained all laboratory values for total white blood cell (WBC) count (n = 4243), ratio of immature to total neutrophil count (I/T ratio; n = 3382), platelet count (n = 6286), glucose (n = 21277), HCO3 (n = 7751), arterial partial pressure of carbon dioxide (Pco2), and pH (n = 28714) from an electronic clinical archive. An abnormal laboratory result was considered to be present when there was leucocytosis or leucopenia (WBC count >20000/μL or <5000/μL), high proportion of circulating immature neutrophils (I/T ratio >0.2), thrombocytopenia (platelet count <100000/μL), hyperglycemia (blood glucose >180 mg/dL), hypercarbia (Pco2 >70 mm Hg), low bicarbonate (HCO3 <15 mg/dL), or low pH (pH <7.2) within the previous 6 hours. Results were similar when laboratory tests were considered as continuous variables (with the important exception of WBC count, as noted below) or when thresholds for abnormal were varied.
We prospectively measured the HRC index every 6 hours (n = 71478).11,12 This is calculated from the HRs and reflects the degree to which reduced variability and transient decelerations are present. These characteristic abnormalities are present in the subclinical phase of some cases of neonatal sepsis and are not part of the spectrum of normal HRV. They can be detected using the SD, sample asymmetry14 (a measure of the symmetry of HR histograms), and sample entropy15,16 (a measure of regularity that falls when both reduced variability and transient decelerations are present) of 4096-beat epochs combined in a logistic regression model that was developed at one NICU and validated at another.11
The HRC index is a continuous measure. To facilitate clinical application of the HRC index, we defined 3 levels of risk on the basis of the model values. The lowest 70% of HRC measurements were less than or equal to the mean value and thus corresponded to a relative risk of 1 or less for acute neonatal illness in the next 24 hours. We categorized these as low risk. Ninety percent of values were less than or equal to twice the mean value, so values between the 70th and 90th percentiles correspond to 1- to 2-fold increase in relative risk, and we categorized these as intermediate risk. HRC values above the 90th percentile correspond to 2-fold and higher risk, and we categorized these as high risk. HRC values were available for 137 episodes of sepsis.
We tested hypotheses and calculated odds ratios using multivariable logistic regression modeling adjusted for repeated measures with a Wald χ2 test for whether predictive variables added independent information to one another.11
We studied 678 infants, 320 of whom were VLBW. The study database contained 49 infant-years of HRC monitoring and laboratory test results. Table 1 gives the demographic characteristics and incidence of sepsis in this study. As expected, the majority of episodes were in VLBW infants, in whom 29% had 1 or more episodes of sepsis.
Availability of HRC Monitoring and Laboratory Tests
As shown in Fig 1, HRC monitoring was available 92% of the time. The remaining time could be accounted for by absence from the NICU, poor-quality electrocardiogram waveform, or skin fragility that precluded use of electrode patches. By our definition, the result was high risk 10% of that time. When the duration of a laboratory test result was 6 hours, laboratory tests were available 4% to 38% of the time and were abnormal up to 3% of the time.
Some laboratory tests were obtained more frequently near the time of sepsis, as shown in Fig 2. The upper line in each panel is the proportion of episodes of sepsis for which laboratory test results were available. The filled regions are the proportion of episodes for which the test was abnormal. Blood cultures obtained for suspicion of sepsis were drawn at time 0. The HRC index (Fig 2A) was available most of the time, and there was a gradual increase before the illness, as we previously found.10,11 Within 6 hours of the positive blood culture, 42% of readings were in the high-risk range and another 30% were in the intermediate-risk range. Complete blood counts (CBCs) with differential analysis were more commonly performed when sepsis was suspected (Fig 2B–D). As a result, most I/T ratio measurements were made within 6 hours of the blood culture, and 44% were abnormal. The other laboratory tests that we analyzed were obtained more uniformly (Fig 2E–H), with a concentration of abnormal glucose and pH values near the time of sepsis.
HRC Monitoring and Laboratory Tests Near the Time of Neonatal Sepsis
Inspection of Fig 2B–D shows an abrupt increase in the number of CBCs obtained in the 6-hour block during which the blood culture for suspicion of sepsis was done and in the 6-hour block preceding it. The interpretation is that physicians obtained CBCs and, either simultaneously or after seeing the result or additional clinical changes, blood cultures when infants began to display signs of illness. To compare the HRC index and laboratory tests in this scenario, we analyzed a 24-hour test window that included these two 6-hour blocks and the two 6-hour blocks that followed.
The significance of changes in the HRC index and laboratory results was tested using regression analysis, and representative results are shown as odds ratios in Fig 3. The inset above is a timeline highlighting the time window for which HRC and laboratory tests were assessed for association with sepsis. After accounting for other variables, there was very highly significant association of the I/T ratio, high-risk HRC index, and abnormal WBC count with the diagnosis of sepsis. High glucose and low pH were also significantly associated with sepsis, but high Pco2 and low HCO3 were not, even when the thresholds for abnormal were varied. It is interesting that any I/T ratio result was highly associated with sepsis, including 0, and the cutoff value of 0.2 did not help to enhance the association. Thus, the clinician's decision to order a CBC was associated with sepsis, but the I/T ratio was not.
To test the idea that HRC monitoring might add a dimension of information because it does not require the physician decision to order a test, we repeated the analysis for an earlier 24-hour test window ending before the 6-hour block in which the blood culture for suspicion of sepsis was obtained. This is the time window used in developing and validating the HRC index,11 where we purposely neglected data from the 6-hour block that contained the blood culture itself. This time window is shown in the inset at the top of Fig 4 and overlaps that of the previous analysis by 6 hours, during which approximately one third of the laboratory tests were obtained. As shown in Fig 4, the odds ratio for HRC index was unchanged, and all others fell. Only I/T ratio and low pH remained significantly associated with an upcoming (+) blood culture.
HRC and Laboratory Tests Add Independent Information
Table 2 shows representative results of logistic regression modeling to test hypotheses about the association of laboratory test results and HRC with sepsis. Variables were added singly or in groups to test for the presence of independent information, and abnormal tests were significantly associated with sepsis (model 1). High-risk HRC index was also significantly associated with sepsis (model 2) and added independent information to laboratory tests (model 3). Models that incorporated both laboratory tests and HRC had very good performance, with the receiver-operating characteristic area exceeding 0.82. In other words, a model prediction chosen at random from a time near sepsis would exceed any other randomly chosen result 82% of the time. Analysis of laboratory results as continuous variables gave similar results with the exception of WBC count, which did not have a significant association with sepsis. We attribute this to the fact that abnormally low or high WBC values may be associated with illness, and thus there is no simple linear relationship between WBC count and risk of sepsis.
We studied HRC and laboratory tests in the diagnosis and prediction of neonatal sepsis in a tertiary care NICU. We confirmed the findings of Fanaroff et al in the NICHD Neonatal Research Network that I/T ratio, abnormal WBC count, hyperglycemia, and low pH were significantly associated with neonatal sepsis.7 A major new finding was that the HRC index, which is available from continuous and noninvasive monitoring, added independent information to conventional laboratory tests.
Our study differs from that of Fanaroff et al in the time windows of test results that were analyzed. The earlier study analyzed laboratory test results in a single 48-hour time window from 24 hours before to 24 hours after the positive blood culture. We analyzed two 24-hour windows and found that laboratory test results other than pH had better performance when obtained in the 24-hour window after and shortly before the positive blood culture. Their performance fell in the 24-hour window leading up to but not including the time of the positive blood culture. It is interesting that the performance of the HRC index was the same in both time windows.
In this observational study, physicians obtained blood tests when they were clinically indicated. CBCs with differential analysis were most commonly obtained to evaluate an infant who was ill with suspected sepsis, so I/T ratio results were more likely to be available near the time of illness. Indeed, we found that the presence of an I/T ratio result had as much information as its value, or whether it exceeded the usual cutoff of 0.2. A more comprehensive comparison would require round-the-clock blood tests, an impractical goal. Another limitation of our study is the absence of C-reactive protein and newer tests for neonatal sepsis, which are likely also to add information in the early diagnosis5 but are not in routine use at our hospital.
Is the HRC Index a Useful Tool for Assessing Risk for Neonatal Sepsis?
Screening tests for infrequent events have long been important diagnostic tools, but most positive results are false positives. Although many infants have episodes of sepsis, the illness is not frequent: we found 149 episodes of sepsis in 49 infant years, or ∼1 episode per infant every 4 months. This compares to the results recently reported by Stoll et al2 in the NICHD Neonatal Research Network, who found a rate of 1 positive blood culture event per 10 months for all VLBW infants. The highest rate was 1 infection per 6 months in the lowest birth weight group, 400 to 500 g. Because the HRC index was calculated every 6 hours and we expect HRC abnormalities for 12 to 24 hours before clinical diagnosis, we expect up to 4 “positive” tests in every several hundred measurements, an event rate of <1%. Viewed in this way, the clinical illness of neonatal sepsis occurs infrequently.
Thus, HRC monitoring is a screening test for an infrequent disease, and we expect that not all abnormal readings inevitably indicate imminent sepsis or other untoward events. The spirit of its use should be that abnormal readings indicate higher-than-expected risk of such events. In this way, HRC monitoring is similar to measuring troponin levels in adults with chest pain or C-reactive protein levels or calcium scores from electron beam computed tomography in adults with risk factors for coronary artery disease, prostate-specific antibody in men, or total-body magnetic resonance imaging in the asymptomatic population.
This study does not address the important question of whether a blood culture that is drawn in a septic infant before clinical signs solely because of increased HRC index would also be positive. That is the subject of future clinical studies.
Support for this study was provided by the American Heart Association, Mid-Atlantic Research Affiliate; University of Virginia Children's Medical Center Research Fund; Virginia's Center for Innovative Technology; and NIGMS-64640 from the National Institute of General Medical Sciences.
- Accepted August 30, 2004.
- Reprint requests to (M.P.G.) Department of Pediatrics, Box 800386, University of Virginia, Charlottesville, VA 22908. E-mail:
Conflict of interest: Medical Predictive Science Corp (Charlottesville, VA) has a license to market technology related to heart rate characteristics monitoring of newborn infants and supplied partial funding for this study. After completion of this study, Drs Griffin and Moorman acquired an equity share in this company.
This work was presented in part at the annual meeting of the Pediatric Academic Societies; May 1–4, 2004; San Francisco, CA.
- ↵Hamilton BE, Martin JA, Sutton PD. Births: preliminary data for 2002. Natl Vital Stat Rep.2003;51 :1– 20
- ↵Stoll BJ, Hansen N, Fanaroff AA, et al. Late-onset sepsis in very low birth weight neonates: the experience of the NICHD Neonatal Research Network. Pediatrics.2002;110(suppl) :285– 291
- ↵Fanaroff AA, Korones SB, Wright LL, et al. Incidence, presenting features, risk factors and significance of late onset septicemia in very low birth weight infants. The National Institute of Child Health and Human Development Neonatal Research Network. Pediatr Infect Dis J.1998;17 :593– 598
- ↵Griffin MP, Moorman JR. Toward the early diagnosis of neonatal sepsis and sepsis-like illness using novel heart rate analysis. Pediatrics.2001;107 :97– 104
- ↵Lake DE, Richman JS, Griffin MP, Moorman JR. Sample entropy analysis of neonatal heart rate variability. Am J Physiol.2002;283 :R789– R797
- ↵Richman JS, Moorman JR. Physiological time series analysis using approximate entropy and sample entropy. Am J Physiol.2000;278 :H2039– H2049
- Copyright © 2005 by the American Academy of Pediatrics