abstract BACKGROUND AND OBJECTIVES: Blood culture is the gold standard to diagnose bloodstream infectionbut is usually time-consuming. Prediction models aim to facilitate early preliminary diagnosisand treatment. We systematically reviewed prediction models for health care–associatedbloodstream infection (HABSI) in neonates, identified superior models, and pooled clinicalpredictors. Data sources: LibHub, PubMed, and Web of Science.
METHODS: The studies included designed prediction models for laboratory-confirmed HABSI orsepsis. The target population was a consecutive series of neonates with suspicion of sepsishospitalized for $48 hours. Clinical predictors had to be recorded at time of or beforeculturing. Methodologic quality of the studies was assessed. Data extracted includedpopulation characteristics, total suspected and laboratory-confirmed episodes and definition,clinical parameter definitions and odds ratios, and diagnostic accuracy parameters.
RESULTS: The systematic search revealed 9 articles with 12 prediction models representing 1295suspected and 434 laboratory-confirmed sepsis episodes. Models exhibit moderate-goodmethodologic quality, large pretest probability range, and insufficient diagnostic accuracy.Random effects meta-analysis showed that lethargy, pallor/mottling, total parenteral nutrition,lipid infusion, and postnatal corticosteroids were predictive for HABSI. Post hoc analysis withlow-gestational-age neonates demonstrated that apnea/bradycardia, lethargy, pallor/mottling,and poor peripheral perfusion were predictive for HABSI. Limitations include clinical andstatistical heterogeneity.
CONCLUSIONS: Prediction models should be considered as guidance rather than an absoluteindicator because they all have limited diagnostic accuracy. Lethargy and pallor and/ormottling for all neonates as well as apnea and/or bradycardia and poor peripheral perfusionfor very low birth weight neonates are the most powerful clinical signs. However, the clinicalcontext of the neonate should always be considered.
aGhent University, Belgium, Ghent, Belgium; bUniversity of Antwerp, Belgium, Antwerp, Belgium; cAntwerp University Hospital, Antwerp, Belgium; dGhent University Hospital, Ghent, Belgium; andeBurns, Trauma and Critical Care Research Centre, The University of Queensland, Brisbane, Australia
Ms Verstraete conceived and designed the study, contributed to the search of published work and data acquisition, and drafted the report; Mr Blot contributed to thedata acquisition, data analysis, and data interpretation and revised the statistical portions of the report; Dr Mahieu contributed to interpreting results and criticallyrevised the report; Dr Vogelaers critically revised the report and made substantial contributions on the final manuscript; Dr Blot contributed to the search of publishedwork, the acquisition of data, and critically revised the report; and all authors approved the final manuscript as submitted.
www.pediatrics.org/cgi/doi/10.1542/peds.2014-3226
DOI: 10.1542/peds.2014-3226
Accepted for publication Jan 27, 2015
Address of correspondence Evelien Verstraete, RN, MNSc, Department of Internal Medicine, Ghent University, De Pintelaan 185, 9000 Ghent, Belgium. E-mail: [email protected]
Health care–associated bloodstreaminfection (HABSI) is the mostfrequent infectious complication inNICUs. Previous studies documentincidence rates ranging from 5% to32%.1–4 For neonates with very lowbirth weight (#1500 g), the NationalInstitute for Child Health and HumanDevelopment reported an incidenceof 21%. HABSIs result in longerhospitalization (on average, +23 days)and a rise in mortality rate up to24% for very low birth weightneonates.5–10 Likewise, for pediatricand adult intensive care patients,HABSI is a common infectiouscomplication.11–13 Blood culture isthe gold standard test to diagnosisHABSI but prone to false-negativeor false-positive results.14–16 Bloodcultures positive for coagulase-negative staphylococci or other skincommensals might represent false-positive results due to contamination.Conversely, low blood culturevolumes, which are a major issue inpremature neonates, and previousantimicrobial therapy may beresponsible for false-negativeresults.16–18 The test is not onlyimprecise but also time-consuming.Hence, with respect to bothdiagnostic and therapeutic strategy,clinicians often must makepreliminary decisions based onlargely nonspecific signs, especially invery low birth weight neonates.Because of the possibly devastatingconsequences of HABSI, a lowthreshold for initiating antimicrobialtherapy is generally accepted.19
Nonetheless, inadequate,inappropriate, or unnecessaryempirical treatment can fosterantimicrobial resistance, compromisegastrointestinal immunity, and isassociated with adverseoutcomes.20,21 In this context,prediction models with clinicalparameters, in particular, clinical signs,have been developed to facilitatepreliminary sepsis diagnosis and theinitiation of antimicrobial therapy.
The first aim of this study was tosystematically review prediction
models for HABSI in hospitalizedneonates and to evaluate theirdiagnostic accuracy to find superiormodels. The second aim was to poolodds ratios (ORs) of individualclinical parameters to detect superiorclinical predictors of HABSI inneonates.
METHODS
A systematic literature review,diagnostic accuracy assessment ofthe prediction models, and randomeffects meta-analysis of clinicalparameters were performed. Theresults are reported in accordance withthe PRISMA guidelines (PreferredReporting Items for SystematicReviews and Meta-Analysis).
Literature Search Strategy
A systematic search was done by 2independent researchers (E.V., S.B.)on PubMed, LibHub, and Web ofScience without language and timeperiod restrictions (up to April 2014).After record screening based onabstract, language restrictions wereapplied. All keywords or Mesh termswere applied for [Title and/orAbstract] or [Topic] and containedfour parts: (1) “bloodstreaminfection*” or “blood streaminfection” or “sepsis” or “septic(a)emia*” or “bacter(a)emia*”; (2)“prediction model” or “diagnosticmodel” or “screening model” or“prediction score” or “screeningscore” or “diagnostic score” or“diagnostic markers” or “diagnostictool” or “clinical markers” or “clinicalsigns” or “clinical characteristics” or“predictors”; (3) “neonate” or“newborn” or “preterm” or “neonatal”or “NICU”; (4) “healthcare(-)associated” or “hospital(-)acquired”or “nosocomial or late(-)onset”.
Additional searches were performedby reviewing the bibliography of theretrieved full text articles and bya manual search of expert authors.
Study Selection Criteria
We included studies creating a predictionmodel for laboratory-confirmed
HABSI with clinical signs. Laboratory-confirmed HABSI or sepsis wasdefined as a positive culture of blood,cerebrospinal fluid, pleural fluid,and/or urine; culturing was at least48 hours after birth or admission.Target population was a consecutiveseries of hospitalized neonates withsuspicion of sepsis. It was requiredthat in these studies, relationshipsbetween clinical parameters andsepsis were assessed by univariateanalyses, whereas prediction modelswere developed by regressionmodeling; sensitivities, specificities,or ORs needed to be reported. Clinicalparameters under research must berecorded preceding or at time ofculturing. The final predictionmodel needed to include at least3 predictors; because neonatalsepsis is a complex clinical syndrome,prediction models based on2 possible parameters may not bejustified. A checklist with all inclusioncriteria was used to assess eligibilityof the studies, and when in doubt,issues for inclusion or exclusion werediscussed between the 2 independentresearchers.
The following items were collected:population characteristics, setting,methods, statistical methods,exclusion criteria, applied casedefinition, definition for suspectedHABSI/sepsis, total suspectedepisodes, total HABSI/sepsisepisodes, assessment time ofa suspected episode, clinicalparameter definitions, predictor/event ratio, and prediction modelaccuracy. Model diagnostic accuracywas assessed by pre- and posttestprobabilities, sensitivities,specificities, and positive andnegative likelihood ratios.
Quality Assessment
The Quality Assessment of DiagnosticAccuracy Studies–2 scale was used toevaluate the methodologic quality ofthe included studies.22 This validatedscale assesses criteria on 4 domains:patient selection (consecutive orrandom sampling, inappropriate
PEDIATRICS Volume 135, number 4, April 2015 e1003 by guest on July 27, 2018www.aappublications.org/newsDownloaded from
exclusions, description of includedpatients), index test (description,execution, blinded interpretation),reference standard (accuracy, blindedinterpretation), and flow and timing(application of and interval betweenreference standard and index test,exclusions for analysis). Because oflimitations in quality scales, wealso examined other individualcomponents of methodologicstandards for prediction models:clearly defined outcome andpredictive variables, description ofpatient characteristics, andcrossvalidation.23–25 In addition,overfitting, which is a substantialthreat in multivariate regressionmodeling, was assessed by calculationof the number of predictorsincluded in the regression model pertotal events; this is termed thepredictor/event ratio.
Statistical Methods
Comparisons between groupswere performed by using theMann-Whitney U test forcontinuous variables and theFischer exact test for categoricalvariables. The statistical package ofsocial science version 21 was usedfor these tests (SPSS Inc, Chicago,IL).
A random effects meta-analysis usingthe inverse-variance methodobtained ORs and 95% confidenceintervals (CIs) for clinical predictors.This was performed by using theComprehensive Meta-Analysisversion 2 program (Biostat Inc,Englewood, NJ). Statisticalheterogeneity was predefined byusing the Higgins I2 statistic (I2
#25% for low, 25% , I2 , 50% formoderate, and I2 $ 50% for high).Only those parameters that weresimilar in definition were pooled toavoid clinical heterogeneity. A 100%interobserver agreement (by E.V.,K.B., S.B.) concerning theseparameter definitions was requiredto be included in the random effectsmeta-analysis. Model diagnosticaccuracy variables were calculated
with a Web-based diagnostic testcalculator.26
Funnel plots will be constructed forassessment of publication bias whenat least 10 studies are included.
RESULTS
In total, 80 studies were retrieved ofwhich 9 were included, representing12 prediction models or scores, atotal of 1295 suspected and 434laboratory-confirmed HABSIepisodes. Figure 1 shows the resultsof the search strategy. The researchperiod spanned from 1993 to 2007with 3 western European,27–29
4 South Asian,30–33 1 Canadian,34 and1 Turkish study.35 All researcheswere performed in level III settingsof which one33 was conducted ina low-resource level III hospital.
Excluded Studies Not MeetingInclusion Criteria
One study6 was excluded because thetime frame of recorded clinicalparameters encompassed a 24-hourfollow-up postsepsis onset. Sixstudies developed a prediction modelincluding early-onset36–38 orcommunity-acquired episodes ofsepsis.39–41 In several studies byGriffin et al,42–44 heart ratecharacteristics and clinicalparameters are under study, andprediction models are developed.However, the studies of Griffin andcolleagues as well as the researchof Modi et al45 considered all NICUpatients rather than neonates withsuspicion of HABSI.
Methodologic Quality of the IncludedStudies
The methodologic performance ofthe included studies could beconsidered as low-medium risk forbias. Patient selection was mostlywell described, as was the motive forpatient exclusion. Selected patientswere all under suspicion of sepsis andunderwent the reference test(ie, blood, urine, or cerebrospinalfluid culturing) and an index test
(ie, clinical prediction score ormodel). Index tests were overall welldeveloped using similar statisticalmethods. The reference test bloodculturing is considered medium riskfor bias because it is known not to be100% accurate. Because bloodculture results are available only afterat least 12 hours and the index testwas performed before the bloodculture test, the studies can beconsidered as double blinded, exceptfor the 2 retrospective studies.31,35
Assessment of methodologicstandards for prediction modelsexhibited clearly defined outcome forall studies, moderate description ofpatient characteristics, andoccasionally clearly defined clinicalparameters. In particular, thedefinitions of the clinical signs andthe interobserver variability ininterpretation might be a mediumrisk for bias. Two studies29,33 had novalidation cohort but useda bootstrapping statistical techniquefor internal validation. Most studiesdefined the time of assessment as“day of sepsis workup,” so a timeframe for data collection ofa maximum of 24 hours precedingsepsis workup is acknowledged. The2006 retrospective study of Dalgicet al35 did not describe the applicabletime frame for data collection.Concerning the issue of overfitting,the 2005 study of Okascharoen et al31
overruled the generally accepted 1:10predictor/event ratio, so this mightbe a risk for bias. A visualpresentation rating the risk of bias inlow, medium, and high is presented inSupplemental Appendix 1.
General Description of the IncludedStudies
Characteristics of the 9 includedarticles are shown in Table 1. Pretestprobability of sepsis ranged between17% to 55%, indicating an importantvariation in study population. Threestudies did not include all neonateswith suspicion of sepsis in theirresearch; selection was made forneonates with gestational age
e1004 VERSTRAETE et al by guest on July 27, 2018www.aappublications.org/newsDownloaded from
,34 weeks29,33 and for birth weight$1000 g and #2500 g.32 One study35
did not report information on patientcharacteristics but did focus onnosocomial sepsis in neonates onneonatal intensive care. Two studieswere internal34 and externalvalidation studies28 of their formerdeveloped prediction score study.Another 2 studies32,33 are adaptedexternal validation studies of Singhet al.30
Characteristics and DiagnosticAccuracy of the Prediction Models
Characteristics and diagnosticaccuracy of the 12 prediction modelswith several cutoff scores aretabulated in Table 2. It is observedthat 3 models with a particular cutoffscore have a sensitivity of at least95%. Of these 3 models, Mahieuet al’s nosocomial sepsis score(NOSEP),27 with a point score of 8 orhigher, displays the highest specificityand positive likelihood ratio. Figure 2
is a visual comparison of thesensitivity and specificity of 7 models.If a model has several cutoff scores, itis represented with its highestsensitivity model.
Random Effects Meta-analysis
Odds ratios of clinical parametersin univariate analysis of 5studies27,29–31,33 were included in therandom effects meta-analysis. Ten ofthe 29 clinical signs and 8 of the22 risk factors within 2 to 5 studiescould be pooled. Eleven clinical signsand 12 risk factors were researchvariables in 1 study; another 6 clinicalsigns and 2 risk factors could not bepooled due to heterogeneity invariable definition. Definitions of thepooled and nonpooled clinical signsand risk factors are presented inSupplemental Appendix 2. Cutoffvalues of biological markers werelargely different, and data onlaboratory techniques used were notavailable, so biological markers were
not pooled. Pooled OR and statisticalheterogeneity of all 18 individualclinical parameters are tabulated inTable 3. Figure 3 displays the forestplots of the 5 significant clinicalparameters predictive for HABSI inneonates. Because 10 of 18 clinicalparameters exposed medium to highstatistical heterogeneity, a post hocanalysis was performed with 2studies,29,33 including solely lowgestational age neonates (,34weeks) and 9 clinical signs. PooledOR and statistical heterogeneity ofthese 9 signs are tabulated in Table 4.Figure 4 displays the forest plots ofthe 4 significant clinical signspredictive for HABSI in lowgestational age neonates (,34 weeks).
DISCUSSION
To our knowledge, this is thefirst systematic review of clinicalprediction models for HABSI inneonates. Three prediction models
FIGURE 1Summary of the literature search and study selection.
PEDIATRICS Volume 135, number 4, April 2015 e1005 by guest on July 27, 2018www.aappublications.org/newsDownloaded from
e1006 VERSTRAETE et al by guest on July 27, 2018www.aappublications.org/newsDownloaded from
TABLE2
Characteristicsof
12PredictionModels
Study(Year)
Model
Name
Model
Variables
Score
ModelApplication
Pre-TP
(%)
Sensitivity
(%)
Specificity
(%)
Post-TP+
Post-TP–
LR+
LR–
Area
Under
theRO
CCurve
(SE)
Mahieu27(2000)
NOSEP1,NO
SEP2
CRP$14
mg/L
5Score
0.82
(0.04)
Neutrophilfraction.50%
3$8
4195
4354
81.67
0.12
Temperature
.38.2°C
5$11
4160
8472
253.75
0.48
TPN$14
d6
$14
4126
100
100
349999
0.74
Thrombocytopenia,1503
103 /mL
5NO
SEP1+
positivehubandexitsite
cultures
Scoreof
$11+
positiveculture
4172
8779
185.50
0.32
0.86
(0.04)
Mahieu28(2002)
NOSEP-new-I
CRP$30
mg/L
5Score
5584
4264
321.45
0.38
0.71
(0.05)
Neutrophilfraction.63%
3$11
Temperature
.38.1°C
5TPN$15
d6
Thrombocytopenia,
1903
103 /mL
5NO
SEP-new-II
NOSEPnew-1+recent
surgery+
maternal
hypertension
+ventilation
Scoreof
$11
+3
risk
factors
5582
6775
252.48
0.27
0.82
(0.04)
Singh30(2003)
Weightedclinical
scorefordefinite
sepsis
Abdominal
distention
2Score
Nodata
Chestretraction
1$1
2987
2933
161.23
0.45
Grunting
2$2
2953
8052
192.65
0.59
Hypertherm
ia1
Increasedaspirates
1Lethargy
1Tachycardia
1Okascharoen31
(2005)
Bedsideprediction
score
Abnorm
albody
temperature
3Score
0.80
(nodata)
Hypotension
4$4
1782
7439
53.15
0.24
Neutrophilbandem
ia$1%
2$5
1770
8244
73.89
0.37
Respiratoryinsufficiency
2$6
1747
9671
1012.0
0.55
Thrombocytopenia,1503
103 /mL
2Um
bilical
line#7d
2Um
bilical
line.7d
4Dalgic35
(2006)
Clinical
Score
Abdominal
distention
2Score
Nodata
Bradycardia
26–12
3956
7155
281.93
0.62
Feedingintolerance
2Hypotension
2Lowest-highest
body
temperature
difference
2Respiratorysymptom
2Kudawla32
(2008)
Clinical
score
Abdominal
distention;chestretractions;
grunting;hypertherm
ia;lethargy;
tachycardia;prefeed
aspirates
Noscore
$1clinical
sign
2790
2330
141.17
0.43
Nodata
$2clinical
signs
2752
6536
211.49
0.74
Sepsisscreen
Absolute
neutrophilcount;CRP;
microerythrocytesedimentationrate;
neutrophilratio
(immature/total)
Noscore
$2markers
2748
7037
211.60
0.74
Nodata
PEDIATRICS Volume 135, number 4, April 2015 e1007 by guest on July 27, 2018www.aappublications.org/newsDownloaded from
have a sensitivity of at least 95% butlack further diagnostic accuracy.Lethargy and pallor and/or mottlingappear to be the best predictiveclinical signs, whereas the use oftotal parenteral nutrition (TPN),lipid infusion, and postnatalcorticosteroids are significantiatrogenic risk factors that shouldstrengthen suspicion for HABSI whenclinical signs occur.
The strengths of this study includemethodologic quality and diagnosticaccuracy evaluation of the modelsand a random effects meta-analysisdesign.
Diagnostic Accuracy of thePrediction Models
The variables of the includedprediction models were largelydifferent, so diagnostic test accuracymeta-analysis had no clinicalconnotation and was not executed.Considering diagnostic accuracy, pre-and posttest probability, sensitivity,specificity, and positive and negativelikelihood ratios were assessed. First,a large range in pretest probabilitywas detected, reflecting thefundamental variation in clinicalcondition of the study population andinfluencing the posttest probabilityresults. In addition, some modelsexhibit minor progress whencomparing pre- and posttestprobability (range of progress,3%–59%); major improvement(.50%) was noted for the NOSEPscore of 14 and the bedsideprediction score of 6, although thelatter experienced a low pretestprobability. Mahieu et al’s27,28
2 NOSEP score models, which havehigh pretest probability, have showngood pre- to posttest probabilityprogress, which might indicateeffectiveness. Second, diagnosticperformance is a major concern. It isobserved that only 3 models havesufficient sensitivity. A sensitivity of$95% is required for a potentiallylethal condition such as HABSI. TheNOSEP score of 8 of Mahieu et al,27
the clinical score (1 sign) and sepsisTABLE2
Continued
Study(Year)
Model
Name
Model
Variables
Score
ModelApplication
Pre-TP
(%)
Sensitivity
(%)
Specificity
(%)
Post-TP+
Post-TP–
LR+
LR–
Area
Under
theRO
CCurve
(SE)
Clinical
score+
sepsisscreen
Abdominal
distention;chestretractions;
grunting;hypertherm
ia;lethargy;
tachycardia;prefeed
aspirates;absolute
neutrophilcount;CRP;micro-erythrocyte
sedimentationrate;neutrophilratio
Noscore
$1clinical
sign
+27
9518
309
1.16
0.28
Nodata
$2markers
Rosenberg33
(2010)
Clinical
sepsisrisk
score
Apnea,hepatomegaly,jaundice,lethargy,
pallor
Noscore
$1clinical
signs
5477
5064
351.54
0.46
0.70
(0.04)
$2clinical
signs
5442
8273
452.33
0.71
Bekhof29
(2013)
Reducedmodel
for
proven
sepsis
Capillary
refill.2s,centrallineinpast24
h,increasedrespiratorysupport,lethargy
Noscore
1of
the4signs
2797
3736
31.54
0.08
0.83
(0.06)
CRP,C-reactiveprotein;
LR+or
LR2,likelihoodratio
ofapositiveor
negativetest;R
OC,receiveroperatingcharacteristic;TP,test
probability
(+or
–,ofapositiveor
negativetest).
e1008 VERSTRAETE et al by guest on July 27, 2018www.aappublications.org/newsDownloaded from
screen (2 biomarkers) of Kudawlaet al,32 and the reduced model(1 sign) of Bekhof et al29 exhibit 5%or fewer false-negative cases. Incontrast, the specificity of these 3models is low (range, 18%–43%); ofthese 3, the NOSEP score of 8 reportsthe highest specificity (43%),nonetheless indicating 33.7%false-positive cases (n = 35) and onlya 1.67 chance on a positive test forcases versus noncases. However,when clinical deterioration isa criterion for culturing, those33.7% false-positive cases mightbe considered true-positive casesfor another illness, justifyingantimicrobial therapy. In contrast,when a C-reactive protein of 1 mg/dLis a criteria for culturing and a NOSEPscore of 8 is detected, the clinical
condition of the neonate might notyet be taken into consideration; forexample, a neonate receiving TPN$14 days with a neutrophil fraction.50% has a NOSEP score of 9. Inthis case, the score can indicateclose monitoring of the neonateand increased or more frequentobservation of clinical signs.Clinically, antibiotic treatment mightbe postponed until clinicaldeterioration or a score of 11 isreached so that overtreatment can berepressed.46,47
Overall, prediction models havebeen developed to streamline theplethora of signs and risks for HABSIand thus to facilitate medicaljudgment and decision-makingconcerning treatment. In clinical
practice, the interpretation of thenonspecific clinical signs is pivotal,although not always obvious andsubject to interobserver variability.The consistency with other symptomsas well as underlying conditionsmust be considered. Important hereis the weight assigned to a clinicalobservation. It is not only thepresence of a clinical sign butprimarily a change in thepresentation of that clinical sign thatmight lead to accurate prediction ofHABSI.45,48,49 For example,premature neonates can have morerespiratory distress than full-termneonates, so signs such as apnea,chest retraction, and grunting mightbe less appropriate to predict HABSI.However, when a sign is defined as“an increase of” or “acute onset of,”the clinical relevance is emphasized.Suggestions for risk stratificationbased on setting, birth weight, orgestational age could also beconsidered in this context.50–53
Random Effects Meta-analysis
Lethargy and pallor/mottling aresignificant clinical signs predictive forHABSI (P , .050) and are observedwith a high statistical heterogeneityof $50%; post hoc analysis with lowgestational age neonates revealeda reduction to 0%, thus heterogeneitymight be caused by clinicaldifferences. Caution is warranted ininterpretation regarding post hocanalysis, but omitting the results ofOkascharoen et al,31 which showedthe strongest association for lethargyand pallor/mottling, can also beinterpreted as a sensitivity analysis.Therefore, it is likely that our results
FIGURE 2Paired forest plot of sensitivities and specificities of 7 models represented with their highest sensitivity model. FN, false-negative; FP, false-positive;TN, true-negative; TP, true-positive.
TABLE 3 Pooled ORs and Statistical Heterogeneity of Clinical Parameters Predictive for HealthCare–Associated Bloodstream Infections in All Studies
Predictor Pooled OR 95% CI P Heterogeneity (I 2), % Studies/ Neonates, n
PEDIATRICS Volume 135, number 4, April 2015 e1009 by guest on July 27, 2018www.aappublications.org/newsDownloaded from
are reliable. Lethargy is often foundto be associated with sepsis for allneonates, including low gestational
age or very low birth weightneonates.6,44,45,54,55 Regardingpallor/mottling, several studies
included pallor and/or mottling intheir research. One study,56 whichincluded all neonates, could not finda significant association betweenpallor and sepsis. Anothermulticenter study,45 which alsoincluded all neonates, clustered pallorand/or mottling under impairedperipheral perfusion. In this study,impaired peripheral perfusion wasa strong predictor for a positive bloodculture (OR 8.5, CI 5.4–13.2). In Ohlinet al’s52 study, pallor/mottling wasclustered with hypotension, anda significant association was found(OR 2.47, CI 1.48–4.14). Van den
FIGURE 3Forest plots of the 5 significant parameters predictive for health care–associated bloodstream infections in neonates. Neg, negative; Pos, positive.
TABLE 4 Pooled ORs and Statistical Heterogeneity of Clinical Sings Predictive for HealthCare–Associated Bloodstream Infection in the 2 Studies Including Exclusively LowGestational Age (,34 weeks)
Predictor Pooled OR 95% CI P Heterogeneity (I2), % Studies/ Neonates, n
e1010 VERSTRAETE et al by guest on July 27, 2018www.aappublications.org/newsDownloaded from
Bruel et al57 systematically reviewedclinical signs and their diagnosticvalue in ambulatory care settings forchildren aged 1 month to 18 years.They described impaired peripheralperfusion resulting in pallor/mottlingand lethargy/reduced consciousnessas red flags in the detection of seriousinfection. Additionally, convulsions, rapidbreathing, and cyanosis were stronglyassociated with serious infection.
Concerning risk factors, TPN, lipidinfusion, and postnatal corticosteroidadministration were significant afterpooling. TPN and lipid infusion aregenerally accepted as important riskfactors, particularly for HABSIscaused by coagulase-negativeStaphylococcus.50,58–62 Interestingly,in our analysis, very low birth weightwas not a significant predictor
despite being associated with HABSIin many studies.2,61,63 The pooledvery low birth weight OR involved2 studies, Okascharoen et al31
(unadjusted OR 9.1, CI 3.0–26.5) andMahieu et al27 (unadjusted OR 1.1,CI 0.5–2.4), representing 180 neonates,60 of whom had laboratory-confirmedHABSI. The lack of power might bea concern here.
Limitations and Recommendations
Although a systematic search wasdone by 2 independent researchers,incomplete retrieval of studies, as wellas publication bias and heterogeneity,might influence our results.
Concerning the power of the randomeffects meta-analysis, some pooledparameters are based on ,200suspected cases or ,100 laboratory-
confirmed cases. For future research,it might be interesting to include allstudies in which the objective was tofind clinical parameters predictive forHABSI or nosocomial sepsis byunivariate analysis. As such, power ofthe pooled results will increase.
In the past decade, much research hasconsidered heart rate characteristicsas an early diagnostic sign forneonatal sepsis.64–68 Including anindex of heart rate characteristics ina prediction model for neonatalsepsis seems promising.64,66,69,70
Differing characteristics of theneonatal populations in varioussettings are not always reported; thismay lead to concerns with externalvalidity of the prediction models. Theaddition of clinical parametersrelated to specific neonatal services
FIGURE 4Forest plots of 4 significant clinical signs predictive for health care–associated bloodstream infections in low gestational age (,34 weeks) neonates. Neg,negative; Pos, positive.
PEDIATRICS Volume 135, number 4, April 2015 e1011 by guest on July 27, 2018www.aappublications.org/newsDownloaded from
and subgroups is recommended forfuture work.
Furthermore, future models mightpotentially benefit from bedsidetechnology capable of quantifyingsubjective symptoms. For example,pallor/mottling could be replaced bycontinuous monitoring of peripheralmicrocirculation, and lethargy could bemeasured by continuous functionalmonitoring of brain activity.
CONCLUSIONS
A prediction model should beconsidered as guidance rather than
an absolute indicator because allmodels have limited diagnosticaccuracy. However, the use ofprediction models might decreasethe use of antimicrobial therapy. Inclinical practice, a NOSEP score of8 or more by Mahieu et al27 hasthe greatest potential but maynecessitate additional considerationof the patient’s clinical condition,depending on the criteria forculturing in a given setting.Furthermore, these findings suggestthat lethargy and pallor/mottling forall neonates and apnea and/orbradycardia and poor peripheral
perfusion for very low birth weightneonates are the most powerfulclinical signs in the prediction ofHABSI. Nonetheless, other clinicalsigns should not be discarded. Asstated earlier, the particular clinicalcontext should always be taken intoaccount.
ACKNOWLEDGMENT
The authors thank Dr EllenDeschepper for her biostatisticaladvice, particularly regardingdiagnostic accuracy.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Evelien Verstraete is supported by a Special Research Fund at Ghent University. Stijn Blot holds a research mandate of the Special Research Fund at Ghent
University.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
REFERENCES
1. Verstraete E, Boelens J, De Coen K, et al.Healthcare-associated bloodstreaminfections in a neonatal intensive careunit over a 20-year period (1992–2011):trends in incidence, pathogens, andmortality. Infect Control Hosp Epidemiol.2014;35(5):511–518
2. Stoll BJ, Hansen N, Fanaroff AA, et al.Late-onset sepsis in very low birthweight neonates: the experience of theNICHD Neonatal Research Network.Pediatrics. 2002;110(2 pt 1):285–291
3. Brodie SB, Sands KE, Gray JE, et al.Occurrence of nosocomial bloodstreaminfections in six neonatal intensive careunits. Pediatr Infect Dis J. 2000;19(1):56–65
4. Gupta N, Crockett DC, Anthony M,Webster DP. Late onset bloodstreaminfections in a tertiary neonatalintensive care unit. Arch Dis Child FetalNeonatal Ed. 2011;96(3):F234–F237
5. Mahieu LM, Buitenweg N, Beutels P, DeDooy JJ. Additional hospital stay andcharges due to hospital-acquiredinfections in a neonatal intensive careunit. J Hosp Infect. 2001;47(3):223–229
6. Fanaroff AA, Korones SB, Wright LL, et al;The National Institute of Child Health andHuman Development Neonatal ResearchNetwork. Incidence, presenting features,
risk factors and significance of lateonset septicemia in very low birthweight infants. Pediatr Infect Dis J. 1998;17(7):593–598
7. Stoll BJ, Hansen NI, Adams-Chapman I,et al; National Institute of Child Healthand Human Development NeonatalResearch Network. Neurodevelopmentaland growth impairment amongextremely low-birth-weight infants withneonatal infection. JAMA. 2004;292(19):2357–2365
8. Pessoa-Silva CL, Miyasaki CH, de AlmeidaMF, Kopelman BI, Raggio RL, Wey SB.Neonatal late-onset bloodstreaminfection: attributable mortality, excessof length of stay and risk factors. Eur JEpidemiol. 2001;17(8):715–720
9. Adams-Chapman I. Long-term impact ofinfection on the preterm neonate. SeminPerinatol. 2012;36(6):462–470
10. Stoll BJ, Hansen N. Infections in VLBWinfants: studies from the NICHD NeonatalResearch Network. Semin Perinatol.2003;27(4):293–301
11. Gray JW. A 7-year study of bloodstreaminfections in an English children’s hospital.Eur J Pediatr. 2004;163(9):530–535
12. Blot S, Cankurtaran M, Petrovic M, et al.Epidemiology and outcome of
nosocomial bloodstream infection inelderly critically ill patients:a comparison between middle-aged, old,and very old patients. Crit Care Med.2009;37(5):1634–1641
13. Gastmeier P, Behnke M, Breier AC, et al.Healthcare-associated infection rates:measuring and comparing. Experiencesfrom the German National NosocomialInfection Surveillance System (KISS) andfrom other surveillance systems [inGerman]. BundesgesundheitsblattGesundheitsforschungGesundheitsschutz. 2012;55(11–12):1363–1369
14. Paolucci M, Landini MP, Sambri V. Howcan the microbiologist help indiagnosing neonatal sepsis? Int JPediatr. 2012;2012:120139. Available at:http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3272815. Accessed February5, 2015
15. Pammi M, Flores A, Leeflang M,Versalovic J. Molecular assays in thediagnosis of neonatal sepsis:a systematic review and meta-analysis.Pediatrics. 2011;128(4). Available at:www.pediatrics.org/cgi/content/full/128/4/e973
16. Chiesa C, Panero A, Osborn JF, SimonettiAF, Pacifico L. Diagnosis of neonatal
e1012 VERSTRAETE et al by guest on July 27, 2018www.aappublications.org/newsDownloaded from
sepsis: a clinical and laboratorychallenge. Clin Chem. 2004;50(2):279–287
17. Fowlie PW, Schmidt B. Diagnostic testsfor bacterial infection from birth to 90days—a systematic review. Arch DisChild Fetal Neonatal Ed. 1998;78(2):F92–F98
18. Beekmann SE, Diekema DJ, Doern GV.Determining the clinical significance ofcoagulase-negative staphylococciisolated from blood cultures. InfectControl Hosp Epidemiol. 2005;26(6):559–566
19. Fischer JE, Ramser M, Altermatt S, NadalD, Waldvogel K. Rational utilization ofantibiotics in critically ill children [inGerman]. Schweiz Med Wochenschr.2000;130(42):1564–1571
20. Apisarnthanarak A, Holzmann-Pazgal G,Hamvas A, Olsen MA, Fraser VJ.Antimicrobial use and the influence ofinadequate empiric antimicrobialtherapy on the outcomes of nosocomialbloodstream infections in a neonatalintensive care unit. Infect Control HospEpidemiol. 2004;25(9):735–741
21. Gordon A, Jeffery HE. Antibiotic regimensfor suspected late onset sepsis innewborn infants. Cochrane DatabaseSyst Rev. 2005; (3):CD004501
23. Laupacis A, Sekar N, Stiell IG. Clinicalprediction rules. A review and suggestedmodifications of methodologicalstandards. JAMA. 1997;277(6):488–494
24. Altman DG, Vergouwe Y, Royston P, MoonsKG. Prognosis and prognostic research:validating a prognostic model. BMJ.2009;338:b605
25. Royston P, Moons KG, Altman DG,Vergouwe Y. Prognosis and prognosticresearch: Developing a prognosticmodel. BMJ. 2009;338:b604
26. Schwartz A. Diagnostic test calculator.2002–2006. Available at: http://araw.mede.uic.edu/cgi-bin/testcalc.pl.Accessed October 11, 2014
27. Mahieu LM, De Muynck AO, De Dooy JJ,Laroche SM, Van Acker KJ. Prediction ofnosocomial sepsis in neonates by meansof a computer-weighted bedside scoring
system (NOSEP score). Crit Care Med.2000;28(6):2026–2033
28. Mahieu LM, De Dooy JJ, Cossey VR, et al.Internal and external validation of theNOSEP prediction score for nosocomialsepsis in neonates. Crit Care Med. 2002;30(7):1459–1466
29. Bekhof J, Reitsma JB, Kok JH, VanStraaten IH. Clinical signs to identifylate-onset sepsis in preterm infants.Eur J Pediatr. 2013;172(4):501–508
30. Singh SA, Dutta S, Narang A. Predictiveclinical scores for diagnosis of late onsetneonatal septicemia. J Trop Pediatr.2003;49(4):235–239
31. Okascharoen C, Sirinavin S, ThakkinstianA, Kitayaporn D, Supapanachart S. Abedside prediction-scoring model forlate-onset neonatal sepsis. J Perinatol.2005;25(12):778–783
32. Kudawla M, Dutta S, Narang A. Validationof a clinical score for the diagnosis oflate onset neonatal septicemia in babiesweighing 1000–2500 g. J Trop Pediatr.2008;54(1):66–69
33. Rosenberg RE, Ahmed AS, Saha SK, et al.Nosocomial sepsis risk score forpreterm infants in low-resource settings.J Trop Pediatr. 2010;56(2):82–89
35. Dalgic N, Ergenekon E, Koc E, Atalay Y.NOSEP and clinical scores fornosocomial sepsis in a neonatalintensive care unit. J Trop Pediatr. 2006;52(3):226–227
37. Wang K, Bhandari V, Chepustanova S,et al. Which biomarkers reveal neonatalsepsis? PLoS ONE. 2013;8(12):e82700
38. Zeitoun AA, Gad SS, Attia FM, Abu MaziadAS, Bell EF. Evaluation of neutrophilicCD64, interleukin 10 and procalcitoninas diagnostic markers of early- andlate-onset neonatal sepsis. Scand JInfect Dis. 2010;42(4):299–305
39. Harrell FE Jr, Margolis PA, Gove S, et al;WHO/ARI Young Infant Multicentre StudyGroup. Development of a clinical
prediction model for an ordinal outcome:the World Health OrganizationMulticentre Study of Clinical Signs andEtiological agents of Pneumonia, Sepsisand Meningitis in Young Infants. StatMed. 1998;17(8):909–944
40. Bachur RG, Harper MB. Predictive modelfor serious bacterial infections amonginfants younger than 3 months of age.Pediatrics. 2001;108(2):311–316
41. Weber MW, Carlin JB, Gatchalian S,Lehmann D, Muhe L, Mulholland EK; WHOYoung Infants Study Group. Predictors ofneonatal sepsis in developing countries.Pediatr Infect Dis J. 2003;22(8):711–717
42. Griffin MP, O’Shea TM, Bissonette EA,Harrell FE Jr, Lake DE, Moorman JR.Abnormal heart rate characteristicspreceding neonatal sepsis andsepsis-like illness. Pediatr Res. 2003;53(6):920–926
43. Griffin MP, Lake DE, Moorman JR. Heartrate characteristics and laboratory testsin neonatal sepsis. Pediatrics. 2005;115(4):937–941
44. Griffin MP, Lake DE, O’Shea TM, MoormanJR. Heart rate characteristics andclinical signs in neonatal sepsis. PediatrRes. 2007;61(2):222–227
45. Modi N, Doré CJ, Saraswatula A, et al.A case definition for national andinternational neonatal bloodstreaminfection surveillance. Arch Dis ChildFetal Neonatal Ed. 2009;94(1):F8–F12
46. Yang YN, Tseng HI, Yang SN, Lu CC, ChenHL, Chen CJ. A strategy for reduction ofantibiotic use in new patients admittedto a neonatal intensive care unit. PediatrNeonatol. 2012;53(4):245–251
47. Patel SJ, Saiman L. Principles andstrategies of antimicrobial stewardshipin the neonatal intensive care unit.Semin Perinatol. 2012;36(6):431–436
48. Fischer JE, Harbarth S, Agthe AG, et al.Quantifying uncertainty: physicians’estimates of infection in critically illneonates and children. Clin Infect Dis.2004;38(10):1383–1390
49. Dewhurst CJ, Cooke RW, Turner MA.Clinician observation of physiologicaltrend monitoring to identify late-onsetsepsis in preterm infants. Acta Paediatr.2008;97(9):1187–1191
50. Holmes A, Doré CJ, Saraswatula A, et al.Risk factors and recommendations forrate stratification for surveillance of
PEDIATRICS Volume 135, number 4, April 2015 e1013 by guest on July 27, 2018www.aappublications.org/newsDownloaded from
51. Auriti C, Ronchetti MP, Pezzotti P, et al.Determinants of nosocomial infection in6 neonatal intensive care units: an Italianmulticenter prospective cohort study.Infect Control Hosp Epidemiol. 2010;31(9):926–933
52. Ohlin A, Björkqvist M, Montgomery SM,Schollin J. Clinical signs and CRP valuesassociated with blood culture results inneonates evaluated for suspectedsepsis. Acta Paediatr. 2010;99(11):1635–1640
53. Phillips P, Cortina-Borja M, Millar M,Gilbert R. Risk-adjusted surveillance ofhospital-acquired infections in neonatalintensive care units: a systematic review.J Hosp Infect. 2008;70(3):203–211
54. Bang AT, Bang RA, Reddy MH, et al.Simple clinical criteria to identify sepsisor pneumonia in neonates in thecommunity needing treatment orreferral. Pediatr Infect Dis J. 2005;24(4):335–341
55. Healy CM, Baker CJ, Palazzi DL, CampbellJR, Edwards MS. Distinguishing truecoagulase-negative Staphylococcusinfections from contaminants in theneonatal intensive care unit. J Perinatol.2013;33(1):52–58
56. Gonzalez BE, Mercado CK, Johnson L,Brodsky NL, Bhandari V. Early markers oflate-onset sepsis in premature neonates:clinical, hematological and cytokineprofile. J Perinat Med. 2003;31(1):60–68
57. Van den Bruel A, Haj-Hassan T, ThompsonM, Buntinx F, Mant D; European Research
Network on Recognising SeriousInfection investigators. Diagnostic valueof clinical features at presentation toidentify serious infection in children indeveloped countries: a systematicreview. Lancet. 2010;375(9717):834–845
58. Avila-Figueroa C, Goldmann DA,Richardson DK, Gray JE, Ferrari A,Freeman J. Intravenous lipid emulsionsare the major determinant of coagulase-negative staphylococcal bacteremia invery low birth weight newborns. PediatrInfect Dis J. 1998;17(1):10–17
59. Freeman J, Goldmann DA, Smith NE,Sidebottom DG, Epstein MF, Platt R.Association of intravenous lipid emulsionand coagulase-negative staphylococcalbacteremia in neonatal intensive careunits. N Engl J Med. 1990;323(5):301–308
60. Samanta S, Farrer K, Breathnach A,Heath PT. Risk factors for late onsetgram-negative infections: a case-controlstudy. Arch Dis Child Fetal Neonatal Ed.2011;96(1):F15–F18
61. Perlman SE, Saiman L, Larson EL. Riskfactors for late-onset health care-associated bloodstream infections inpatients in neonatal intensive care units.Am J Infect Control. 2007;35(3):177–182
62. Anderson-Berry A, Brinton B, Lyden E,Faix RG. Risk factors associated withdevelopment of persistent coagulase-negative staphylococci bacteremia in theneonate and associated short-term anddischarge morbidities. Neonatology.2011;99(1):23–31
63. Auriti C, Maccallini A, Di Liso G, DiCiommo V, Ronchetti MP, Orzalesi M. Riskfactors for nosocomial infections in
a neonatal intensive-care unit. J HospInfect. 2003;53(1):25–30
64. Lake DE, Fairchild KD, Moorman JR.Complex signals bioinformatics:evaluation of heart rate characteristicsmonitoring as a novel risk marker forneonatal sepsis. J Clin Monit Comput.2014;28(4):329–339
65. Griffin MP, Lake DE, Bissonette EA, HarrellFE Jr, O’Shea TM, Moorman JR. Heartrate characteristics: novelphysiomarkers to predict neonatalinfection and death. Pediatrics. 2005;116(5):1070–1074
66. Moorman JR, Lake DE, Griffin MP. Heartrate characteristics monitoring forneonatal sepsis. IEEE Trans Biomed Eng.2006;53(1):126–132
67. Kovatchev BP, Farhy LS, Cao H, Griffin MP,Lake DE, Moorman JR. Sampleasymmetry analysis of heart ratecharacteristics with application toneonatal sepsis and systemicinflammatory response syndrome.Pediatr Res. 2003;54(6):892–898
69. Goldstein B. Heart rate characteristics inneonatal sepsis: a promising test that isstill premature. Pediatrics. 2005;115(4):1070–1072
70. Fairchild KD. Predictive monitoring forearly detection of sepsis in neonatal ICUpatients. Curr Opin Pediatr. 2013;25(2):172–179
e1014 VERSTRAETE et al by guest on July 27, 2018www.aappublications.org/newsDownloaded from
DOI: 10.1542/peds.2014-3226 originally published online March 9, 2015; 2015;135;e1002Pediatrics
Evelien Hilde Verstraete, Koen Blot, Ludo Mahieu, Dirk Vogelaers and Stijn BlotAssociated Sepsis: A Meta-analysis−Prediction Models for Neonatal Health Care
ServicesUpdated Information &
http://pediatrics.aappublications.org/content/135/4/e1002including high resolution figures, can be found at:
Referenceshttp://pediatrics.aappublications.org/content/135/4/e1002#BIBLThis article cites 68 articles, 13 of which you can access for free at:
Subspecialty Collections
http://www.aappublications.org/cgi/collection/neonatology_subNeonatologysubhttp://www.aappublications.org/cgi/collection/fetus:newborn_infant_Fetus/Newborn Infanthttp://www.aappublications.org/cgi/collection/critical_care_subCritical Carefollowing collection(s): This article, along with others on similar topics, appears in the
Permissions & Licensing
http://www.aappublications.org/site/misc/Permissions.xhtmlin its entirety can be found online at: Information about reproducing this article in parts (figures, tables) or
Reprintshttp://www.aappublications.org/site/misc/reprints.xhtmlInformation about ordering reprints can be found online:
by guest on July 27, 2018www.aappublications.org/newsDownloaded from
DOI: 10.1542/peds.2014-3226 originally published online March 9, 2015; 2015;135;e1002Pediatrics
Evelien Hilde Verstraete, Koen Blot, Ludo Mahieu, Dirk Vogelaers and Stijn BlotAssociated Sepsis: A Meta-analysis−Prediction Models for Neonatal Health Care
http://pediatrics.aappublications.org/content/135/4/e1002located on the World Wide Web at:
The online version of this article, along with updated information and services, is
