Early and late markers for the detection of early-onset neonatal sepsis

Material and methods

The data of the study was collected in the NICU of Aalborg Hospital, University of Aarhus, between September 1999 and September 2001. Altogether, 123 neonates were consecutively included. Inclusion criteria were: clinical signs of EONS with at least one significant symptom indicating possible EONS ( Table 1 ); age less than 72 hours at admission and a birth weight more than 1200 g. We recommended the weight limit of 1200 g based on our standards for the maximum sizes of individual blood samples related to the body mass. The study was approved by the local ethical committee. Both verbal and written informed consent was given by the parents.

The patients were retrospectively allocated into a sepsis and no sepsis group before the cytokine and PCT analyses were done:

1A

Sepsis. Infants with blood culture verified bacteraemia (four infants: one with Group B Streptococcus (GBS), two with Escherichia coli and one with Staphylococcus aureus ).

1B

Strongly suspected sepsis (25 infants). Significant symptoms and inflammatory response defined by CRP > 50 μ g/ml at any time point, in conformity with other studies on EONS (12-14). This strategy was chosen as it is known that many true bacteraemia are not found by the standard blood culture [15, 16].

2A

No sepsis (treated with antibiotics) (37 infants). This group included infants where antibiotic treatment was initiated because of clinical symptoms but the inflammatory response was low to moderate (CRP ≤ 50 mg/l). The antibiotic treatment was withdrawn after few days (median length of treatment was three days) because dynamics and severity of symptoms never confirmed clinical EONS. Antibiotic treated infants routinely started up with ampicillin (200 mg/kg/day) and gentamycin (3-4 mg/kg/day) intravenously.

2B

No sepsis (not treated) (57 infants). These infants were observed closely because of sepsis relevant symptoms. The severities of symptoms were not, though, considered relevant for initial treatment. The following dynamics of the paraclinic and clinical symptoms supported this strategy. All infants recovered without need of antibiotics.

Blood sampling was performed 0, 8, 16, 24, 48 and 72 hours after the neonate was suspected of EONS. At time 0 hours a blood culture and leukocyte count including differential counts were made. CRP was immediately analysed at all sampling times. At times 0, 8, 16 and 24 hours, 300 μ l plasma was separated for later PCT and cytokine analysis. The capillary blood sample was taken using heel prick. The blood was then transported on ice, centrifuged, and the plasma was frozen at -70 °C.

The leukocyte differential counts were done manually by microscope.

CRP was routinely analysed by fixed-point immuno-rate colorimetri (Vitros 950, Ortho-Clinical Diagnostics, Rochester, USA). The detection limit was 10 μ g/ml.

PCT was analysed using an immunoluminometric assay (LUMItest R PCT; BRAHMS Diagnostica, Berlin, Germany). The kit was used according to manufacturer´s recommendations. Detection limit was 0.08 ng/ml.

For the evaluation of IL-6, IL-8, IL-10, IL-18, IFN- γ and TNF- α , we used an in-house test based on the Luminex platform [17, 18]. The following antibody pairs were used: IL-6 (R&D Systems cat. nr. DY206 incl. standard), IL-8 (CLB Sanquin cat. nr. M9318. The standard was from NIBSC cat. nr. 89/520), IL-10 (CLB Sanquin cat. nr. M9310 incl. standard), IL-18 (Medical & Biological Laboratories Co. cat. nr. D044-3/D045-6 standard B001-5), IFN- γ (R&D Systems cat. nr. DY285 incl. standard) and TNF- α (R&D Systems cat. nr. DY210 incl. standard). 12.5 ml patient serum was used for the assay. Detection limits for the cytokines were: IL-6 4.6 pg/ml; IL-8 13.7 pg/ml; IL-10 13.7 pg/ml; IL-18 13.7 pg/ml; IFN- γ 7.8 pg/ml and TNF- α 13.7 pg/ml.

Statistical methods

In the evaluation of sensitivity and specificity of the different cut-off values, we compared the results in the sepsis group with the results found in the group with no sepsis.

Differences between groups were analysed with Kruskal-Wallis One-Way ANOVA on Ranks - not corrected for ties. Differences were considered significant at p< 0.05.

The statistical calculations were made by the NCSS statistical program.

Results

Demographic and clinical characteristics

Except for GA (π= 0.01), the two patient groups were not statistically significantly different from each other concerning background characteristics (p = 0.14-0.87 One-Way ANOVA on ranks) ( Table 2 ). The median gestational age was 38 weeks, the median birth weight 3170 g and the median age when included in the study was eight hours.

Symptoms

When the clinical symptoms at inclusion were categorized as given in Table 1, the clinical score per patient (maximum one point from each of the five groups of symptoms) was 2.1 (0.94) (mean (SD)) for the sepsis group and 1.56 (0.82) for those with no sepsis.

Dynamic analysis

The dynamics of the markers are illustrated in Table 3 ; 50, 90 and 100 percentiles are shown, as we find these values are the most interesting when high specificity is wanted.

IL-6, IL-8 and IL-10 median concentrations peaked at the time of the initial blood sampling. PCT and CRP median values in group 1 created a characteristic "dome" peaking at 8-16 hours and 16-24 hours, respectively.

The median concentrations of IL-6, IL-8, IL-10, I/T ratio and PCT (Table 3) were significantly different between the two groups at time 0, 8, 16 and 24 hours (p = 0-0.03 ), except the IL-8 values at 24 hours (p = 0.68) and Il-10 at 16 and 24 hours (p = 0.11 and 0.85).

IFN- γ only exceeded detection level in five patients and two of these patients were in the group of no sepsis (one treated and one not treated). TNF- α was measured above detection level in only seven patients, five from the sepsis group and two from the group of no sepsis (treated). For IL-18, IFN- γ and TNF- α there were no significant differences between the two groups (p = 0.10-0.66).

Cut-off values

Using receiver operating characteristic (ROC) curves, the area under the curve (AUC ) was greatest at time 0 hours for the cytokines IL-6 (0.77), IL-8 (0.70) and IL-10 (0,65). For PCT it was largest at 8 hours (0.83). AUC value for I/T ratio at time 0 was 0.81.

At time 0 hours we found cut-off values with specificities around 95% using ROC curves and tables. These cut-off values had rather low sensitivities ranging from 21 to 57% ( Table 4 A ). The highest sensitivity was seen for IL-6 and the second highest for I/T ratio.

Combination of early and late markers at time 0 hours

Combining IL-6 with PCT gave the highest sensitivity: sensitivity 71%, specificity 88% (I/T ratio combined with PCT was 59% and 88%) ( Table 5 A).

Alternative cut-off values often used in other studies

Cut-off values for each marker were calculated using the points closest to the upper left corner of the ROC curves (where sensitivity and specificity are approximately identical). This strategy is often applied in other studies [2, 19-21]. In our study, sensitivities then varied from 64 to 71% and specificities from 61 to 79% (Table 4B). When early markers were combined with PCT the sensitivity rose, but as expected, the specificity dropped drastically. The values for IL-6 combined with PCT now were 93% and 46% (Table 5B).

PCT

We found that PCT > 25 ng/ml was a fair late marker of EONS at blood sampling at time 8 hours (AUC 0.83, Likelihood Ratio 4, sensitivity 70% and specificity 86%), table not shown.

Individual marker curves

To illustrate the dynamics of the markers in verified EONS, the data from the four infants are shown in Figure 1 .

Discussion

We found that IL-6, IL-8 and IL-10 were reacting with maximum plasma concentrations at the time when the infants were suspected of having sepsis. PCT and CRP were late reacting with maximum concentrations at 8-16 and 16-24 hours, respectively. Especially we noted that when the cytokine values were decreasing, PCT and CRP had just started to increase.

When we looked at the high cut-off values, i.e. with specificities around 95%, of the single early markers at time 0 hour, IL-6 was the best marker followed by the I/T-ratio; however, the sensitivity was only 57% and 48%, respectively.

To improve the effectiveness of the tests we combined an early and a late marker. The combined value of IL-6 and PCT at the first blood sample was a fair EONS predictor (sensitivity 71% and specificity 88%). Observe that I/T ratio is almost as efficient as IL-6. By introduction of these markers it may be possible to reduce the diagnostic "non-conclusive" period of paraclinical values.

In the sepsis group 26 of 29 infants had significant paraclinic values for EONS at time 0. In the no sepsis group 21 of 94 infants (37 were treated in the study) had paraclinic values significant for EONS within the first 16 hours. So using paraclinic values alone for the evaluation of EONS, 16 infants less would need to be treated. This way the total days of antibiotic treatment in the no sepsis group might be reduced from 111 to 63 days.

We found that IL-18, IFN- γ and TNF- α were of no use in the diagnosis of the EONS. To our knowledge IL-18 has not previously been studied in infected infants. Lauw et al [22] found that IL-18 was raised in infected adults. Ng et al [18] found that IFN- γ was increased in septic neonates. In five out of six studies summarized by Mehr and Doyle [2] TNF- α was increased in infected infants, especially if they had septic shock. In our study very few infants had septic shock, which might be the reason for the difference between the results.

There are no rules concerning how to choose cut-off values from an ROC curve. The values at which sensitivity and specificity are equal are often used. Others prefer a high sensitivity to avoid missing a relevant patient [23]. In our study we chose cut-off values with high specificity (approximately 95%) for the detection of sepsis. But to be able to compare our study with others [2, 19-21] we calculated the diagnostic power when combining low cut-off values on the markers, i.e. the situation where sensitivity and specificity were almost equal. This strategy resulted in too many false positive results (specificity 44-60% in the group of no sepsis).

So if combinations of early and late markers are used at time 0 it is important to use high cut-off values. This way a restrictive antibiotic politic is supported. Remembering though, that when the clinical symptoms are considered serious, treatment should always be initiated without waiting for initial paraclinic values.

The strength of this study

1. We analysed the dynamics of the markers every eight hour in the first 24 hours. This way we could show that the values of the cytokines fell rapidly, while the late markers peaked 16-24 hours after the suspicion of EONS.

2. Therefore our high cut-off technique focusing on peak values of early and late markers is valuable. Screening for a high early or late marker minimises the risk of reacting on an insignificant raise in values caused by unspecific inflammation in the newborn.

3. Further, by using the LUMINEX flow cytometric technique, we could analyse many markers although we only had a small serum volume.

The weakness of the study

1. We only had very few infants with a positive blood culture.

2. Our paraclinic definition of sepsis was based on CRP values above 50 mg/l. Many studies have used other values and parameters [2, 19-21] but unfortunately no consensus exists on this subject. We find our method acceptable. Especially as the symptom scores on every infant has been evaluated by a senior neonatologist before the analyses of the new markers were made.

3. Markers related to CRP are biased to better results, as CRP are part of our definition of sepsis [10, 24].

4. We have not included the most vulnerable infants with a birth weight less than 1200 g.

5. The markers from the present study and the interpretation resulting from the chosen cut-off values ought to be tested in a prospective clinical study.

Conclusion

PCT combined with IL-6 (or I/T ratio) is fair for detecting EONS at the first blood sample. By the introduction of IL-6 and PCT it may be possible to reduce the number of infants treated with antibiotics by reducing the diagnostic "non-conclusive" period of the paraclinical values.

Acknowledgements

We thank our medical staff for their help in collecting the data, the laboratory staff of the Department of Clinical Biochemistry, Aalborg Hospital, the laboratory assistant Eva Poulsen, Department of Clinical Immunology, Aalborg Hospital, for setting up new tests for this study, and finally Henrik Toft Sørensen and Gunnar Lauge Nielsen, Department of Epidemiology, Aarhus University Hospital, for providing ideas about how to analyse the results.

This work was supported by grants from Aalborg Frivillige Bloddonoreres Fond, Den Obelske familiefond and Nordjyllands Lægekredsforenings Forskningsfond. The PCT kit was sponsored by BRAHMS Diagnostica, Berlin, Germany.