Introduction

Sepsis is an important cause of morbidity and mortality for neonates [1, 2], especially in the developing countries [3]. Rapid and accurate diagnosis of neonatal sepsis is often difficult in routine clinical practice because the clinical manifestations of this condition can overlap with those of non-infectious conditions, such as the meconium aspiration syndrome, respiratory distress syndrome, and hemodynamic instability of various underlying etiologies. Microbiological cultures aid in the identification of serious bacterial infection, but often yield false-negative results, particularly after maternal antibiotic use [4, 5], and might also yield false-positive results because of specimen contamination.

The use of several biochemical markers has been studied with the aim to improve the clinical management of neonates with suspected bacterial infection [6–8]. Yet, no single laboratory test is considered to reliably predict neonatal sepsis at the time of initial presentation. Thus, neonates with clinical manifestations of sepsis or with risk factors for serious bacterial infection are commonly treated empirically with antibiotics, awaiting the results of microbiological and other investigations [9]. This inevitably leads to overuse of antibiotics, which in turn can pose a selection pressure for multidrug-resistant bacteria in the neonatal intensive care unit [10].

Serum procalcitonin (PCT) is a biological marker of increasing interest for detecting serious bacterial infections [11], including sepsis, in adults [12], or pediatric patients and newborns [13, 14]. However, regarding neonates particularly, a physiological postnatal increase of serum PCT occurs in healthy term and preterm neonates, with peak values at 24 h of age [15–17]. Taking all the above into consideration, we aimed to assess the value of PCT for the diagnosis of neonatal sepsis by performing a diagnostic test accuracy meta-analysis of relevant studies.

Methods

Data sources

We systematically reviewed PubMed, Scopus, and the Cochrane Library databases up to 8 June 2009. The PubMed combined search term used was: (procalcitonin OR PCT) AND (neonatal sepsis OR neonatal infections OR sepsis). The search terms applied to the Scopus and the Cochrane Library were “procalcitonin and sepsis” and “procalcitonin,” respectively. The bibliographies of relevant articles were also hand-searched.

Study selection criteria

A study was considered eligible for inclusion in our review if it provided data on serum PCT for neonates with and without sepsis (either microbiologically or clinically documented). In addition, PCT blood measurement had to be performed at the time of clinical presentation with suspected sepsis before the administration of antimicrobial therapy or for asymptomatic neonates at the time of inclusion in the study. We excluded studies that used PCT measurements that were made only on maternal or umbilical cord blood samples. Conference abstracts or studies written in languages other than English, Spanish, French, German, Italian, and Greek were also excluded.

Data extraction

Data extracted from each of the included studies referred to the type of the study design, the size and characteristics of the study population, the number of patients with early/late onset of sepsis, as well as the number of patients >28 days old that were included, and the number and specific characteristics of the patients in the septic and non-septic groups. Specific data regarding the cutoff level of serum PCT evaluated, the sensitivity/specificity, and the positive/negative predictive value (PPV/NPV) of PCT for the diagnosis of neonatal sepsis were also extracted.

In cases in which major discrepancies between the data reported in the included studies and the data calculated were observed, we contacted the first or last authors of the individual studies via e-mail, requesting clarification regarding the raw data of the studied patient groups.

Definitions

Patients included in the septic group had either microbiologically (culture-proven) or clinically diagnosed sepsis, whereas patients included in the non-septic group were patients for whom the diagnosis of sepsis was excluded based on the microbiological/clinical symptoms and signs and/or if they had a benign clinical course. We considered that neonates who presented with a clinical suspicion of sepsis and required antibiotic therapy for no more than 3 days had a negative diagnosis for sepsis, in the case that this was reported in the included studies. In addition, neonatal sepsis was considered as early onset (EOS) if it was diagnosed in the first 72 h of life and late-onset (LOS) if it was diagnosed after this period.

Data analysis

In our primary analysis we included all studies that evaluated PCT in neonates with microbiologically or clinically diagnosed sepsis in comparison with ill neonates that had other conditions. Studies in which the control group consisted of healthy neonates as well as studies in which pediatric patients >28 days old constituted >25% of the total population were also excluded. Studies for which the accurate number of patients with true-/false-positive and true-/false-negative PCT results could not be calculated, such as those that only provided data for PCT sensitivity and specificity that were derived from a computed ROC curve, were also excluded from the analysis. In order to evaluate the performance of PCT for the diagnosis of early onset and late-onset neonatal sepsis separately, we performed two sub-analyses limited to studies that involved exclusively or in the majority (>85%) neonates with early onset and late-onset sepsis, respectively.

We performed a diagnostic test meta-analysis using a bivariate meta-analysis model [18] to calculate the pooled sensitivity, specificity, positive/negative likelihood ratios, as well as the diagnostic odds ratio. We also constructed the respective hierarchical summary receiver-operating characteristic (HSROC) curve that plots sensitivity versus specificity and calculated the area under the curve (AUC) [19]. Moreover, the Spearman correlation coefficient between the logits of sensitivity and specificity was used to evaluate the presence of a threshold effect in the accuracy of PCT. The presence of statistical between-study heterogeneity was assessed by the I 2 test [20]. Values of 25, 50, and 75% for the I 2 test were regarded as indicative of low, moderate, and high statistical heterogeneity, respectively. All the above analyses were performed using the Midas Module in Stata software version 10 [21, 22].

Quality assessment of the included studies

The methodological quality of the analyzed studies was assessed using the QUADAS tool [23]. Nine of the 14 items of the QUADAS tool were considered relevant for the studies included in our review. These were: representative spectrum, clear description of study selection criteria, acceptable reference standard, avoidance of partial/differential verification and incorporation biases, detailed description of index test and reference standard, and explanation of study withdrawals.

We considered the spectrum of the patients to be representative of the target population if all the evaluated patients were neonates (0–28 days) and had critical illness and/or clinical manifestations consistent with possible sepsis. If healthy neonates were included in the non-septic group, the study population was considered as non-representative. The acceptable reference standard consisted of diagnostic criteria for neonatal sepsis matching those presented in the International Pediatric Sepsis Consensus conference [24]. Partial and differential verification biases were considered to have been avoided if all the included children were evaluated with the same reference standard method used in each study, regardless of the PCT results. All calculations and analyses, including the methodological quality analysis, were performed with the use of the Review Manager (RevMan) v. 5.0 Software (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2008).

Results

Study selection process

In Fig. 1, we present the flow diagram showing the process of selection of the studies included in our review. Specifically, our search in PubMed generated 623 potentially relevant articles, of which 28 were eligible for inclusion in our review. The search performed in the Cochrane Library yielded no additional articles eligible for inclusion, whereas one additional study that qualified for inclusion was identified from the search of the Scopus database. No additional eligible study was identified from hand searching of the bibliographies of relevant articles. Overall, 29 individual articles were eligible for inclusion in our review [25–53].

Fig. 1
figure 1

Flow diagram of the detailed process of selection of articles for inclusion in our review

Full size image

Characteristics of the included studies

In Table 1, we present the main characteristics of the 29 included studies. Four studies had a retrospective design [27, 30–32], whereas the remaining were prospective cohort or case control studies. Seven studies exclusively involved patients with early onset neonatal sepsis [4, 27, 31, 40, 47, 49, 52], and six studies exclusively involved patients with late-onset neonatal sepsis [26, 33, 34, 42, 44]. In 4 of the 29 included studies, the comparator group consisted of patients that potentially had sepsis, on the basis of the relevant diagnostic criteria provided, and thus we did not include them in the analysis [29, 30, 41, 48]. With regard to the remaining 25 studies comparing patients with neonatal sepsis to patients without neonatal sepsis, 6 did not provide accurate PCT patient diagnostic data [37, 39, 42, 49, 50, 52], 2 involved healthy subjects as controls [33, 36], and 1 did not provide accurate PCT patient diagnostic data and also involved pediatric patients >28 days old in a percentage >25% [43]. The above-mentioned nine studies were also excluded from the analyses. In Table 2, we present the data derived from each of the 16 analyzed studies regarding the value of serum PCT for the diagnosis of neonatal sepsis. In 13 of these studies, the septic group consisted of neonates with both culture-proven sepsis and clinically diagnosed sepsis [25–27, 31, 34, 35, 38, 40, 44–47, 51], and in the remaining 3 with culture-proven sepsis alone [28, 32, 53]. The non-septic group consisted of ill neonates with other conditions that were hospitalized in the pediatric ICU in all but three of these studies [34, 51, 53].

Table 1 Main characteristics of the studies included in the meta-analysis
Full size table
Table 2 Data derived from the studies included in the analysis regarding patients with sepsis (bacteriologically or clinically documented) versus patients without sepsis
Full size table

Methodological quality of the included studies

In Fig. 2, we summarize the results of the methodological assessment for the total of the 29 studies included in the meta-analysis.

Fig. 2
figure 2

Methodological quality graph depicting the cumulative findings of the methodological quality analysis of the studies included in the meta-analysis

Full size image

Diagnostic accuracy of PCT

Sixteen studies, involving a total of 1,959 neonates, were included in our primary analysis [25–28, 31, 32, 34, 35, 38, 40, 44–47, 51, 53]. The pooled (95% CI) sensitivity of PCT for the diagnosis of neonatal sepsis was 81% (74–87%), and the specificity was 79% (69–87%) (Fig. 3). The pooled (95% CI) diagnostic odds ratio was 16 (8–32), whereas the pooled (95% CI) positive and negative likelihood ratios were 3.9 (2.5–6.0) and 0.24 (0.17–0.34), respectively. The area under the HSROC curve (95% CI) for PCT was 0.87 (0.84–0.90) (Fig. 4). The I 2 index (95% CI) was 96% (92–99%). The effect of the diagnostic threshold (cutoff value) was not found to be important, since a weak negative correlation between the logits of sensitivity and specificity was observed (Spearman correlation coefficient = −0.08).

Fig. 3
figure 3

Forest plot of pooled sensitivity and specificity of BDG for the diagnosis of neonatal sepsis. The point estimates and the respective 95% confidence intervals for each one of the included studies are represented by the circles in the squares and the horizontal lines. The point estimate is represented by the dotted line, whereas the 95% CIs are represented by the diamond shape

Full size image
Fig. 4
figure 4

Hierarchical summary receiver-operating characteristic curve of the sensitivity versus specificity of PCT for the diagnosis of neonatal sepsis. The curve is represented by the straight line; each of the analyzed studies is represented by a circle; the point estimate to which summary sensitivity and specificity correspond is represented by the diamond shape and the respective 95% confidence intervals by the dashed line, whereas the 95% confidence area in which a new study will be located is represented by the dotted line

Full size image

Sub-analyses

Diagnostic accuracy of PCT for the diagnosis of EOS

Six studies, involving a total of 780 neonates, were included in the analysis regarding EOS [15, 27, 31, 40, 45, 47]. The pooled (95% CI) sensitivity of PCT for the diagnosis of EOS was 76% (68–82%), and the specificity was 76% (60–87%). The pooled (95% CI) diagnostic odds ratio was 10 (5–22), whereas the pooled (95% CI) positive and negative likelihood ratios were 3.2 (1.8–5.7) and 0.32 (0.23–0.43), respectively. The area under the HSROC curve (95% CI) for PCT was 0.78 (0.74–0.81) (Fig. 5). The I 2 index (95% CI) was 89% (77–100%).

Diagnostic accuracy of PCT for the diagnosis of LOS

Five studies, involving a total of 535 neonates, were included in the analysis regarding LOS [25, 26, 34, 35, 44]. The pooled (95% CI) sensitivity of PCT for the diagnosis of LOS was 90% (73–97%), and the specificity was 88% (72–96%). The pooled (95% CI) diagnostic odds ratio was 67 (23–200), whereas the pooled (95% CI) positive and negative likelihood ratios were 7.7 (3.1–18.9) and 0.11 (0.04–0.31), respectively. The area under the HSROC curve (95% CI) for PCT was 0.95 (0.93–0.97) (Fig. 6). The I 2 index (95% CI) was 93% (86–99%).

Fig. 5
figure 5

Hierarchical summary receiver-operating characteristic curve of the sensitivity versus specificity of PCT for the diagnosis of early onset neonatal sepsis. The curve is represented by the straight line; each of the analyzed studies is represented by a circle; the point estimate that summary sensitivity and specificity correspond to is represented by the diamond shape and the respective 95% confidence intervals by the dashed line, whereas the 95% confidence area in which a new study will be located is represented by the dotted line

Full size image

Discussion

The main finding of our meta-analysis is that PCT has very good diagnostic accuracy for the diagnosis of neonatal sepsis. Specifically, in our primary analysis involving all studies evaluating PCT in neonates with and without sepsis, the area under the HSROC curve was 0.87, and the pooled sensitivity and specificity were 81 and 79%, respectively. Additionally, the area under the curve for the analyses regarding early-onset and late-onset sepsis was 0.78 and 0.95, respectively. However, marked statistical heterogeneity was present in all analyses, a fact that must not be overlooked in the interpretation of the above findings.

Fig. 6
figure 6

Hierarchical summary receiver-operating characteristic curve of the sensitivity versus specificity of PCT for the diagnosis of late-onset neonatal sepsis. The curve is represented by the straight line; each of the analyzed studies is represented by a circle; the point estimate to which summary sensitivity and specificity correspond is represented by the diamond shape and the respective 95% confidence intervals by the dashed line, whereas the 95% confidence area in which a new study will be located is represented by the dotted line

Full size image

We should note that an important advantage of any biological marker used for neonates with suspected sepsis would be to correctly identify the septic episodes that are culture-negative and require antibiotic therapy. Ruling out sepsis is important, as if the number of neonates treated with antibiotics can be minimized, the length of hospitalization can be shortened; there may be less selection pressure for the emergence of resistant organisms, with medical and financial advantages that could offset the financial costs of measuring PCT [54]. Indeed, recent randomized controlled studies (RCTs) have suggested that PCT-guided algorithms are associated with a reduction in antibiotic exposure and antibiotic treatment duration [10, 55]. However, as observed with other diagnostic tests, including C-reactive protein (CRP) and total leukocyte count [56, 57], it cannot correctly identify 100% of the septic neonates by itself. Thus, relying on this biomarker has the risk of withholding antibiotic therapy in septic neonates that could otherwise benefit from such potentially life-saving therapy. The use of a lower cutoff value of serum PCT could theoretically increase the sensitivity and negative predictive value of this test for the diagnosis of neonatal sepsis [58].

The considerable heterogeneity regarding the definition of neonatal sepsis observed among the studies included in our review illustrates the lack of a universally acceptable definition of neonatal sepsis, particularly for the clinically septic but culture-negative newborns [9]. Although in neonatology the concept of clinical sepsis is widely used and considerable attempts have been made [24], a uniform definition for this common diagnosis is still lacking. This can be a cause of variability in the criteria for the definition of neonatal sepsis used in the studies that evaluate clinically diagnosed sepsis. Thus, in all likelihood, the spectrum of disorders and disease severity encompassed under the term neonatal sepsis differed among the various studies included in this review, a fact that may potentially account for the considerably high statistical heterogeneity observed in our analyses.

The different cutoff values of PCT incorporated in the analyzed studies were not found to account for a considerable proportion (threshold effect) of the observed statistical heterogeneity. Another potential source of heterogeneity may be the age of the involved pediatric patients. In order to address this issue, we excluded from our analysis the studies that involved a substantial proportion (>25%) of pediatric patients older than 28 days. Moreover, we excluded studies that involved healthy neonates as controls, as they cannot be regarded as representative of the population to whom PCT will be applied in routine clinical practice. The inclusion of premature neonates in the evaluated studies may also be another source of heterogeneity. However, since data regarding the percentage of preterm neonates among the involved patients were scarcely reported in the included studies, we could not assess the effect of this specific factor on the performance of PCT regarding the diagnosis of neonatal sepsis.

Taking into consideration the physiological postnatal increase of serum PCT concentration that is observed in healthy preterm neonates [17], as well as in healthy term neonates [15], with peak values at 24 h of age [16], we performed two sub-analyses limited to studies evaluating the performance of PCT for the diagnosis of early (<72 h) and late-onset (>72 h) neonatal sepsis, respectively. Although our findings suggest that PCT has better diagnostic accuracy for late-onset compared with early onset neonatal sepsis, the available data for late-onset sepsis were not sufficient to allow any firm conclusions.

Several limitations should be taken into consideration in the interpretation of the findings of this meta-analysis, particularly the heterogeneity between the included studies regarding the characteristics of the enrolled neonates (particularly the postnatal age), as well as the broad definition criteria of neonatal sepsis. Until a uniform definition of neonatal sepsis is available, this important limitation will continue to be inherent in the research in this field. Moreover, a considerable proportion of neonates included in the septic group had possible (not microbiologically documented) sepsis. Indeed, possible neonatal sepsis is a diagnosis frequently encountered in routine clinical practice. PCT may aid in the classification of these patients as septic or non-septic. However, since specific data regarding the diagnostic performance of PCT for this sub-group of patients were scarcely reported from the included studies, we did not assess the potential influence of this factor on our study findings. Finally, it is possible that PCT may perform differently in neonatal sepsis because of gram-positive, gram-negative, or fungal pathogens [59]. Hence, not only the clinical characteristics of the enrolled neonates, but also the local microbiological profile in a given ICU are likely to affect the value of PCT in predicting neonatal sepsis. However, we were unable to explore this further because the necessary information was usually unavailable in the studies included in this review.

Conclusion

In conclusion, our findings suggest that serum PCT, measured at the time of clinical presentation, seems to have very good diagnostic accuracy for the discrimination between ill neonates with sepsis and those with other conditions. However, the considerable differences between the analyzed studies, as well as the lack of a uniform definition of neonatal sepsis that can be used as a reference diagnostic standard, cannot allow the establishment of any firm conclusions. Larger studies using appropriate methodology are required to validate the routine use of PCT as a diagnostic marker of neonatal sepsis.