Early-onset sepsis risk calculator: a review of its effectiveness and comparative study with our evidence-based local guidelines

Background According to most early-onset sepsis (EOS) management guidelines, approximately 10% of the total neonatal population are exposed to antibiotics in the first postnatal days with subsequent increase of neonatal and pediatric comorbidities. A review of literature demonstrates the effectiveness of EOS calculator in reducing antibiotic overtreatment and NICU admission among neonates ≥34 weeks’ gestational age (GA); however, some missed cases of culture-positive EOS have also been described. Methods Single-center retrospective study from 1st January 2018 to 31st December 2018 conducted in the Division of Neonatology at Santa Chiara Hospital (Pisa, Italy). Neonates ≥34 weeks’ GA with birth weight ≤ 1500 g, 34–36 weeks’ GA neonates with suspected intraamniotic infection and neonates ≥34 weeks’ GA with three clinical signs of EOS or two signs and one risk factor for EOS receive empirical antibiotics. Neonates ≥34 weeks’ GA with risk factors for EOS or with one clinical indicator of EOS undergo serial measurements of C-reactive protein and procalcitonin in the first 48–72 h of life; they receive empirical antibiotics in case of abnormalities at blood exams with one or more clinical signs of EOS. Two hundred sixty-five patients at risk for EOS met inclusion criteria; they were divided into 3 study groups: 34–36 weeks’ GA newborns (n = 95, group A), ≥ 37 weeks’ GA newborns (n = 170, group B), and ≥ 34 weeks’ GA newborns (n = 265, group A + B). For each group, we compared the number of patients for which antibiotics would have been needed, based on EOS calculator, and the number of the same patients we treated with antibiotics during the study period. Comparisons between the groups were performed using McNemar’s test and statistical significance was set at p < 0.05; post-hoc power analysis was carried out to evaluate the sample sizes. Results 32/265 (12.1%) neonates ≥34 weeks’ GA received antibiotics within the first 12 h of life. According to EOS calculator 55/265 (20.7%) patients would have received antibiotics with EOS incidence 2/1000 live births (p < 0.0001). Conclusion Our evidence-based protocol entails a further decrease of antibiotic overtreatment compared to EOS calculator. No negative consequences for patients were observed.

As the incidence of EOS has decreased over the last two decades, clinicians raised concerns about antibiotic exposure among uninfected newborns: according to Group B Streptococcus (GBS) EOS prevention guidelines, approximately 10% of the total neonatal population are exposed to antibiotics in the first postnatal days, and almost 100% of the extremely preterm population are exposed to ampicillin and an aminoglycoside [10]. Early antibiotic exposure is associated with the emergence of antibioticresistant pathogenic microorganisms and with the decrease of intestinal microbial diversity, which can cause very difficult to treat infections [10]. Antibiotics administration in the neonatal period has also been linked with late onset sepsis, necrotizing enterocolitis, increased mortality and long term health outcomes such as childhood asthma, obesity, inflammatory bowel disease, celiac disease and type 1 diabetes [10]. Furthermore, administration of antibiotics to neonates often results in admission to intensive care unit, decreased breastfeeding, invasive procedures and increased hospital costs [11].
For all these reasons, it is important to avoid unnecessary antibiotics administration to patients during the early post-natal period [11]. However, the clinical diagnosis of sepsis is challenging for neonatologists because many signs of sepsis are nonspecific and are observed with other non-infectious conditions [7]. On the other side, low-level bacteremia (4 colony-forming units/mL or less), inadequate blood specimens (less than 1 mL) or maternal antibiotic treatment before or during delivery may result in negative blood cultures [1,7]. It has been estimated that the incidence of culture-negative EOS is 6 to 16 times higher than that of culture-confirmed EOS [1]. Total white blood cell (WBC) count with its subcomponents and platelet count have also shown a poor predictive accuracy, and the specificity and selectivity of genetic biomarkers are yet to be fully evaluated [7,12]. Protein biomarkers demonstrate high specificity and sensitivity and include C-reactive protein (CRP) and Procalcitonin (PCT), which are the most commonly used protein biomarkers for the diagnosis of sepsis and monitoring of antibiotic therapy [12][13][14][15]. Both CRP and PCT have a physiologic increase over the first 24-48 h of life; baseline concentrations of both markers are mainly affected by birth weight and gestational age (GA) [16]. On these basis, different attempts have been done to establish the appropriate cut-off values of both PCT and CRP [17][18][19]. Umbilical blood PCT and CRP have also been tested for EOS diagnosis; cut-off values were different among studies (0.5-2 ng/ ml for PCT and 1-10 mg/l for CRP) [20].
After June 2005, several studies have assessed the safety of monitoring neonates at risk for EOS with serial physical examinations: this approach resulted in less laboratory exams and antibiotics exposure without missing any case of EOS [21][22][23].
In December 2012 the Kaiser Permanente EOS calculator has been developed with the purpose of avoiding antibiotic overtreatment [24]. The EOS calculator is based on a multivariate predictive risk model which allows clinicians to estimate a newborn's individual risk for EOS given objective maternal risk factors and the infant's clinical presentation [24]. This model permits to overcome some disadvantages of the CDC algorithm, such as the dichotomization of the continuous variables and the inclusion of maternal chorioamnionitis (CAM) as an impactful risk factor for starting antibiotic therapy [24]. A vast majority of studies about the EOS calculator demonstrates its efficacy in reducing antibiotic overtreatment, laboratory testing, painful procedures and NICU admission with increased opportunities for mother-child bonding and breastfeeding (Table 1) [11,.
The objective of our study was to compare the administration of antibiotics based on our local EOS guidelines derived from current evidence with the calculator's recommendations in neonates born at ≥34 weeks' GA.

Methods
This was a single-center retrospective study from 1st January 2018 to 31st December 2018 conducted in the Division of Neonatology at Santa Chiara Hospital (Pisa, Italy). The parents of all subjects signed a written consent form and the study was approved by the ethics committee of the Meyer Children's Hospital of Florence. Based on our local guidelines, neonates born at ≥34 weeks' GA are divided into three categories (high, medium and low EOS risk) as shown in Table 2, and managed as shown in Table 3.
Neonates born at ≥34 weeks' GA during the study period at our institution were identified using admission logs. Thus, we retrospectively reviewed maternal and neonatal charts and collected data for input into the EOS calculator. We also collected data about other risk factors for EOS, mode of delivery, duration of labor, presence and type of clinical indicators of EOS with the time in which they appeared, relevant laboratory results, type and duration of antibiotic therapy. These data were obtained in order to verify physicians' compliance with our local guidelines and the correct classification of medium-and high-risk patients into non-septic patients, patients with culture-positive EOS and patients with    Table 4. For each study group, we compared the number of patients for which antibiotics would have been needed, based on EOS calculator, and the number of the same patients we treated with antibiotics during the study period. Data were collected into a designated database. We therefore used R Software version 3.6.2 for statistical evaluations; comparisons between the groups were performed using McNemar's test for paired nominal data and statistical significance was set at p < 0.05. Post-hoc power analysis was carried out to evaluate the sample sizes.

Results
A total of 1667 neonates born at ≥34 weeks' GA during the study period at our institution were identified using admission logs. Patients at low risk for EOS (1394/1667, 83.6%) and those who met exclusion criteria (8/1667, 0.5%) were excluded from the study. Thus, a total of 265 (15.9%) patients fulfilled inclusion criteria and were enrolled in the study. Demographic characteristics and risk factors for EOS of the study subjects are shown in Table 5.
According to our guidelines, 32/265 (12.1%) neonates were initiated on antibiotics in the first 12 h of life; none was initiated on antibiotics at 13-72 h of life.
After entering the data into the EOS calculator with local EOS incidence of 2/1000 live births, the recommendations were as follows: 1) No culture, no antibiotics, routine vitals (168 patients); 2) No culture, no antibiotics, vitals every 4 h for 24 h (7 patients); 3) Blood culture, vitals every 4 h for 24 h (35 patients); 4) Strongly consider starting empiric antibiotics, vitals per NICU (1 patient); 5) Empiric antibiotics, vitals per NICU (54 patients). Thus, according to EOS calculator, antibiotics were needed in 55/265 (20.7%) patients in the first 12 h of life. The difference with our local guidelines resulted statistically significant (p < 0.0001). Data are shown in Fig. 2.
As no cases of culture-positive EOS were observed during the study period, we also entered the same data into the EOS calculator with the lowest possible local EOS incidence (0.1/1000 live births). The recommendations were as follows: 1) No culture, no antibiotics,  Low-risk patients Routinely observation for ≥48 h before discharge The whole study population included all newborns ≥34 weeks' GA consecutively admitted to the Neonatology Department of Santa Chiara Hospital (Pisa, Italy) during the study period. The selection process of the study participants is represented in Fig. 1; the study groups included all newborns ≥34 weeks' GA, newborns 34-36 weeks' GA, and newborns ≥37 weeks' GA. All included patients have been managed in accordance with our local guidelines CBC Cell blood count, CRP C-reactive protein, EOS Early-onset sepsis, GA Gestational age, PCT Procalcitonin a Cord blood CRP and PCT or measurement of both markers within the first hour of life or at the onset of symptoms, blood culture and complete blood count (CBC) before receiving empirical antibiotics b The quantities of blood used for laboratory analyses are the following ones: 200 μL for CRP or PCT, 300 μL for both CRP and PCT, 400 μL for CBC and 1 mL for blood culture. Measurement of CRP and/or PCT is also possible on capillary blood samples. Cord blood CRP ≥ 10 mg/L and cord blood PCT ≥ 0.6 ng/mL are considered pathological; CRP ≥ 10 mg/L is considered abnormal even when performed on neonatal blood samples. PCT requires adjustment of the cut-off point with time according to the age-specific 95% reference intervals by Chiesa  A full-term newborn with culture-negative EOS starting with respiratory distress 6 h after birth received antibiotics according to our local guidelines; when using EOS calculator, this patient was classified as "equivocal" and would not have received antibiotics with EOS incidence 0.1/1000 live births.  As regards treatment, overlap between EOS calculator recommendations and our local guidelines was 88.3% (234/265 patients) when using EOS calculator with EOS incidence 2/1000 live births, and 90.9% (241/265 patients) when using EOS calculator with EOS incidence 0.1/1000 live births. Data are shown in Fig. 2.
Inborn infants ≥37 weeks' GA were 170/265 (64.2%). According to our local guidelines, 6/170 (3.5%) of these neonates were initiated on antibiotics in the first 12 h of life. A retrospective analysis of blood culture, CRP and PCT results showed no cases of culture-positive EOS and 4 cases of culture-negative EOS among the 1532 inborn infants ≥37 weeks' GA during the study period. Thus, the calculated incidence rate of EOS was 2.6/1000 live births. After entering data into the EOS calculator with local EOS incidence of 2/1000 live births, the recommendations were as follows: 1) No culture, no antibiotics, routine vitals (131 patients); 2) No culture, no antibiotics, vitals every 4 h for 24 h (4 patients); 3) Blood culture, vitals every 4 h for 24 h (17 patients); 4) Strongly consider starting empiric antibiotics, vitals per NICU (0 patients); 5) Empiric antibiotics, vitals per NICU (18 patients). Thus, according to EOS calculator, antibiotics were needed in 18/170 (10.6%) patients in the first 12 h of life; the difference with our local guidelines resulted statistically significant (p = 0.001). Data are shown in Fig. 4.
As no cases of culture-positive EOS were observed among inborn infants ≥37 weeks' GA, we also entered the same data into the EOS calculator with the lowest Post-hoc power analysis for statistically significant differences revealed that sample sizes were appropriate.

Discussion
Early diagnosis and treatment decision-making of neonatal EOS are challenging for clinicians; at the same time antibiotic resistance is an increasing problem, thus antibiotic overexposure among neonates should be avoided. For this purpose, we revisited the antibiotic stewardship program at our institution and drew up a protocol for management of neonates at risk for EOS.
The neonatal EOS calculator has been introduced to support the clinician's treatment decision-making of neonatal EOS. Most of Authors agree on the efficacy of the EOS calculator in reducing antibiotic overtreatment; however, some Authors have also reported patients with culture-positive EOS who would not have received antibiotics based on the EOS calculator. In our study, according to our local guidelines, antibiotics were needed in 32/265 enrolled patients ≥34 weeks' GA in the first 12 h of life; based on the EOS calculator, antibiotics would have been needed in 44/265 patients when using an EOS incidence of 0.1/1000 live births, and in 55/265 patients when using an EOS incidence of 2/1000 live births. As both differences resulted statistically significant, the use of our protocol is advantageous in clinical practice.
According to our study, a missed case of culturenegative EOS was observed when using the EOS calculator with EOS incidence 0.1/1000 live births; however, this EOS incidence includes only culture-positive EOS cases and, probably, underrates the true incidence of EOS at our institution. Furthermore, when using the EOS calculator all patients classified as "clinical illness" would have received antibiotics regardless of EOS incidence; according to our local guidelines 26/44 of these neonates received antibiotics with no negative consequences.
We think that the effectiveness of our protocol results from the inclusion of anamnestic data, clinical evaluation and laboratory exams. Anamnestic data permitted us to identify neonates with risk factors for EOS or elements which could explain clinical presentation (for example gestational diabetes, meconium aspiration or short labor in patients with respiratory distress). Thus, not all neonates classified as "clinical illness" received antibiotics according to our protocol. Clinical evaluation is very important since none of the patients with culture-negative EOS had risk factors for EOS, thus they were identified for the presence of clinical signs. Clinical evaluation is even crucial to decide whether to start antibiotic therapy because of the possibility of false-negatives with blood culture and false-positives with measurement of CRP and PCT. However, serial CRP and PCT measurements allowed us to identify patients with EOS, to control the efficacy of antibiotic therapy, and to decide when to stop antibiotic treatment. We think EOS calculator is an effective tool to reduce unnecessary antibiotics administration to neonates but it also has several limitations. First, the highest possible EOS incidence is 4/1000 live births, thus EOS calculator cannot be utilized in contexts with EOS incidence higher than 4/1000 live births. Second, its use is limited in the first 12 h of life but EOS can manifest itself between 12 and 72 h of life, although rarely, and serial measurements of CRP and PCT in the first 72 h of life allow us to identify all cases of EOS. Third, antibiotics are indicated to all neonates classified as "clinical illness" (persistent need for nCPAP/HFNC/ mechanical ventilation outside of the delivery room, hemodynamic instability requiring vasoactive drugs, neonatal encephalopathy/perinatal depression, need for supplemental O 2 ≥ 2 h to maintain oxygen saturations > 90% outside of the delivery room); we think that careful consideration of risk factors for EOS, anamnestic data and alternative diagnoses should further reduce unnecessary antibiotics administration. Fourth, equivocal patients can present with tachycardia, tachypnea, temperature instability or respiratory distress; however other clinical indicators of possible EOS (altered behaviour or responsiveness, feeding difficulties etc.) should be considered. Fifth, laboratory exams should be considered to reduce the number of patients receiving unnecessary antibiotics, above all among patients classified as "clinical illness", and to identify patients with EOS appearing after 12 h of life. Furthermore, our protocol incorporates the new definition for CAM: this disease is now defined as intraamniotic infection or "Triple I" and requires more clinical features for diagnosis. Thus, we make difference between neonates from mothers with "Triple I" and those with isolated maternal fever: this contributes in reducing the number of neonates receiving antibiotics.
However, even our protocol has many limitations. First, the number of 34-36 weeks' GA neonates receiving antibiotics is too high (26/95, none with EOS). Thus, we should re-evaluate clinical criteria for starting antibiotics and the optimal cut-off point for both CRP and PCT in late-preterm infants. Birth weight ≤ 1500 g should also be re-evaluated as a criteria to start antibiotics. Second, reducing antibiotics administration is money-saving. However, laboratory exams, above all PCT, are quite expensive. Third, we need serial clinical evaluations to identify neonates with clinical signs of EOS, especially those without maternal risk factors. However, even applying the EOS calculator requires serial clinical evaluations in the first 12 h of life. Fourth, serial blood samplings are needed for measurement of CRP and PCT; however, the first measurement is usually performed on cord blood, and blood sampling at 48 ± 4 h of life is the same for the newborn screening test. The remaining measurements sometimes coincide with blood samplings for gas analysis or glycemia evaluation. Fifth, our study is retrospective. Even if the course of each patient is well documented, the classification of neonates into well-appearing, equivocal or clinically ill is partly dependent on whomever is analyzing the medical records. Sixth, we should consider an earlier interruption of antibiotics at 48 h of life in well-appearing neonates with negative laboratory exams in order to reduce both antibiotics exposure and laboratory exams. Seventh, the EOS calculator has already been validated on more than 180.000 newborns, thus we should also validate our protocol on large scale to definitively prove its superiority.

Conclusion
EOS calculator has been proven to be an effective tool for treatment decision-making of neonatal EOS, however we have shown a further decrease in antibiotics administration through a continuous evidence-based update of local guidelines. Thus, continuous review of recommendations and updated guidelines are necessary to reduce both antibiotics administration and microbial resistance, with consequent reduction of related comorbidities, and to pursue the best possible antibiotic stewardship.