Objective: To examine the reliability of “low-risk” criteria (LRC) to exclude serious bacterial infection (SBI) in febrile neonates (⩽28 days), according to age in weeks.
Design: Epidemiological and clinical data and final diagnosis of all febrile neonates presenting to the emergency room from June 1997 to May 2006 were reviewed. Neonates who fulfilled specific LRC for the presence of SBI were classified as LRC+. The prevalence of SBI and the percentage of LRC+ neonates who had SBI were calculated for each of the first 4 weeks of life.
Results: A total of 449 neonates were evaluated. Eighty-seven (19.4%) neonates had an SBI. The prevalence of SBI among infants 3–7, 8–14, 15–21 and 22–28 days of age was 21.6%, 26.1%, 17.9% and 12.1%, respectively (p = 0.007 for linear trend after second week of life). Of the 226 LRC+ neonates, 14 (6.2%) had an SBI, including one case of bacteraemia and meningitis and 13 cases of urinary tract infection (UTI). The negative predictive value (NPV) of the LRC for SBI was 93.8% (95% CI 90.1% to 96.4%). The prevalence of SBI among LRC+ infants 3–7, 8–14, 15–21 and 22–28 days of age was similar, with rates of 15%, 6.3%, 3.0% and 6.7%, respectively.
Conclusion: LRC are not sufficiently reliable to exclude the presence of SBI, including bacteraemia and meningitis in febrile neonates of all ages. All febrile neonates should therefore be hospitalised, undergo a full “sepsis evaluation” and receive empirical intravenous antibiotic therapy.
Statistics from Altmetric.com
The higher risk of serious bacterial infection (SBI) among young infants coupled with the difficulty in identifying all infants with SBI, mandates a cautious approach to the evaluation and management of fever in such patients. Historically, febrile infants under the age of 3 months were routinely subjected to a full “sepsis evaluation” including procurement of blood, urine and cerebrospinal fluid (CSF) cultures, were hospitalised, and begun on empirical antimicrobial therapy.1–3 Over the past three decades, the increased awareness of potentially deleterious effects of hospitalisation4–6 as well as the realisation that SBI, while more common in younger infants still occurs at a relatively low incidence, has prompted considerable study as to whether the age until which this policy must be adopted may be safely lowered. The 1993 practice guidelines for the management of febrile infants recommended outpatient management with empirical intramuscular antibiotics for febrile infants older than 28 days of age who fulfil “low-risk criteria”.7 These guidelines, as well as those more recently published8 call for all febrile neonates (⩽1 month of age) including those classified as low risk, to undergo a full sepsis evaluation, hospitalisation and empirical antimicrobial therapy. Studies which have sought to assess the reliability of the low-risk criteria among neonates have reached disparate conclusions.9–14 Importantly, most of them considered neonates of all ages as a homogeneous group, and none evaluated the potential reliability of these criteria during the first month of life according to age in weeks. We hypothesised that younger neonates are at higher risk for SBI since they were more recently exposed to bacterial pathogens as they traversed the birth canal. We also hypothesised that urinary tract infection (UTI) would be more common in younger male neonates since ritual circumcision is performed on most Jewish males on the eighth day of life. Hence, the low-risk criteria may be a more reliable predictor of the absence of SBI among older neonates in comparison with younger ones.
The aim of this study therefore, was to assess the reliability of the low-risk criteria in excluding SBI among febrile neonates according to age in weeks.
PATIENTS AND METHODS
All available medical records of neonates (⩽28 days) presenting to the paediatric emergency room (ER) of Shaare Zedek Medical Center for evaluation of fever during the 10-year period from June 1997 to May 2006 were reviewed. Infants with a rectal temperature of ⩾38°C measured in the ER or at home before arrival were considered eligible for study. Exclusion criteria were: birth before 37 weeks’ gestation, prior hospitalisation or receipt of antibiotics, known chronic disease, and a source of infection apparent on physical examination other than acute otitis media (AOM). Throughout the 10-year study period, all neonates with fever of ⩾38°C underwent the same sepsis evaluation, which included complete blood count (CBC), blood culture, bladder catheterisation or suprapubic aspiration to obtain a sterile urine specimen for dipstick analysis (UA) and culture, and lumbar puncture to obtain CSF for cell count, chemistry, culture and, when indicated, Gram stain. A chest radiograph (CXR) was obtained when respiratory signs or symptoms were present. All infants, independent of initial CBC, UA and CSF findings, were hospitalised and treated with intravenous (IV) antibiotics pending culture results.
Data retrieval was performed by four senior physicians under the guidance of the principal investigator (SS), who personally performed all of the data entry, and also reviewed a randomly selected large sample of questionnaires to assure the accuracy of the data. The authors were not, and could not be, blinded to the culture results since these results were readily visible in the records.
Data collected from the chart of each infant included:
Gender, ethnic background, age in days on presentation to ER, and for circumcised males, age in days when ritual circumcision was performed.
Height of fever in ER, or at home if fever in ER was <38°C.
General appearance in ER, either “not ill appearing” or “ill appearing” based on the medical note written by the examining physician. If not specifically mentioned in the chart, “ill appearance” was assigned if the infant was noted to show signs of illness such as lethargy, irritability, pallor, cyanosis, dehydration or respiratory distress.
Results of CBC, UA and CSF cell count.
Results of blood culture, urine culture and CSF culture. If performed, results of CXR, stool culture and middle ear fluid culture were also noted.
Criteria for being at low risk for SBI (LRC+) were: not ill appearing, peripheral white blood cell (WBC) count of 5000–15 000/mm3, absence of leucocyte esterase (LE) in non-centrifuged urine on dipstick test (Multistix®10 SG, Bayer Corporation, Elkhart, Indianapolis, USA), and <23 WBC/high power field (hpf) on microscopic examination of the CSF.15 Infants who did not fulfil these four criteria were classified as LRC-.
All infants were also classified with respect to the presence or absence of SBI. SBI was diagnosed (SBI+) if: a culture of blood, urine, CSF or stool grew a known bacterial pathogen, a CXR revealed a lobar infiltrate, or a bone or soft tissue infection not present on admission to the ER became apparent after hospitalisation. All other infants, including those with otitis media, were considered not to have an SBI (SBI-).
UTI was diagnosed if there was: any growth of a single known urinary bacterial pathogen, isolated growth of >1000 cfu/ml of a single skin bacteria, >1000 cfu/ml of at least one known urinary bacteria pathogen if two bacteria were isolated, and >10 000 cfu/ml of at least one known urinary pathogen if three organisms were isolated.16
The prevalence of SBI and the percentage of LRC+ cases which were SBI+ were calculated for each of the 4 weeks of life.
Data were typed into a computerised questionnaire written in Epi-Info 6.04d and analysed by this programme and by Epi-Info3.3.2 (both from CDC, Atlanta, USA) and PEPI for Windows (Abramson JH & Gahlinger PM, www.brixtonhealth.com). We applied the t test, chi square test (Fisher exact where applicable) and the linear trend analysis. Ninety-five per cent CIs were calculated using the established mid-P method.
There were 644 febrile neonates who presented to the ER during the study period. One hundred and ninety-five were excluded because of missing data; 14 lacked culture results and 181 because other laboratory parameters were unavailable. Thus, the study group comprises the 449 neonates for whom the medical records contained all the data necessary for evaluation. A sub-analysis was performed for the 181 neonates with known culture results who were excluded, to ensure that there was no selection bias affecting the rate of SBI. Indeed, in this subgroup there were 32 (17.6%) neonates with SBI, as compared with the study group in which there were 87 (19.4%, two-tailed p value = 0.65).
Two hundred and ninety of the 449 (65%) neonates were males. Four hundred and twenty-six neonates (95%) were Jewish, and the remaining were of Moslem descent.
The date of life on which ritual circumcision was performed was documented in the medical records of 85 Jewish males, of whom 82 (96.5%) were circumcised on the eighth day of life.
Serious bacterial infection
SBI was diagnosed in 87 (19.4%) neonates, of whom 69 (79%) were males. The distribution of types of infections and causative organisms is shown in table 1. For the six cases of UTI in which more than one organism was isolated, the predominant one is listed.
The prevalence of SBI among infants 3–7, 8–14, 15–21 and 22–28 days of age is displayed in table 2. The decline in SBI rate after the second week of life was significant (p = 0.007 for linear trend).
Fourteen (6.2%) of the 226 LRC+ neonates had an SBI. Streptocccus pneumoniae grew from the blood and CSF of an 18-day-old female who had six WBCs in her CSF. This same organism was isolated from her middle ear fluid obtained by tympanocentesis. The remaining 13 neonates ranging in age from 6 to 27 days had an isolated UTI. The negative predictive value (NPV) of the low-risk criteria to exclude SBI among the 449 neonates was 93.8% (95% CI 90.1% to 96.4%). The overall prevalence of SBI among LRC+ neonates was similar between the four different age groups.
Urinary tract infections
Sixty-seven (23.1%) of the 290 males and 15 (9.4%) of the 159 females had a UTI (p⩽0.001). Ritual circumcision was documented in the medical records as having been performed on the eighth day of life in 33 of the 60 Jewish males older than 8 days of age who had a UTI. Among 30 of these 33 (91%) male infants, UTI occurred within 8 days subsequent to their circumcision.
The urine dipstick revealed LE in 60 of the 82 (73.2%) cases of UTI, yielding a sensitivity of 73.2% and an NPV of 94.1% (95% CI 91.7% to 96.3%).
Risk of SBI and various clinical parameters (table 4)
Males had a higher rate of SBI than females (24% vs 11%; p<0.001).
The mean (SD) ER temperature was slightly higher among those with SBI than those without (38.3 (0.61)°C vs 38.1 (0.63)°C, p = 0.006). Yet, high fever as measured in the ER did not confer a statistically significant greater risk for SBI. There was no statistically significant association between otitis media and bacteraemia or other SBI.
Though none of the 16 neonates with leucopenia (WBC range 3500–4900/mm3) had SBI, this finding lacks statistical significance due to the small sample size. Leucocytosis (WBC>15 000/mm3) was strongly associated with SBI, but had a positive predictive value of only 33%. Infants with bronchiolitis had a lower prevalence of SBI than those without this diagnosis; however this difference was not statistically significant.
Ill appearance was associated with SBI. Importantly, there were six neonates with SBI (3 UTI, 2 bacterial pneumonia, 1 bacteraemia) whose only characteristic warranting their being categorised as LRC-, was their ill appearance. The reliability of this parameter improved significantly at 2 weeks of age, and steadily thereafter (table 2). Overall however, clinical observation demonstrated very low sensitivity (21%) and a poor NPV (82.5%, 95% CI 78.5% to 86.0%).
The findings of our study suggest that the low-risk criteria may not be safely relied upon to exclude SBI in febrile neonates. Of the infants in our study who met these criteria, 6.2% had an SBI. Analysed differently, 14 of the 87 (16.1%) cases of SBI would not have been detected by the low-risk criteria. This is the first study which sought to analyse the predictive accuracy of the “low-risk criteria” according to age in weeks. Our relatively large sample of 449 neonates, 226 of whom were LRC+, facilitated this analysis. Contrary to our hypothesis, these criteria are no more reliable among older neonates than among younger ones.
The reliability of the low-risk criteria in febrile neonates has been previously addressed in several studies9–14 17 and one case report18 (table 5). The specific clinical criteria used in each study are essentially similar to each other as well as to ours, with only minor variations. As apparent in the table, the disparate findings of these investigations have led to differing management recommendations. Dagan19 20 pioneered the investigation of the efficacy of low-risk criteria in excluding SBI among young febrile infants in two separate studies. Specific data regarding the neonates studied were not reported, precluding their inclusion in the table. However, based on his findings, Dagan recommended outpatient management of low-risk neonates without empiric antibiotics.20 Recently, Marom et al14 reported only one case of SBI (a UTI) among 166 febrile low-risk neonates, and thus arrived at the same conclusion regarding management. Though our low-risk criteria and urinalysis method were very similar to those of Marom et al, we found the former to have a lower NPV regarding SBI (99.4% vs 93.8%), and the latter, lower sensitivity for UTI (87.0% vs 73.2%).
The overall rate of SBI in our study population, 19.4%, is slightly greater than that of previous reports of febrile neonates, which range from 6.5% to 17.8%.9–12 19 21 22 Marom et al14 reported an overall SBI rate of 28%. In that study however, AOM was regarded as an SBI, and accounted for 20% of all cases. Like most authors,9–13 17 19 21 22 we did not consider AOM an SBI.
Our less restrictive definition of UTI contributed minimally to our relatively high SBI rate. Other studies required the growth of a single organism in order to qualify as a UTI. We, in accordance with criteria established by the American Society of Microbiology,16 classified urine cultures with growth of more than one organism as positive, if the bacterial count exceeded a minimum threshold. Nonetheless, only seven of the 82 UTI cases in our study had a urine culture with growth of more than one organism.
What is already known on this topic
Practice guidelines established for the management of febrile well-appearing infants between the ages of 1 and 2 months support outpatient management for those, who according to clinical criteria, are considered at low risk for harbouring a serious bacterial infection.
Studies which have evaluated the reliability of these criteria in febrile neonates (age ⩽28 days) have led to disparate findings.
What this study adds
“Low-risk” criteria do not reliably exclude the presence of a serious bacterial infection (SBI) among febrile neonates, and the sensitivity of these criteria to exclude SBI does not improve with increasing age of neonates.
All febrile neonates should undergo a full sepsis evaluation including lumbar puncture, be hospitalised, and treated empirically with intravenous antibiotics.
The higher SBI rate in our population is undoubtedly also due to the significant number of male infants who underwent ritual circumcision on the eighth day of life. This procedure may cause a UTI to develop during the subsequent 1–12 days.23–26 Urinary retention due to tight wrapping of the penile shaft is the most likely mechanism,26 though non-sterile technique25 and pain-induced urinary retention25 may also play a causative role.
Urinalysis in our ER during the 10-year study period was performed by the dipstick method. The sensitivity of the dipstick urinalysis in detecting UTI (73.2%), while seemingly disappointing, is within the range of previous reports regarding urine dipstick analysis to detect UTI in young children.27–29 Since in all studies the most common SBI among LRC+ neonates is a UTI (13 of the 14 LRC+ neonates in our study), a screening test for UTI with greater sensitivity could render the low-risk criteria significantly more reliable. Microscopic analysis for pyuria was recently shown to be inferior to the dipstick method in detecting UTI in children, while the performance of Gram stain and dipstick were found to be similar.28 “Enhanced urinalysis”, which combines Gram stain and WBC count/mm3 as measured by a cytometer of a non-centrifuged urine specimen, was reported to have a sensitivity and NPV, of 96% and 99.7%, respectively, when used to detect UTI in febrile infants less than 60 days of age.13 This urinalysis method, though arguably superior, is more costly, less specific, and not available in most hospitals.27 The method of urinalysis used in our study is more representative of that commonly used to screen for UTI in febrile infants.
The presence of nitrites on the dipstick urinalysis was not consistently recorded in the medical chart. Therefore the absence of nitrites was not included as part of our low-risk criteria. It is unlikely though that the test for nitrites would have significantly improved the sensitivity of the urinalysis. Urinary nitrites are a poor indicator of UTI, with a sensitivity of 10% in children under 2 years of age.29 Very frequent voiding in neonates limits the time necessary for bacteria to reduce nitrates to nitrites,28 and thus would likely lower the sensitivity even further. A positive nitrite result in the absence of LE on dipstick analysis would also be a very rare event.
One limitation of our study is the omission of the absolute band count (ABC) or band-to-neutrophil ratio from the low-risk criteria. Most, but not all14 similar studies of febrile neonates considered either an ABC of <1500/mm3 9 10 13 17 19 20 or a band-to-neutrophil ratio of <0.211 12 as part of their low-risk criteria. Since our hospital laboratory does not identify band forms, we were unable to evaluate this parameter. In separate studies of infants 28–8930 and 29–5631 days of age, increased band forms were found to be predictive of SBI. Yet, such data are not presented in the studies of febrile neonates less than 28 days of age.
Another limitation of this study is that the outcome measure by which the predictive accuracy of the low-risk criteria were judged was microbiological or clinical diagnosis rather than the ultimate outcome of the neonate, that is, morbidity and mortality. We had no control group of febrile LRC+ neonates who were initially observed without empirical antibiotic therapy. Most similar investigations also suffer from this limitation, a point which has been previously addressed.32 A delay in the institution of antimicrobial therapy until the infant clinically deteriorated or until culture results became available, may not necessarily have led to a worse outcome among some or all of the 13 LRC+ neonates with UTI. This contention however, is less reasonable in the case of the neonate with bacteraemia and meningitis.
In conclusion, the principal finding of this study is that the accepted “low-risk” criteria used to assess febrile infants 28–60 days of age are not sufficiently reliable in neonates of all ages, as their use would lead to an unacceptable percentage of serious bacterial infections being unidentified. UTI represents the majority of SBI cases undetected by the low-risk criteria. Nonetheless, cases of bacteraemia and meningitis are rarely “missed” as well. The search for more sensitive laboratory indicators of sepsis in febrile neonates continues, but thus far they remain elusive.33 Until they are identified, we suggest that all febrile neonates be hospitalised, undergo a full “sepsis evaluation” and receive empirical intravenous antibiotic therapy.
Competing interests: None.
Ethics approval: Approval for this study was granted by the Human Rights Committee of Shaare Zedek Medical Center.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.