Introduction

Healthcare-associated sepsis (HAS) is the most frequent infectious complication in neonatal intensive care units (NICUs) with substantial risk for neurological sequelae and mortality [13, 28, 30, 31, 33]. The blood culture is the gold standard for sepsis diagnosis and is one of the most commonly executed diagnostic procedure in infants admitted to a NICU. The majority of the cultures are sampled on the day of birth and approximately 40% are sampled in suspicion of HAS [4]. Guidelines on clinical indications for blood culturing in neonates suspected for HAS are lacking so various thresholds for sampling are assumed. Moreover, due to the potential severity and lethality of HAS, a low threshold for sampling and antimicrobial therapy (ABT) initiation is accepted [32]. In addition, blood culturing is time-consuming and imprecise so early and correct diagnosis of HAS is challenging and mostly based on judgment of clinical signs and inflammatory biomarkers [6, 25, 26]. Much attention is for the predictive value of biomarkers such as C-reactive protein (CRP), procalcitonin, and anti-inflammatory cytokines. However, those biomarkers can be of noninfectious origin and may not have the ability to differentiate between several inflammation sources [2, 21, 22, 37]. Likewise for individual clinical signs which are mostly subtle and difficult to interpret, predictive value is questionable [34].

We hypothesize that blood culture sampling practice differs between NICUs because of the lack of guidelines. For this, our objective is to describe blood culture sampling frequencies, clinical indications for sampling, and ABT administration in three NICUs.

Methods

Study design and population

We performed a multicenter prospective cohort study. All neonates admitted to the NICU of three Belgian tertiary referral hospitals for ≥ 48 h were eligible for inclusion. During an 18-month study period (July 2013–December 2014), a consecutive series of blood cultures obtained as part of a sepsis evaluation were included; cultures sampled within 7 days of each other and after ABT initiation were excluded. The outcome of a suspected episode was classified in “lab-confirmed sepsis,” “clinical sepsis,” or “no sepsis.” Classification was made by the principal investigator (E.V.).

Center 1 and center 2 are academic hospitals with respectively a 32-bed and 28-bed NICU; center 3 is a non-academic hospital with a 32-bed NICU. The yearly mean NICU admission rate of center 1 is approximately 550 of which circa 18% are very low birth-weight (VLBW) neonates (≤ 1500 g); specialist services include pediatric cardiology and cardiac surgery. For centers 2 and 3, the yearly NICU admission rate is approximately 400 and 470 respectively of which 25 and 15% are VLBW. Only newborns (i.e., less than 28 days of life) are admitted to these units.

Sampling practice: data collection

Prospective registration of nosocomial sepsis (NOSEP) score [17] (including CRP ≥ 14 mg/L, neutrophil fraction > 50%, temperature > 38.2 °C, total parenteral nutrition [TPN] ≥ 14 days, thrombocytopenia < 150 × 103/μL), 10 predefined clinical signs [19] (including increased/acute onset of tachypnea/oxygen need, apnea/bradycardia, hypotension, glucose intolerance, impaired peripheral perfusion, lethargy, temperature instability, ileus or feeding intolerance, oliguria, base deficit), CRP value, and ABT initiation were completed at time of sampling by the attending neonatologist using sticker labels; assessment of clinical signs encompasses an interpretation period of ≤ 24 h preceding sampling. Also, registration of risk factors for HAS was performed (including gender, birth-weight, gestational age, intra-uterine growth restriction, type of catheterization, TPN days, mechanical ventilation for ≥ 24 h prior to culturing, recent major surgery ≤ 7 days prior to culturing, postnatal steroid administration prior to culturing, pathogen, days of antibiotics, other infection, and length of stay). Blood cultures were sampled by venipuncture or from a newly inserted central line; volume was standardized at 1 mL; skin or catheter-hub disinfection time was standardized at 30 s. In center 2, sampling was indicated when a NOSEP score of ≥ 8 was attained. Center 1 and center 3 did not use any protocol or procedure in guidance of sampling. In all centers, blood cultures were assessed using the BacT/Alert 3D automated microbial detection system (bioMérieux Inc., Durham, NC).

Definitions and classification

HAS: occurs ≥ 48 h after birth [15].

Lab-confirmed sepsis: a pathogen is detected in a blood, urine, or cerebrospinal fluid culture or X-ray-confirmed ventilator-associated pneumonia and necrotizing enterocolitis. When the causative pathogen is a skin commensal, mainly coagulase-negative staphylococci, additional criteria are needed: two clinical signs and a CRP of > 20 mg/L [11].

Clinical sepsis: presence of two clinical signs and ABT initiation of ≥ 5 days when no pathogen was detected in one or more cultures and no apparent other site of infection is assumed [11], or a skin commensal was detected plus presence of two clinical signs and ABT initation for 5 days or more and no apparent other site of infection is assumed.

No sepsis: not meeting criteria of clinical sepsis or contamination (skin commensal not meeting criteria for clinical sepsis).

Statistics

Logistic regression was used to determine the association between the indicators and lab-confirmed sepsis (“no sepsis” and “clinical sepsis” as controls); the association of patient-specific risk factors (“birth-weight,” “gestational age,” “gender,” and “intra-uterine growth restriction”) was only evaluated for the first suspected HAS episode. Unadjusted odds ratios (ORs) and 95% confidence intervals (CI) were calculated. The statistical package of social science version 21 was utilized for all tests (SPSS Inc., Chicago, IL).

Ethics committee

The study was approved by the ethics committee of the participating centers.

Results

Patients

A total of 413 blood culture samples in 301 neonates were registered. Of these 413 samples, 95 were excluded because of prior ABT and an additional 19 because of reiteration ≤ 7 days. For final analysis, 299 samples in 212 neonates were included of which 118 (55.7%) had birth-weight ≤ 1500 g; median birth-weight, gestational age, and length of stay were respectively 1363 g (interquartile range [IQR], 919–2165), 31 weeks (IQR 27–35), and 41 days (IQR 20–77). Characteristics of the 212 included neonates are displayed per center in Table 1.

Table 1 Neonatal characteristics per center
Full size table

Pathogen distribution of the lab-confirmed episodes is presented in Table 2.

Table 2 Pathogen distribution of lab-confirmed episodes
Full size table

Episode classification and incidence

Classification of the suspected episodes and incidence per center are in Table 3. Among the 299 included samples, respectively 66 (22.1%), 74 (24.7%), and 157 (52.6%) were classified as “lab-confirmed sepsis,” “clinical sepsis,” and “no sepsis.”

Table 3 Classification of sepsis episodes and incidence
Full size table

The proportion of blood cultures sampled in suspicion of HAS per total admissions during the 18-month study period was respectively 23.3% (center 1, 192/825), 10.0% (center 2, 60/600), and 6.7% (center 3, 47/704) (p < 0.001); the proportion of lab-confirmed + clinical sepsis per suspected episodes was respectively 36% (70/192, center 1), 77% (46/60, center 2), 55% (26/47, center 3) (p < 0.001).

Sampling practice

Differences between centers in sampling practice are presented in Table 4.

Table 4 Sampling practice
Full size table

For the total cohort, the median number of clinical signs per suspected episodes was significantly lower for the cohort “no sepsis” (1, IQR 0–1) compared to “lab-confirmed sepsis” (3, IQR 1–4) (p < 0.001); in contrast, median CRP was not significantly different between these two cohorts, respectively 10.0 mg/L (IQR 0.0–22.3) for “no sepsis” and 9.0 mg/L (IQR 1.8–27.5) for “lab-confirmed sepsis” (p = 0.81).

Of all suspected episodes with ≥ 5-day ABT administration, respectively 48% (70/146, center 1), 100% (46/46, center 2), and 84% (26/31, center 3) (p < 0.001) were classified as for “lab-confirmed sepsis” or “clinical sepsis.”

Concerning the NOSEP score of 8 as indicator for sampling in center 2, 39 suspected episodes (65%) did not reach this score. The median number of clinical signs at time of culturing of these 39 cases was 2 (IQR 1–4) and the median number of clinical signs of the 21 cases with a NOSEP score of 8 was 3 (IQR 2–4) (p = 0.019).

Association of indicators with lab-confirmed sepsis

Unadjusted odds of clinical indicators and risk factors for lab-confirmed HAS are reported in Table 5.

Table 5 Association of clinical indications and risk factors with lab-confirmed sepsis
Full size table

Discussion

We described sampling practice in suspicion of HAS; substantial differences between the centers were observed. Additionally, bivariate analysis of indicators for lab-confirmed sepsis revealed that most clinical signs were not associated with lab-confirmed sepsis; in contrast, the ad random combination of 2 and 3 clinical signs was associated.

Incidence

Significant difference in incidence between the centers (3.6% center 1, 4.6% center 2, 1.1% center 3) and in the proportion of samples per total admissions (23.3% center 1, 10% center 2, 6.7% center 3) can indicate a case-mix in population and/or variance in sampling practice. Since in center 1 risk profile for HAS seems lower (i.e., higher median birth-weight and gestational age) and proportion of samples per total admission is higher, difference in sampling practice is assumed [7]. To our knowledge, research on sampling frequencies in neonates is limited. Blackburn et al. [4] reported a 48% proportion of samples/admissions in a tertiary neonatal unit in London at “day of birth” of which 15% were lab-confirmed on day 8. Unfortunately, our results are incomparable because of the focus on blood cultures sampled after 48 h of admission.

Sampling practice

For center 1, a proportion of 27% zero clinical signs per suspected episode, a median of 1 clinical sign, and a median CRP of 11 mg/L at time of culturing might indicate a sampling practice based on CRP value or probably other biomarkers or risk factors. Another possible explanation for the high proportion of samples in the absence of clinical signs might be that some doctors perform a systematic weekly sampling in specific cases, although this was not defined as a standard procedure. Contrariwise for the other centers, sampling practice was more driven by clinical signs and not by CRP value. Similar to centers 2 and 3 are the results of Bekhof et al. [3] wherein more than half of their suspected HAS cohort (65%) had normal CRP (i.e., ≤ 10 mg/L). This result is in contrast with the results of Ohlin et al. [23] conducted on 401 neonates with suspected bloodstream infection of which 129 were in suspicion of HAS; they found that only 22% had a CRP < 10 mg/L.

Remarkable are the results for center 2. The highest proportion of lab-confirmed + clinical sepsis per suspected episode was observed (77 vs. 36% center 1 and 55% center 3) potentially indicating a more efficient sampling practice. Since the scope of this study was solely a description of sampling practice features, we did not assess the clinical consequences for the neonate. For this, we cannot judge whether the policy of center 2 is truly more efficient than the policies of the other centers. Furthermore, this center uses a NOSEP score of 8 in guidance of blood culturing; however, only 35% of the suspected episodes had a score of 8 or more, so in 65% of the episodes sampling was executed because of other indication(s). For the 65% not reaching a NOSEP score of 8, a median of two clinical signs at time of culturing was observed, so it is assumed that the presence of two clinical signs could also indicate sampling in this center. We note that episodes with a NOSEP score of 8 had significantly more clinical signs at time of blood culturing than episodes without a NOSEP score of 8; this may implicate an association with NOSEP and the clinical condition of the neonate in suspicion of HAS.

Because of the possible devastating consequences of HAS, a low threshold for initiating ABT is generally accepted [9]. Nonetheless, inadequate, inappropriate empiric treatment can foster antimicrobial resistance, compromise gastrointestinal immunity, and be associated with adverse outcomes [1, 12]. We observed that in all centers, blood cultures were taken without ABT initiation (note that follow-up cultures, as a part of ABT evaluation procedure, are excluded). The rationale for not starting antibiotics might also be that some doctors perform a systematic weekly sampling in specific cases. Implementing antimicrobial stewardship programs can be helpful to reduce antimicrobial initiation and duration in NICU [5, 20, 27, 36].

Association of indicators with lab-confirmed sepsis

Unadjusted regression analysis confirms that most clinical signs are in separation no specific sepsis indicators; only “ileus/feeding intolerance” and “impaired peripheral perfusion” appear to be specific symptoms for lab-confirmed sepsis. Since heterogeneity in symptom definition and population is observed, comparison with other research results is difficult. Modi et al. [19] searched for clinical signs associated with lab-confirmed sepsis in low-gestational-age infants (≤ 31 weeks) and found that all 10 signs, as applied in our study, were significantly associated. Although the same definitions of these signs were used, comparison with Modi et al. [19] is probably inappropriate since they included all NICU patients. As the median number of clinical signs is significantly higher for “lab-confirmed” compared to “no sepsis” and the occurrence of two or three ad random clinical signs at time of culturing have higher odds for “lab-confirmed sepsis,” the presence of several clinical signs can be valuable as sepsis indicator.

We did not find an association between “CRP rise” and “lab-confirmed sepsis” which is in contrast to the study results of Ohlin et al. [23] and Mahieu et al. [17] but in agreement with other studies [16, 29]. The predictive value of CRP is controversial since a CRP rise is known as a late diagnostic marker for infection [10, 16, 29].

“Mechanical ventilation” or “respiratory distress,” “peripheral or central line catheterization,” and impaired peripheral perfusion are well-known risk factors for HAS and frequently included in prediction models [3, 8, 13, 24, 35]. We agree with these well-known individual risk factors and clinical characteristics of infection. Though, according to Mahieu et al.[17] when using multivariate analysis other risk factors such as prolonged exposure to TPN seem more important than the confounding factors “mechanical ventilation” and prolonged exposure to “central line catheterization”. Moreover, at time of the NOSEP score development, only objective clinical characteristics were taken into account to exclude the subjective interpretation by the bedside physician and/or nurse. Therefore, “respiratory distress” and “poor peripheral perfusion” were not taken into account for the development of the NOSEP prediction model.

Prediction of HAS is a major issue in neonatal intensive care; unspecific signs and case-mix hamper the development of accurate prediction models. However, the NOSEP score is perceived with moderate diagnostic accuracy and is externally validated with good AUC after redefining cutoff of NOSEP items [17, 18]. Our observations in clinical practice revealed that a score of 8 as a separate indicator did not seem sufficient in guidance of blood culturing. Additionally, the unadjusted odds of NOSEP score 8, i.e., 1.6 (0.9–2.8), are only borderline significant. In contrast, for center 2, the combination of NOSEP sore 8 and several clinical signs might have led to an efficient sampling practice as stated earlier. For this and because it is our belief that clinically ill neonates cannot be left without initiation of diagnostic procedures and treatment, it may be appropriate to use the presence of two clinical signs as a threshold for blood culturing.

Limitations and recommendations

It is possible that some infants might be wrongly classified since blood culture test is imprecise. However, our applied definition of lab-confirmed sepsis caused by skin commensals with additional criteria most probably exclude the false-positive cultures.

In clinical practice, the interpretation of the non-specific clinical signs is not always obvious and probably facing inter-observer variability which might have influenced our results. Nevertheless, for most clinical signs, standardized definitions were used. Also, limited incidence of some clinical signs might have introduced bias (type I error) in unadjusted regression analysis.

When a larger cohort (including a control group) is obtained, future research can explore the benefit of using prediction models in guidance of sampling practice; assessment of mortality and morbidity directly attributable to HAS can help to judge the different policies. Less blood culture samples result in less workload for the staff, a financial profit for the hospital, and can result in appropriate ABT [14, 36]. Moreover, we cannot ignore the decreased burden for neonate.

Conclusion

Center heterogeneity in sampling practice is substantial. Optimizing sampling practice can be recommended. Although not obvious to uniform due to case-mix in population, guidelines might be helpful and desirable.