development and implementation of an early-onset sepsis ... jt... · of neonatal gbs disease...

The Joint Commission Journal on Quality and Patient Safety

Volume 42 Number 5May 2016232

One of the most common management decisions pedi-atricians face involves early-onset sepsis (EOS) Which

infants should be evaluated Which infants should receive empiric antibiotics If an infant is clinically ill the decision is straightforward as most pediatricians will treat empirically while they await the results of blood cultures or other diagnos-tic tests However in most cases the newborn is well appearing or only has minor physiologic disturbances (tachypnea tem-perature instability) In general the presence of risk factors such as group B Streptococcus (GBS) inadequate intrapartum anti-biotic prophyl axis (IAP) maternal chorioamnionitis and pro-longed rupture of membranes (ge 18 hours) has been used to decide which infants to evaluate with a blood culture andor treat empirically with antibiotics However making a rational decision to evaluate or treat an infant should depend on the probability of infection as well as the cost-benefit relationship between unnecessarily treating uninfected infants and delaying antibiotic treatment in infected infants

Widespread use of IAP has led to a sharp decline in EOS from GBS1ndash4 The Centers for Disease Control and Prevention (CDC) reports that the rate of EOS secondary to GBS has de-creased from 17 cases per 1000 births (1993) to 028 cases per 1000 births5 During this time the incidence of non-GBS EOS has declined as well67 Although the population risk has declined how do we determine an infantrsquos individual risk The CDC 2002 (and revised 2010) guidelines for the prevention of neonatal GBS disease provide algorithms for the evaluation of infants at risk for GBS EOS67 Presumably infants with a higher probability of EOS should have an evaluation or empir-ic treatment recommended but the algorithm neither gives an explicit risk nor specifies risk levels for evaluation or treatment Although simple to use the CDC algorithm has several draw-backs Information is lost as continuous variables are dichoto-mized (that is rupture of membranes ge 18 hours and gestational

age lt 37 weeks) Using chorioamnionitis a clinical and thus less reliable predictor is also problematic In the current era when internet access and smartphones are ubiquitous more sophis-ticated multivariate risk prediction models can be employed

Under current CDC guidelines the percentage of infants be-ing treated with antibiotics is ~200-fold higher than the inci-dence of EOS In 2008 and 2009 clinicians at Brigham and Womenrsquos Hospital (Boston) treated 8 of well-appearing in-fants born at ge 34 weeksrsquo gestation with antibiotics while the incidence of EOS among those infants was only 04 cases per 1000 live births8 Similarly an internal review of births in Kai-ser Permanente Northern California between 2010 and 2013 demonstrated that 15 of infants born ge 34 weeksrsquo gestation had blood cultures and 54 received antibiotics despite an incidence of EOS of only 03 cases per 1000 live births

Unnecessary evaluations and antibiotic treatment are not risk free In the short term obtaining blood cultures may result in infant discomfort and parental anxiety Given the low incidence of EOS false positive cultures will outnumber the true posi-tive and inevitably lead to more laboratory tests andor unneed-ed antibiotic exposure Infants receiving antibiotics are often admitted to a neonatal ICU for treatment this may interrupt parental bonding and breastfeeding Finally an emerging liter-ature suggests that early antibiotic exposure may be associated with diseases later in childhood Studies have shown an associa-tion between early antibiotic exposure and asthma9ndash16 allergicautoimmune disease17ndash21 and obesity22ndash25 Additional studies are needed to delineate the relationship between early antibiotic ex-posure on the neonatal microbiome and later disease Prudence should dictate a more conservative stance toward exposing large numbers of newborns to broad-spectrum antibiotics

Our goal was to implement an evidence-based approach for the evaluation and management of newborns suspected of EOS To achieve this goal we adapted two predictive models

Tool Tutorial

Development and Implementation of an Early-Onset Sepsis Calculator to Guide Antibiotic Management in Late Preterm and Term NeonatesReaders may submit Tool Tutorial inquiries and submissions to Steven Berman sbermanjcrinccom

Michael W Kuzniewicz MD MPH Eileen M Walsh RN MPH Sherian Li MS Allen Fischer MD Gabriel J Escobar MD

Copyright 2016 The Joint Commission


Volume 42 Number 5May 2016 233

developed by our research team for use in an integrated health care delivery system In this article we describe the strategy and the tools that we employed The central component of our ef-forts was to take advantage of new technologies not used by the CDC guidelinemdashWeb-based analytic tools smart phones and an administrative infrastructure for establishing consensus across a network of 14 delivery hospitals

Tool Development and Description Early-OnsEt sEpsis risk prEdictiOn MOdEl

In two recent reports2627 our team described the development and validation of two linked predictive models for EOS These models employ a Bayesian approach The first model establish-es a newbornrsquos baseline probability of EOS entirely on the basis of the maternal risk factors (or EOS risk at birth) EOS risk at birth is calculated using gestational age and maternal variables that are available at the moment of birth (highest antepartum temperature GBS carriage status duration of rupture of mem-branes and the nature and timing of IAP)

To maximize the reliability of the variables we favored ob-jective variables such as ldquohighest maternal antepartum tempera-turerdquo over more subjective variables such as ldquophysician diagnosis of chorioamnionitisrdquo We used nonparametric smoothing meth-ods to incorporate continuous variables Thus instead of dichot-omizing continuous variables as is done in the CDC algorithm the relationship of each variable to the outcome was analyzed separately and then combined into a multivariate model

Early-OnsEt sEpsis calculatOr

One difficulty with a multivariate risk prediction model is that it is not as simple to use as the CDC flow diagram Ex-pecting clinicians to manually calculate risk using regression model coefficients is unrealistic To facilitate clinical adoption we built Web-based calculators for desktop Web browsers and smartphones that allow clinicians to quickly enter patient-spe-cific data and calculate individualized risk of EOS28 Physi-cians deter mine sepsis risk using the calculator when admitting a patient and writing his or her history and physical (HampP) As had been done in the past under CDC guidelines nursing staff alert pediatric providers of risk factors (GBSndashpositive sta-tus maternal fever obstetriciansrsquo chorioamnionitis diagnoses) or clinical symptoms (tachypnea increased work of breathing fever) particularly in those instances in which a physician was not completing the HampP immediately In those instances pe-diatric providers calculate the initial sepsis risk and determine appropriate management

dissEMinatiOn Of Early-OnsEt sEpsis calculatOr

KPNC serves a population of 39 million members which constitutes nearly 40 of the insured population in Northern California Fourteen hospitals have obstetrical and neonatal ser-vices which care for approximately 35000 newborns each year Newborn care is provided by a combination of hospital-based pediatricians neonatal nurse practitioners and neonatologists Two of the facilities also have pediatric residents involved in the care of newborns The EOS calculator and an explanation of the risk prediction model were presented to the neonatology chiefs with a ldquosoft launchrdquo (no guidance was given to staff with respect to intervention thresholds) in December 2012 This strategy permitted staff to familiarize themselves with the calculator and probability of EOS at birth

Critical to the introduction of the EOS calculator was an ad-ministrative structure that has developed at KPNC during the past decade Three key venues for the introduction of initiatives are the Perinatal Research Unit (PRU) the Neonatology Chiefs Group and the electronic health record (EHR) (Kaiser Perma-nente HealthConnect [KPHC]) governance group Often an idea will be brought forward at the Chiefs group A request goes out to the regionwide journal club for a review of the literature and to the PRU for analysis of internal data After a consensus is built around a specific clinical practice those ideas go back to the journal club for further vetting and then back to the Chiefs for endorsement Whenever possible tools are built into the EHR to support and reinforce that new practice

incOrpOratiOn Of clinical prEsEntatiOn

The second predictive model for EOS that we developed quantifies how the baseline risk is modified by the infantrsquos clin-ical examination27 Using a variety of statistical and visualiza-tion techniques our team analyzed newborn data to define a hierarchical classification scheme for a newbornrsquos evolving clin-ical examination in the first 12 hours of life We defined three clinical-presentation categories (clinical illness equivocal and well appearing) and calculated likelihood ratios (LRs) for each Through a combination of recursive partitioning (classification and regression trees) and logistic regression we stratified infants on the basis of both their clinical presentation and the three lev-els of sepsis risk at birth (lt 065 cases1000 live births 065ndash154 cases 1000 live births ge 154 cases1000 live births) On the basis of the posterior probability of EOS in these strata and estimated numbers needed to treat (NNT) management choic-es were suggested (continued observation observe and evaluate and treat empirically) In August 2013 at a regional KPNC journal club for neonatal staff we formally introduced the EOS




calculator as well as thresholds for treatment based on this strat-ification scheme

prOblEMs with Our Original risk stratificatiOn Our original stratification strategy suffered from several lim-

itations Collapsing EOS risk at birth into three categories re-sults in significant information loss which is most pronounced in the highest-risk group Although the low (lt 065 cases1000 live births) and medium (065ndash1541000 live births) EOS risk at birth groups had relatively narrow ranges of risk a much broader range of risk is present in the high-risk (ge 154 cases1000 live births) group Therefore information loss was particularly problematic among well-appearing infants in the highest-risk group The mean posterior probability in this group was 674 cases1000 infants (NNT 148) However infants whose risk was close to the lower bound of the high-risk range would ac-tually have a posterior probability of lt 1 case 1000 live births (NNT ge 1000) when the LR for a normal exam is taken into account In addition the strata means were determined from the nested case-control study not a population-based sample Because the incidence of EOS is relatively rare the nested-case control study with three controls for case by design results in an EOSndashenriched cohort

Consequently we examined more recent data (KPNC 2010ndash2012 births) and calculated EOS risk at birth for all infants We found a left-skewed risk distribution (Figure 1 right) that is a large percentage of infants in our original high-risk stratum had a risk near the lower bound of the distribution Thus applying our initial stratification schema would result in the treatment of a large number of infants in the high-risk group who if well appearing would have a low individual risk

dEtErMining individual risk Of EOs (incOrpOrating thE clinical prEsEntatiOn)

To remedy this issue we utilized Bayesrsquos theorem (prior odds multiplied by the LR equals the posterior odds) to calculate in-dividual posterior probabilities for EOS risk for each infant The LR equals the probability of a particular finding in individ-uals with disease divided by the probability of a particular find-ing in individuals without disease We converted the probability of EOS risk at birth to odds to determine the prior odds (Prob-ability = odds [odds+1]) The finding of interest is the infantrsquos evolving clinical presentation after birth We used LRs from our prior work27 as follows

Clinical illness 145 (95 confidence interval [CI] 102ndash212)

Equivocal presentation 375 (95 CI 283ndash500)

Well appearing 036 (95 CI 031ndash041) We wanted to give clinicians a single value for EOS risk rath-

er than a range However because the LR estimate has an uncer-tainty bound we reasoned that using the point estimate of the LR could be dangerous (and lead to underestimation of an in-fantrsquos risk) To be conservative and estimate the maximum EOS risk we used the upper limit of the 95 confidence interval of the LR instead of the point estimate of the LR itself (2120 for clinical illness 500 for equivocal presentation and 041 for well-appearing presentation) Figure 2 (page 235) illustrates the effect of clinical presentation on the posterior probability of EOS for a given EOS risk at birth (prior probability) A well- appearing presentation will decrease the posterior probability an equivocal presentation increases the posterior probability and clinical illness greatly increases the posterior probability (and has the most dramatic effect)

sEtting trEatMEnt thrEshOlds

Now that we could determine an individualrsquos risk for EOS we needed to define management thresholds We decided to establish two one for continued observationblood culture and one for empiric antibioticsblood culture Unfortunately there are no randomized controlled trials to inform the probability of infection at which the benefits of starting antibiotics imme-diately exceed the risk and costs of delaying treatment It is important to remember that the treatment choice is not that of

Distribution of Early-Onset Sepsis (EOS) Risk at Birth in High-Risk Group Kaiser

Permanente Northern California 2010ndash2012

Figure 1 The calculated EOS risk at birth of infants in the original high-risk stratum had a risk near the lower bound of the distribution indicating a need to revise the original model

EOS Risk at Birth (Cases per 1000 Live Births)

Per

cent

age




antibiotics versus no antibiotics but rather between (a) starting antibiotics empirically (b) drawing a blood culture and wait-ing to start antibiotics if the infant becomes clinically ill or the blood culture becomes positive or (c) observing the infant and waiting to start antibiotics if the infant becomes clinically ill On September 4 2012 we convened a one-day symposium of local experts in the San Francisco Bay Area and queried the KPNC chiefs of neonatology to help determine a risk threshold for in-tervention We reached consensus for (a) posterior probability of EOS of ge 11000 live births (NNT29 1000) to obtain a blood culture monitor vital signs every 4 hours in the mother-baby unit and remain in the hospital until the culture was incubated for 24 hours and (b) a posterior probability of EOS of ge 31000 (NNT 333) to obtain a blood culture and start empiric antibi-otics We reasoned that these NNT were relatively conservative For a comparison in perinatal medicine the risk of uterine rup-ture complicating vaginal birth after cesarean section (VBAC) is ~1 leading to death or severe neurologic injury in the infant in approximately 10 of those births30ndash33 Thus the risk of death or severe neurologic injury after VBAC is about 11000 VBACs or a number needed to harm of 1000

rEvisiOns tO thE Early-OnsEt sEpsis calculatOr tO iMprOvE clinical accEptancE

In July 2014 we added several enhancements to the calcu-lator We modified some of the specifications in our original papers to improve face validity and clinician acceptance These modifications centered on the need to facilitate dissemination across 14 hospitals To broaden the ability to apply the calcula-

tor to different populations we added the ability to select the baseline prevalence of EOS in the population being managed Because the predictive model was developed using data from the nested-case control study its intercept term needs to be ad-justed to reflect the prevalence of EOS in the population to which it is being applied2634

The original EOS calculator was adjusted for an EOS prev-alence of 06 cases per 1000 live birthsmdashthe prevalence of the population from which the nested cases and controls were drawn8 However in KPNC the current (2010ndash2012) preva-lence of EOS was 03 cases per 1000 live births The revised cal-culator allows the user to select a baseline prevalence of 03ndash06 1000 live births in increments of 01

The original risk prediction model used broad-spectrum an-tibiotics lt 4 hours prior to delivery and GBSndashspecific antibi-otics any time prior to delivery as predictors that reduced the probability of EOS However even with the large study pop-ulation of 608014 there were not enough cases for antibiot-ic administration to be included in the model as a continuous variable Further antibiotic timing as displayed in the calcula-tor lacked face validity because clinical efficacy of antibiotics administered shortly before delivery was questioned Conse-quently we reviewed the IAP timing literature with respect to efficacy of preventing EOS Evidence exists that IAP had some effect even if administered as little as 30 minutes prior to de-livery but the effectiveness of IAP in preventing GBS if given ge 2 hours was ge 8935ndash38 Therefore we elected to be more conservative than the risk-prediction model by stipulating that antibiotic administration must occur at least 2 hours before de-livery which increased face validity and clinician acceptance In deciding on adequate IAP for GBS we instructed clinicians to include erythromycin and clindamycin as adequate prophylaxis only if GBS cultures had shown sensitivity to these antibiot-ics We also modified the clinical presentation descriptions and timing specifications Because infants often have transient respi-ratory distress transitioning to ex-utero life we explicitly state that the need for respiratory support needed to occur beyond the delivery room We added high-flow nasal cannula (which has become more widely used) to the list of respiratory support We expanded the study criterion of seizure to the more inclusive criterion of encephalopathy For the supplemental oxygen re-quirement we added a caveat that supplemental oxygen needed to maintain pulse oximetry of gt 90 for gt 2 hours to avoid mis-classification For the equivocal presentation we clarified time periods regarding vital signs abnormalities

We expanded the time frame to identify clinical illness or equivocal presentation from 12 hours to 24 hours which is

0001 0002 0005 001 002 003 005 007 008 009 010

Sepsis Risk Birth (Prior Probability) = 051000

PROBABILITY of SEPSIS

LR 041

LR 50

LR 212

Well Appearing

Equivocal

Clinical Illness

Effect of Clinical Presentation on Early-Onset Sepsis (EOS) Risk

Figure 2 The effect of likelihood ratios (LRs) for clinical presentation on the posterior probability of EOS is shown




consistent with a more conservative stance Table 1 (above) summarizes our modified criteria The most notable enhance-ment was adding posterior probabilities for each of the three clinical exam categories This provides the user with (a) the in-dividual probability for EOS incorporating both the EOS risk at birth and clinical presentation and (b) a clinical recommen-dation based on it The clinical recommendations play an im-portant purpose as we found that in the comment phase absent recommendations physicians often set their own thresholds for intervention (rather than the consensus thresholds endorsed by the neonatology chiefs) One of the goals of the EOS calcula-tor was to reduce practice variability so that the efficacy as well as the safety of the treatment thresholds can be evaluated Al-though physicians remain free to do what they wish it is much harder to stray from the recommendations if the calculatormdashwhose output is visible to nurses and other physiciansmdashgives a specific recommendation rather than just a number Prior to de-veloping the calculator we saw tremendous variability between centers with regard to antibiotic usage When a clinician en-ters in the maternal and infant data the calculator produces the risk of EOS and clinical recommendations for all three clinical presentations If an infant developed clinical symptoms several hours after birth the physician would apply the clinical recom-mendation based on the new clinical presentation status There is no need to reenter data into the calculator because it provides recommendations on all three clinical presentations

Also we introduced two additional safeguards into the cal-

culator Although the LR for clinical illness was high (21) it was theoretically possibly to have an EOS risk at birth so low such that when multiplied by 21 a posterior probability of lt 3 cases 1000 live births (clinical recommendation would be not to initi-ate antibiotics) would have resulted We did not want physicians to withhold antibiotics in an infant with clinical illness even if his or her posterior probability was below the consensus threshold Therefore the EOS calculator clinical recommendation for an infant with clinical illness and a posterior probability of lt 3 cases 1000 live births reads ldquoStrongly consider antibioticsrdquo

Finally we recognized that a septic infant might appear well at birth and then develop symptoms later Because a ma-jor component of the EOS calculator is clinical presentation and its effect on the posterior probability we wanted to ensure that infants with a higher EOS risk at birth (ge 1 case1000 live births) would not merely have their posterior probability re-duced by a reassuring exam shortly after birth The infant might then be placed in the newborn nursery under routine care and have the identification of clinical symptoms missed or delayed As a result we added an additional clinical recommendation on the basis of the EOS risk at birth If the EOS risk at birth was ge 1 case1000 live births the clinical recommendation is for vital signs every 4 hours for 24 hours to encourage more timely iden-tification of any clinical symptoms that might arise

Use of the Early-Onset Sepsis CalculatorIn December 2012 we made the EOS calculator available in both a computer-formatted website (httpwwwkporgEOScalc [available as of May 1 2016]) and a smartphone-formatted dis-play (httpwwwNewborn sepsiscalculatororg) The computer- formatted website is shown in Figure 3 (page 237 also available in color in online article) Key issues surrounding implementa-tion of the EOS calculator and their solutions are summarized in Table 2 (page 238)

The calculators are being used widely at KPNC The calcula-tor has been accessed by users in all 50 states with 52 of page visits by users in California and by users in 124 other countries Figure 4 (page 238) shows the monthly use of the desktop cal-culator The mobile application currently receives 1500 page views per month

Monitoring the Impact and Safety of the Early-Onset Sepsis CalculatorWe are actively monitoring the impact of the EOS calculator by following monthly rates of sepsis evaluations in the first 24 hours (defined by obtaining a blood culture) as well as antibi-otics administered in the first 24 hours (ascertained from the

Table 1 Classification of Clinical Presentation

Clinical illness 1 Persistent need for NCPAP HFNCmechanical ventilation (outside of the delivery room)

2 Hemodynamic instability requiring vasoactive drugs

3 Neonatal encephalopathyPerinatal depressionbull Seizurebull Apgar Score 5 minutes lt 5

4 Need for supplemental O2 gt 2 hours to maintain oxygen saturations gt 90 (outside of the delivery room)

Equivocal presentation

1 Persistent physiologic abnormality gt 4 hoursbull Tachycardia (HR gt 160)bull Tachypnea (RR gt 60)bull Temperature instability (gt 1004˚F or lt 975˚F)bull Respiratory distress (grunting flaring or

retracting) not requiring supplemental O2

2 Two or more physiologic abnormalities lasting for gt 2 hours

Well appearing 1 No persistent physiologic abnormalitiesNCPAP nasal continuous positive airway pressure HFNC high-flow nasal cannula HR heart rate RR respiratory rate




electronic medication administration record) Rates are deter-mined for KPNC as well as individual medical centers To en-sure that we are not missing infants with EOS with our lower rate of sepsis evaluations we are tracking any readmissions in the first week of life with a positive blood or cerebrospinal fluid (CSF) culture As an integrated system we can track all subse-quent laboratory evaluations and readmissions anywhere in the KPNC system In addition we review every EOS case to de-termine if antibiotics would have been started earlier under the CDC guidelines and if any adverse outcomes occurred such as meningitis need for mechanical ventilation need for inotropic agents and death We plan to publish data on utilization and safety after 18 months of implementation of the EOS calculator

Results and Lessons LearnedWe have developed a Web-based EOS calculator that provides perinatal practitioners with an explicit risk estimate The cal-culatorrsquos Bayesian structure reflects the actual flow of clinical experience while simultaneously providing the capability of up-dating a risk estimate By calculating explicit risk and using a guideline with specific intervention thresholds it is possible to better target the use of antibiotics in high-risk situations while forgoing their use when the risk is low Use of the calculator fits well with standard hospital work flow which has contributed to its broad adoption within KPNC The potential benefits of re-duced antibiotic exposure in low-risk situations is significant as emerging evidence has shown associations between early antibi-otic therapy and subsequent diseases such as asthma inflamma-tory bowel disease and childhood obesity2239ndash43

Reduction in sepsis evaluations and antibiotic usage will have to be weighed against other important clinical outcomes Hospital readmission for an infant with sepsis despite an as-sessment that the infant was ldquolow-riskrdquo is the obvious concern However this scenario is unlikely given that nearly all cases of EOS present with clinical illness before 24 hours and most be-fore 12 hours of age44ndash46 We have an active surveillance system EOS cases (if any) are reviewed on a monthly basis In addition because KPNC is an integrated system in which newborns are covered under their motherrsquos insurance for at least the first 30 days we are tracking any readmissions with a positive blood or CSF culture in the first seven days of life

An important element of the EOS calculator is that it pro-vides individualized risk estimates in conjunction with man-agement recommendations (as opposed to the CDCrsquos isolated management recommendations) First it provides physicians (and potentially parents) with an actual numeric risk value This is beneficial as one weights the benefit and risks of treat-

ment with nontreatment Treatment thresholds become more explicit and transparent enabling the physician to use the ac-tual risk value to explain to parents why treatment is or is not needed The EOS calculator also allows for a further Bayesian approach in select cases with use of the LR for the complete blood count4647

To implement the calculator modifications from the ini-tial research studies were necessary to increase face validity add additional safeguards and promote clinical adoption Our ad-ministrative structure supported the physician education and consensus building necessary to change clinical practice Links to the calculator and documentation templates in the EHR fur-ther support and standardize the new quantitative tool as it is adopted in an entire hospital network Anytime one strays from the specifications in a research study there is the possibility of not obtaining the identical results produced by the study How-ever a protocol that does not garner clinical acceptance will

Early-Onset Sepsis (EOS) Online Calculator

Figure 3 A computer-formatted screenshot of the EOS calculator website as shown at httpwwwkporgEOScalc (available as of May 1 2016)




never be put into practice Because we were departing from the established CDC algorithm it was particularly important that clinicians understood the reasoning behind the EOS calculator that the approach had clinical face validity and that additional safeguards were built into the approach The modifications we made were always more conservative than the original studies

We have demonstrated the translation of a relatively complex statistically sophisticated health services research tool into daily clinical practice The bedrock of any such effort is the quality of the research design and the subsequent data analysis Transla-tion to clinical practice in a large hospital network requires a ro-bust administrative infrastructure that facilitates organizational learning and consensus building Change in clinical practice is enforced by effective implementation of the EHR J

References1 Bizzarro MJ et al Seventy-five years of neonatal sepsis at Yale 1928-2003 Pediatrics 2005116595ndash6022 Moore MR Schrag SJ Schuchat A Effects of intrapartum antimicrobial prophylaxis for prevention of group-B-streptococcal disease on the incidence and ecology of early-onset neonatal sepsis Lancet Infect Dis 20033201ndash2133 Phares CR et al Epidemiology of invasive group B streptococcal disease in the United States 1999-2005 JAMA 2008 May 72992056ndash20654 Puopolo KM Eichenwald EC No change in the incidence of ampicillin- resistant neonatal early-onset sepsis over 18 years Pediatrics 2010125e1031ndash10385 Centers for Disease Control and Prevention Group B Strep (GBS) Clinical Overview (Updated Jun 1 2014) Accessed Mar 28 2016 httpwwwcdcgovgroupbstrepcliniciansclinical-overviewhtml Published 20156 Schrag S et al Prevention of perinatal group B streptococcal disease Revised guidelines from CDC MMWR Recomm Rep 2002 Aug 1651(RR-11)1ndash227 Verani JR McGee L Schrag SJ Division of Bacterial Diseases National Center for Immunization and Respiratory Diseases Centers for Disease Con-trol and Prevention Prevention of perinatal group B streptococcal disease Revised guidelines from CDC 2010 MMWR Recomm Rep 2010 Nov 19 59(RR-10)1ndash368 Mukhopadhyay S Eichenwald EC Puopolo KM Neonatal early-onset sep-sis evaluations among well-appearing infants Projected impact of changes in CDC GBS guidelines J Perinatol 201333198ndash2059 Droste JH et al Does the use of antibiotics in early childhood increase the risk of asthma and allergic disease Clin Exp Allergy 2000301547ndash155310 Muc M Padez C Pinto AM Exposure to paracetamol and antibiotics in early life and elevated risk of asthma in childhood Adv Exp Med Biol 2013 788393ndash40011 Murk W Risnes KR Bracken MB Prenatal or early-life exposure to anti-

Michael W Kuzniewicz MD MPH is Director Perinatal Research Unit Kaiser Permanente Division of Research Oakland California Eileen M Walsh RN MPH is Senior Research Project Manager and Sherian Li MS is Senior Data Analyst Kaiser Permanente Di-vision of Research Allen Fischer MD is Regional Director of Neo-natology and Gabriel J Escobar MD is Regional Director Hospital Operations Research Kaiser Permanente Northern California Oak-land Please address correspondence to Michael W Kuzniewicz MichaelWKuzniewiczkporg

Early-Onset Sepsis (EOS) Online Calculator Page Visits per Month

Figure 4 Page visits per month for the EOS Online Calculator are shown

Table 2 Key Issues in Incorporating Early-Onset Sepsis Calculator into Clinical Practice

Issue SolutionClinical acceptance of calculator

Avoided ldquoblack boxrdquo calculator

Extensive education on development of risk prediction model and components

Information loss due to recom mendations based on EOS risk in large infant subsets

Use of actual likelihood ratios to generate individual risk estimates

Limitations of clinical exam categories in original report

Modification made to original clinical categories

Safety concerns around uncertainty of risk estimates

Use of likelihood ratiosrsquo upper 95 confidence intervals for calculations to maximize risk estimates

Manual calculation of model coefficients impractical

Development of Web-based calculator

How to translate probabilities into discrete clinical actions

Recommendations embedded in calculator

Departing from CDC guideline

Educational component highlighting evidenced-based approach with analyses of more recent KPNC data

Prospective data collection strategy implemented that includes balancing measures (such as readmissions for EOS and complications due to delayed antibiotic treatment) tracking possible adverse outcomes

EOS early-onset sepsis CDC Centers for Disease Control and Prevention KPNC Kaiser Permanente Northern California




biotics and risk of childhood asthma A systematic review Pediatrics 2011 127 1125ndash113812 Ong MS Umetsu DT Mandl KD Consequences of antibiotics and in-fections in infancy Bugs drugs and wheezing Ann Allergy Asthma Immunol 2014112441ndash445e113 Risnes KR et al Antibiotic exposure by 6 months and asthma and allergy at 6 years Findings in a cohort of 1401 US children Am J Epidemiol 2011 Feb 1173310ndash31814 Sobko T et al Neonatal sepsis antibiotic therapy and later risk of asthma and allergy Paediatric Perinat Epidemiol 20102488ndash9215 Stensballe LG et al Use of antibiotics during pregnancy increases the risk of asthma in early childhood J Pediatr 2013162832ndash838e316 Sun W et al Early-life antibiotic use is associated with wheezing among children with high atopic risk A prospective European study J Asthma 2015 52647ndash65217 Tsakok T et al Does early life exposure to antibiotics increase the risk of eczema A systematic review Br J Dermatol 2013169983ndash99118 Schokker D et al Early-life environmental variation affects intestinal microbiota and immune development in new-born piglets PLoS one 2014 Jun 189e10004019 Maringrild K et al Antibiotic exposure and the development of coeliac disease A nationwide case-control study BMC Gastroenterol 2013 Jul 81310920 Li M Wang M Donovan SM Early development of the gut microbiome and immune-mediated childhood disorders Semin Reprod Med 20143274ndash8621 Arvonen M et al Repeated exposure to antibiotics in infancy A predis-posing factor for juvenile idiopathic arthritis or a sign of this grouprsquos greater susceptibility to infections J Rheumatol 201542521ndash52622 Trasande L et al Infant antibiotic exposures and early-life body mass Int J Obes (London) 20133716ndash2323 Murphy R et al Antibiotic treatment during infancy and increased body mass index in boys an international cross-sectional study Int J Obes (London) 2014381115ndash111924 Bailey LC et al Association of antibiotics in infancy with early childhood obesity JAMA Pediatr 20141681063ndash106925 Azad MB et al Infant antibiotic exposure and the development of childhood overweight and central adiposity Int J Obes (London) 2014381290ndash129826 Puopolo KM et al Estimating the probability of neonatal early-onset in-fection on the basis of maternal risk factors Pediatrics 2011128e1155ndash116327 Escobar GJ et al Stratification of risk of early-onset sepsis in newborns ge 34 weeksrsquo gestation Pediatrics 201413330ndash3628 Kaiser Permanente Probability of Neonatal Early-Onset Infection Based on Maternal Risk Factors 2015 Accessed Mar 28 2016 httpswwwdorkaiser orgexternalDORExternalresearchInfectionProbabilityCalculatorOLD aspx29 Walka MM et al Complete blood counts from umbilical cords of healthy term newborns by two automated cytometers Acta Haematol 1998100 167ndash173

30 Barger MK et al Maternal and newborn outcomes following uterine rup-ture among women without versus those with a prior cesarean J Matern Fetal Neonatal Med 201326183ndash18731 Barger MK et al Severe maternal and perinatal outcomes from uterine rupture among women at term with a trial of labor J Perinatol 201232837ndash84332 Ouzounian JG et al Vaginal birth after cesarean section Risk of uterine rupture with labor induction Am J Perinatol 201128593ndash59633 Signore C VBAC What does the evidence show Clin Obstet Gynecol 201255961ndash96834 Kelley CT Solving Nonlinear Equations with Newtonrsquos Method Philadel-phia Society for Industrial and Applied Mathematics 200335 Colombo DF et al Optimal timing of ampicillin administration to preg-nant women for establishing bactericidal levels in the prophylaxis of group B Streptococcus Am J Obstet Gynecol 2006194466ndash47036 de Cueto M et al Timing of intrapartum ampicillin and prevention of ver-tical transmission of group B Streptococcus Obstet Gynecol 199891112ndash11437 Lin FY et al The effectiveness of risk-based intrapartum chemoprophylaxis for the prevention of early-onset neonatal group B streptococcal disease Am J Obstet Gynecol 20011841204ndash121038 Illuzzi JL Bracken MB Duration of intrapartum prophylaxis for neo-natal group B streptococcal disease A systematic review Obstet Gynecol 20061081254ndash126539 Russell SL et al Early life antibiotic-driven changes in microbiota enhance susceptibility to allergic asthma EMBO Rep 2012 May 113440ndash44740 Kozyrskyj AL Ernst P Becker AB Increased risk of childhood asthma from antibiotic use in early life Chest 20071311753ndash175941 Kim BS Jeon YS Chun J Current status and future promise of the human microbiome Pediatr Gastroenterol Hepatol Nutr 20131671ndash7942 Hviid A Svanstroumlm H Frisch M Antibiotic use and inflammatory bowel diseases in childhood Gut 20116049ndash5443 Ajslev TA et al Childhood overweight after establishment of the gut mi-crobiota The role of delivery mode pre-pregnancy weight and early adminis-tration of antibiotics Int J Obes (London) 201135522ndash52944 Escobar GJ et al Neonatal sepsis workups in infants ge 2000 grams at birth A population-based study Pediatrics 2000106(2 Pt 1)256ndash26345 Ottolini MC et al Utility of complete blood count and blood culture screening to diagnose neonatal sepsis in the asymptomatic at risk newborn Pediatr Infect Dis J 200322430ndash43446 Bromberger P et al The influence of intrapartum antibiotics on the clinical spectrum of early-onset group B streptococcal infection in term infants Pediat-rics 2000106(2 Pt 1)244ndash25046 Newman TB et al Combining immature and total neutrophil counts to predict early onset sepsis in term and late preterm newborns Use of the IT2 Pediatr Infect Dis J 201433798ndash80247 Newman TB et al Interpreting complete blood counts soon after birth in newborns at risk for sepsis Pediatrics 2010126903ndash909




developed by our research team for use in an integrated health care delivery system In this article we describe the strategy and the tools that we employed The central component of our ef-forts was to take advantage of new technologies not used by the CDC guidelinemdashWeb-based analytic tools smart phones and an administrative infrastructure for establishing consensus across a network of 14 delivery hospitals

Tool Development and Description Early-OnsEt sEpsis risk prEdictiOn MOdEl

In two recent reports2627 our team described the development and validation of two linked predictive models for EOS These models employ a Bayesian approach The first model establish-es a newbornrsquos baseline probability of EOS entirely on the basis of the maternal risk factors (or EOS risk at birth) EOS risk at birth is calculated using gestational age and maternal variables that are available at the moment of birth (highest antepartum temperature GBS carriage status duration of rupture of mem-branes and the nature and timing of IAP)

To maximize the reliability of the variables we favored ob-jective variables such as ldquohighest maternal antepartum tempera-turerdquo over more subjective variables such as ldquophysician diagnosis of chorioamnionitisrdquo We used nonparametric smoothing meth-ods to incorporate continuous variables Thus instead of dichot-omizing continuous variables as is done in the CDC algorithm the relationship of each variable to the outcome was analyzed separately and then combined into a multivariate model

Early-OnsEt sEpsis calculatOr

One difficulty with a multivariate risk prediction model is that it is not as simple to use as the CDC flow diagram Ex-pecting clinicians to manually calculate risk using regression model coefficients is unrealistic To facilitate clinical adoption we built Web-based calculators for desktop Web browsers and smartphones that allow clinicians to quickly enter patient-spe-cific data and calculate individualized risk of EOS28 Physi-cians deter mine sepsis risk using the calculator when admitting a patient and writing his or her history and physical (HampP) As had been done in the past under CDC guidelines nursing staff alert pediatric providers of risk factors (GBSndashpositive sta-tus maternal fever obstetriciansrsquo chorioamnionitis diagnoses) or clinical symptoms (tachypnea increased work of breathing fever) particularly in those instances in which a physician was not completing the HampP immediately In those instances pe-diatric providers calculate the initial sepsis risk and determine appropriate management

dissEMinatiOn Of Early-OnsEt sEpsis calculatOr

KPNC serves a population of 39 million members which constitutes nearly 40 of the insured population in Northern California Fourteen hospitals have obstetrical and neonatal ser-vices which care for approximately 35000 newborns each year Newborn care is provided by a combination of hospital-based pediatricians neonatal nurse practitioners and neonatologists Two of the facilities also have pediatric residents involved in the care of newborns The EOS calculator and an explanation of the risk prediction model were presented to the neonatology chiefs with a ldquosoft launchrdquo (no guidance was given to staff with respect to intervention thresholds) in December 2012 This strategy permitted staff to familiarize themselves with the calculator and probability of EOS at birth

Critical to the introduction of the EOS calculator was an ad-ministrative structure that has developed at KPNC during the past decade Three key venues for the introduction of initiatives are the Perinatal Research Unit (PRU) the Neonatology Chiefs Group and the electronic health record (EHR) (Kaiser Perma-nente HealthConnect [KPHC]) governance group Often an idea will be brought forward at the Chiefs group A request goes out to the regionwide journal club for a review of the literature and to the PRU for analysis of internal data After a consensus is built around a specific clinical practice those ideas go back to the journal club for further vetting and then back to the Chiefs for endorsement Whenever possible tools are built into the EHR to support and reinforce that new practice

incOrpOratiOn Of clinical prEsEntatiOn

The second predictive model for EOS that we developed quantifies how the baseline risk is modified by the infantrsquos clin-ical examination27 Using a variety of statistical and visualiza-tion techniques our team analyzed newborn data to define a hierarchical classification scheme for a newbornrsquos evolving clin-ical examination in the first 12 hours of life We defined three clinical-presentation categories (clinical illness equivocal and well appearing) and calculated likelihood ratios (LRs) for each Through a combination of recursive partitioning (classification and regression trees) and logistic regression we stratified infants on the basis of both their clinical presentation and the three lev-els of sepsis risk at birth (lt 065 cases1000 live births 065ndash154 cases 1000 live births ge 154 cases1000 live births) On the basis of the posterior probability of EOS in these strata and estimated numbers needed to treat (NNT) management choic-es were suggested (continued observation observe and evaluate and treat empirically) In August 2013 at a regional KPNC journal club for neonatal staff we formally introduced the EOS




















Per

cent

age











0001 0002 0005 001 002 003 005 007 008 009 010



LR 041

LR 50

LR 212

Well Appearing

Equivocal

Clinical Illness





















































































Per

cent

age











0001 0002 0005 001 002 003 005 007 008 009 010



LR 041

LR 50

LR 212

Well Appearing

Equivocal

Clinical Illness












































































0001 0002 0005 001 002 003 005 007 008 009 010



LR 041

LR 50

LR 212

Well Appearing

Equivocal

Clinical Illness



































































development and implementation of an early-onset sepsis ... jt... · of neonatal gbs disease...

Documents