www.elsevier.com/locate/jpedsurg

Journal of Pediatric Surgery (2011) 46, 801–807

CAPS Papers

Institutional practice and outcome variation in themanagement of congenital diaphragmatic hernia andgastroschisis in Canada: a report from the CanadianPediatric Surgery Network☆

Robert Baird a,⁎, Gareth Eesonb, Arash Safavi b, Pramod Puligandla a,Jean-Martin Laberge a, Erik D. Skarsgard b

Canadian Pediatric Surgery NetworkaDivision of Pediatric Surgery, The Montreal Children's Hospital, McGill University, Montreal, Quebec, CanadabDivision of Pediatric Surgery, BC Children's Hospital, the University of British Columbia, Vancouver,British Columbia, Canada

Received 24 January 2011; accepted 11 February 2011

0d

Key words:Gastroschisis;Congenital diaphragmatichernia;

Outcomes;Practice variability

AbstractBackground: Perinatal management of congenital diaphragmatic hernia (CDH) and gastroschisis (GS)remains nonstandardized and institution specific. This analysis describes practice and outcome variationacross a national network.Methods: A national, prospective, disease-specific database for CDH and GS was evaluated over 4years. Centers were evaluated individually and defined as low (low-volume center [LVC]) or high (high-volume center [HVC]) volume based on case mean.Results: Congenital diaphragmatic hernia. Two hundred fifteen liveborn cases were studied (mean,14.3 cases/center) across 15 centers (8 LVCs and 7 HVCs). Significant interinstitutional practicevariation was noted in rates of termination (0%-40%) and cesarean delivery (0%-61%). Centersdemonstrated marked variation in ventilation strategies, vasodilator and paralytic use, timing of surgery,and rates of primary closure. Overall survival was 81.4% (LVC, 76.9%; HVC, 82.4%; P = .43).

Gastroschisis. Four hundred sixteen cases were investigated (mean, 26 cases/center; range, 6-72)across 16 centers (10 LVCs and 6 HVCs). Cesarean delivery rates varied widely between centers(0%-86%) as did timing of closure (early vs delayed, 1%-100%). There was no difference in length ofstay, days on total parenteral nutrition, and overall survival (94.3% vs 97.2%; P = .17) between LVCsand HVCs.Conclusions: The existence of perinatal practice and outcome variation for GS and CDH suggeststargets for improved delivery of care and justifies efforts to standardize treatment on a national basis.© 2011 Elsevier Inc. All rights reserved.

☆ This work was supported by the Canadian Institute for Health Research: MOP-69050.⁎ Corresponding author. Tel.: +1 514 412 4400; fax: +1v 514 412 4289.E-mail address: robert.baird@mail.mcgill.ca (R. Baird).

022-3468/$ – see front matter © 2011 Elsevier Inc. All rights reserved.oi:10.1016/j.jpedsurg.2011.02.008

802 R. Baird et al.

The delivery of complex, multidisciplinary care to infantsborn with surgical birth defects such as congenitaldiaphragmatic hernia (CDH) and gastroschisis (GS) hasimproved, although the magnitude of improvement, asmeasured by standard outcomes, has been difficult toquantify. The persistent lack of standardization of care ineach of the multidisciplinary domains involved makes itextremely difficult, if not impossible, to identify carepractices, which are linked to “best” outcomes. This iscertainly true of most single-center outcome reports, wherecare patterns may have changed over the course of thereporting period, case numbers may be small, and an abilityto stratify patients into low and high risk according to anaccepted measure of illness severity may be lacking.

A significant obstacle in improving outcome through bestpractice change is to know which outcomes to study andwhich practices to target for change. Two essentialrequirements for quality improvement initiatives are interin-stitutional collaboration and a robust database that allowscollection of standardized institutional data related to risk,treatment, and outcome. There are several examples ofneonatal networks including the Australia and New ZealandNeonatal Network [1], Canadian Neonatal Network [2], andVermont Oxford Network [3] that have proven the efficacyof quality improvement initiatives, which are supported bycomprehensive, cohort-specific neonatal data sets. Althoughthe value of these networks in driving outcome improvementinitiatives within specific neonatal intensive care unit(NICU) cohorts (especially premature infants) is unchal-lenged, their specific use in evaluating infants with surgicallycorrectable birth defects is less clear [4].

The Canadian Pediatric Surgery Network (CAPSNet) wasestablished in 2005, and its observational database now hasmore than 4 years of population-based, multidisciplinarytreatment and short-term outcome data on Canadian infantsborn with CDH and GS. The purpose of this study was to usethe CAPSNet database to document the interinstitutionalvariability in treatment (beginning with prenatal diagnosisthrough to death or discharge from the hospital of birth) aswell as mortality and nonmortality outcomes of these 2 birthdefect cohorts, in preparation for the development of nationalquality improvement initiatives.

1. Patients and methods

The CAPSNet database has been described previously[5]. Briefly, this is an externally funded, prospective,disease-specific database for patients with CDH and GSfrom all 16 tertiary-level perinatal centers that collectivelyprovide care for infants from prenatal diagnosis (if one ismade) until death or hospital discharge. We analyzed allcases of CDH and GS entered into the database between May2005 and May 2009. Data were collected and entered on siteby trained abstractors before patient deidentification and data

centralization. Data collection at each center complied withprovincial privacy protection legislation and had siteinstitutional review board approval.

“Treatment” and outcome data for this study werereported by individual anonymized CAPSNet centers andsubsequently stratified by case volume into low-volumecenter (LVC) or high-volume center (HVC) based onwhether individual center volumes were below or abovethe network mean. Treatment was considered to include aprenatal phase (prenatal diagnostic workup); an intrapartumphase, including obstetrical planning and actual intrapar-tum management; and a condition-specific phase ofpostnatal treatment. Reported risk factors included maternal(age, smoking, and drug use for GS) and neonatal criteriasuch as gestational age, birth weight, and Score forNeonatal Acute Physiology (SNAP-II) [6], a validatedoutcome predictor for both CDH and GS [7,8]. Theoutcomes evaluated included mortality and survival out-comes (length of stay [LOS], duration of mechanicalventilation, duration of total parenteral nutrition [TPN],infectious and surgical complications requiring treatment,and comorbidities in survivors at discharge). Statisticalcomparisons of groups were performed where appropriateusing the Fisher exact test for dichotomous variables andStudent t test for continuous variables, with P b .05considered significant.

2. Results

2.1. Congenital diaphragmatic hernia

Over the 4-year study period (May 2005–May 2009),there were 215 liveborn cases of CDH across 15 studycenters. Variation in CDH management and outcome wascharacterized for individual centers and for groups of centersbased on case volume, which formed the basis forcomparison of centers in this study. The definition of LVCand HVC was based on whether the individual institution fellbelow or above the median case number of 9 for the entirestudy period. The overall mean case number per institutionwas 3.6 cases per year (1.2 cases per year for LVC and 6.3cases per year for HVC).

2.1.1. PrenatalComparable rates of prenatal diagnosis of CDH were

observed at LVCs and HVCs (64.3% vs 71.8%; P = .33)with a network average of 70.5%. There was no observeddifference in the rates of chromosomal analysis in fetuseswith a prenatal diagnosis of CDH between LVC and HVC(59.3% vs 54.8%; P = .67). The distribution of prenataloutcomes and live births per center is shown in Fig. 1.Elective termination rates at LVCs were 11.1% comparedwith 18.7% at HVCs; however, this difference was notstatistically significant (P = .34).

Fig. 1 Distribution of prenatal outcomes for CDH by CAPSNet center.

803Practice and outcome variation in management of CDH and GS

2.1.2. ObstetricalThe gestational age at delivery was fairly consistent

across institutions with a mean of 37.4 ± 1.6 weeks forLVCs and 37.8 ± 0.4 weeks for HVCs. The mode ofdelivery varied widely among institutions with cesareandelivery rates ranging from 0% to 60% (data not shown).Overall, the cesarean delivery rate was 30%, with nodifference observed between LVCs and HVCs (33% vs27%; P = .57). The timing of delivery of infants with anantenatal diagnosis differed between institutions, with ahigher proportion of overnight deliveries occurring atHVCs (18.8% vs 36.3%; P = .007). Low-volume centersand HVC were comparable with respect to gestational age,birth weight, sex, side of defect, and SNAP-II.

2.1.3. PostnatalConsiderable interinstitutional variability in the use of

various modes of ventilation was observed among individualcenters. No significant differences were noted betweenpatterns of use between LVCs and HVCs. Although high-frequency oscillation was used in 51.3% of cases in LVCscompared with 40.8% of cases in HVCs, this difference didnot, however, achieve statistical significance. The use ofvasodilators was relatively consistent across institutions,with 41.4% of infants receiving some form of vasodilator.Low-volume centers saw the use of vasodilators in 35% ofinfants compared with 42.6% of infants managed at HVCs (P= .38). Routine use of muscle relaxation, as indicated by theinitiation of paralytic agents at the time of NICU admissionin ventilated infants, demonstrated considerable variabilityacross institutions. The frequency of routine musclerelaxation varied between 0% and 83% across centers, witha network average of 46%. No difference was observed in the

rates of routine muscle relaxation between LVCs and HVCs(P = .11). Extracorporeal membrane oxygenation was usedin 4 of 15 centers, 3 of which were classified as HVCs,whereas 1 was considered an LVC. Overall, extracorporealmembranous oxygenation (ECMO) was used in only 7.4% ofCDH infants over the study period.

The timing of surgical closure of the diaphragmatic defectranged from days of life (DOL) 0 to 77 (mean DOL, 5.8 ±8.3; mode DOL, 2). Variability of mean time to surgicalclosure is demonstrated in Fig. 2. Although no differencewas observed in closure technique between LVCs andHVCs, marked variability was demonstrated betweenindividual centers (Fig. 3). Overall, 64% of infants wereclosed primarily, 30% received a patch, and in 6%, theclosure technique was not reported.

2.1.4. OutcomesThe overall survival of liveborn CDH cases across the

network was 81.3%, with an interinstitutional variationranging from 40% to 100% (Fig. 4). There was no observedsurvival difference between LVC and HVC. Althoughinterinstitutional variation existed between selected survivaloutcomes (LOS, days mechanical ventilation, incidence ofbronchopulmonary dysplasia (BPD), need for feeding tube,supplemental O2 at discharge), there were no significantdifferences in outcome between LVC and HVC (Table 1).

2.2. Gastroschisis

During the 4-year study period, 416 liveborn patients withGS were captured by the CAPSNet data set, with all 16participating centers providing patients. The mean casenumber per center was 26 (range, 6-72 cases) over the entire

Fig. 2 Timing of CDH repair by CAPSNet center.

804 R. Baird et al.

study period, which enabled stratification of centers into highor low volume relative to the network mean. One hundredforty-three patients were managed in LVC (mean, 3.6 casesper year), and 273 in HVCs (mean, 11.4 cases per year).

2.2.1. PrenatalOverall, a prenatal diagnosis of GS was established in

94% of cases, with institutions varying from 82% to 100%.When comparing LVCs and HVCs, there was no differencein maternal age (LVC, 22.3 years; HVC, 23.5 years) or druguse (LVC, 23.1%; HVC, 22.3%), although HVCs were morelikely to undertake chromosomal analysis (LVC, 76.2%;HVC, 87.2%; P b .01) and perform a detailed prenatalultrasound after the establishment of the diagnosis of GS.

2.2.2. ObstetricalGestational age (36.2 vs 36.1 weeks), birth weight (2.52

vs 2.54 kg), and SNAP-II scores (10.1 vs 8.5) werecomparable between LVC and HVC, respectively. Therewas marked variation in cesarean delivery rates for GSbetween centers as shown in Fig. 5, although no differencewas demonstrated between LVC and HVC. The median timeof birth was 1:37 PM ± 6.6 hours vs 12:50 PM ± 6.5 hours forLVC and HVC, respectively.

2.2.3. PostnatalThere was great interinstitution variation in the approach to

defect closure, with the frequency of urgent closure (within 6hours of NICU admission) ranging from 4.5% to 93% (Fig. 6).

Fig. 3 Type of CDH repa

Low-volume centers were noted to close the defect morerapidly than HVCs at both times points of less than 6 hours(65% vs 36.3%; P b .01) and less than 24 hours (77.2% vs48.8%; P b .01). There was also variation in the closurelocation. The practice of closure in the NICU ranged from 0%to 48% among institutions, but no difference between LVCand HVC was appreciated (14.9% vs 16.5%; P = .67).

2.2.4. OutcomesThe overall survival rate in the cohort was 96.2%, with no

difference between LVCs and HVCs (LVC, 94.3%; HVC,97.2%). Selected survival outcomes (LOS, TPN days, days tofull enteral nutrition, culture-proven bacteremic episodes, needfor feeding tube at discharge, severe cholestasis [conjugatedbilirubin N80 mmol/L]) at discharge are reported in Table 2.

3. Discussion

Pediatric surgeons are involved in the care of vastlydifferent cohorts of patients for whom improvement in thequality of care would have a substantial impact. One suchgroup is the population of infants born with surgicallycorrectable birth defects, of which CDH and GS are but 2examples. An inherent barrier to elucidating the influence ofsurgery on the outcome of infants with surgical birth defects isthe complex multidisciplinary nature of NICU care. Necro-tizing enterocolitis (NEC) is an illustrative example of a

ir by CAPSNet center.

Fig. 4 Congenital diaphragmatic hernia survival to discharge by CAPSNet center.

805Practice and outcome variation in management of CDH and GS

condition for which surgical treatment in isolation likely haslittle impact on overall outcome. Other critical outcomedeterminants that must be considered include patient riskfactors, disease factors, institutional and treating surgeon bias,and the unavoidable outcome impact that “nonsurgicaltreatments” (timing and quality of diagnosis, supportivecare, drug therapy, and timing of surgical intervention to namea few) have on overall outcome. These variables are virtuallyimpossible to control in the context of a randomizedcontrolled trial (RCT) targeting the evaluation of surgicaltreatment, which remains the preferred method to determine“best practice” in an evidence-based model of qualityimprovement. Two well-designed, adequately poweredmulticenter trials that have attempted to determine the bestsurgical management of perforated NEC both failed toprovide a definitive answer as to whether infants withperforated NEC should undergo peritoneal drainage orlaparotomy [9,10]. The logical explanations to the findingsof both studies are that either there is truly no outcomedifference attributable to surgical treatment or, perhaps morelikely, despite the best of intentions, it is extremely difficult, orperhaps impossible, to design a surgical trial that completelyeliminates uncontrolled biases, known or unknown.

An alternative approach to RCTs as the primary tool in thequality improvement “toolkit” is the use of large clinical datasets that contain variables for risk adjustment as well astreatment and outcome data. These “observational” data setscan be powerful tools, if carefully designed with fields that

Table 1 Selected outcomes in CDH infants in LVC and HVC

LVC HVC Significance

Total liveborn case volume 39 176 –Mean case number perinstitution

4.9 25.14 –

No. of institutions 8 7 –Duration of mechanicalventilation (d)

9.3 9.5 P = .20

LOS (d) 30.1 34.4 P = .60Respiratory support at 28 d 35.9% 38.1% P = .80Oxygen support at discharge 17.9% 24.4% P = .39Reflux medications at discharge 17.9% 24.4% P = .39Feeding tube at discharge 12.8% 26.1% P = .08Survival 71.4% 83.2% P = .43

allow accurate and discriminating characterization oftreatment (surgical and nonsurgical) and clinically relevantoutcomes. When sufficient data are collected from multiplecenters, it is possible to create risk-adjusted outcomeprediction models that establish “benchmarks” for outcomecomparison between centers. The identification of “outliers,”which are centers or individual practitioners whose actualoutcomes are at significant variance from predicted out-comes, are then targeted for improvement by site-drivenquality improvement projects. This has been the foundationalconcept underlying the National Surgical Quality Improve-ment project and the recently adapted pediatric module [11].

In addition to providing benchmarking of outcomes, thesedata sets can be used to document variation (ie, number ofoptions) in the treatment approach as well as the network“amplitude” of outcome variation. Currently, CAPSNetpublishes and distributes an annual report to all participatingcenters that reports case numbers, treatment, and keyoutcomes so that each center gets an annual performance“snapshot” relative to other centers and the network mean.The selection of outcomes for targeted national improvementwould be based on their clinical relevance and demonstratedvariation. For example, there would be relatively little to begained by quality improvement projects targeting mortalityreduction in GS because survival is already greater than 90%in most published reports [12]. The assessment of secondaryoutcomes (LOS and rate of liver dysfunction) allows fordisease-specific benchmarking, as opposed to survival alonethat has been suggested by others [13]. On the other hand,mortality reduction in CDH would be widely celebrated,given the marked variation in reported survival and the“room for improvement.”

The observation of large variations in treatment (eg,urgent vs silo closure for GS) usually reflects either a lack ofevidence to guide the choice of treatment or a stronginstitutional/practitioner bias (or both). Provided that patientscan be risk stratified, multivariate analysis of the effect oftreatment variables on outcome may suggest the clearsuperiority of 1 approach or may provide the foundationalevidence for a carefully designed RCT. An alternative toRCTs to guide best practice change is a methodologyreferred to as evidence-based practice improvement inquality (EPIQ) [14]. This approach uses a combination ofbest evidence (and where none exists, expert consensus) to

Fig. 5 Routes of delivery for GS by CAPSNet center.

806 R. Baird et al.

guide the selection of interventions that are intended to targeta specific outcome. Each practice change is introduced in acontrolled manner via “plan-do-study- act” cycles that allowpreinterventional and postinterventional measurement of theeffect of the intervention on outcome. The CanadianNeonatal Network has undertaken several EPIQ projects,resulting in a significant and sustained reduction in the ratesof nosocomial infection and bronchopulmonary dysplasiaamong premature infants in Canadian NICUs [15].

In the current study, we report population-based,condition-specific risk, treatment, and outcome data forCDH and GS and have identified significant variation acrossthe network for both conditions. Although the case numbersat individual centers are still too small to permit robustcomparison, significant variation between centers suggestspotential variables of interest. For example, for both CDHand GS, there is a highly variable approach to obstetrical

Fig. 6 Time to GS abdominal wa

management across Canada. We and others have attemptedto address the question of impact of the intended and actualdelivery on outcome for GS and CDH, without any clearbenefit shown to delivery by a planned vaginal route vscesarean delivery [16-18]. The type of abdominal wall repair(urgent vs delayed silo closure) used to treat GS is alsohighly variable across Canada, with some centers clearlyfavoring one approach over the other. Techniques ofcardiorespiratory stabilization before CDH repair and thetiming and type of repair are all highly variable acrossCanada. Indeed, there are variations in outcome for bothconditions across Canada. For CDH, there are variations inrisk-adjusted mortality that we had previously shown wasdirectly related to the volume of cases treated by individualcenters [19]. Perhaps, the most clinically important outcomefor GS (which is also highly variable in this study) is thepresence of severe cholestasis at discharge. This adverse

ll closure by CAPSNet center.

Table 2 Selected outcomes in GS infants in LVC and HVC

LVC HVC Significance

Mean case number perinstitution per year

3.6 11.4 –

LOS (d) 49.2 48.3 P = .16Days on TPN 36.3 40.1 P = .31Days to full enteral nutrition 13.4 13.2 P = .92Culture-proven bacteremicepisodes, n (%)

16(11.2)

43(15.8)

P = .24

Need for feeding tube atdischarge, n (%)

24(16.8)

54(19.8)

P = .51

Severe cholestasis, a n (%) 37(26.4)

44(16.1)

P = .02

a Defined as conjugated bilirubin greater than 80 mmol/L duringhospital admission.

807Practice and outcome variation in management of CDH and GS

outcome clearly identifies a group at risk for increasedmorbidity and mortality and is likely related to site-specificvariation in feeding practices, including the variable use ofnovel parenteral nutrition formulations and cycling lipidadministration. These examples of treatment and outcomevariation across a population-based network representopportunities for studies using EPIQ methodology and mayrepresent potential quality assurance benchmarks. Furthervalidation is required to use these parameters (cholestasis inGS, mortality for CDH, and mode of delivery for both) asstandards for hospitals caring for these patients.

In summary, the CAPSNet database has use, not only forfacilitating hypothesis-driven research projects but also forproviding national averages or benchmarks for the outcomesof specific birth defects. Where variations from the networknorm exist, individual centers can undertake local qualityaudits in an attempt to explain an unfavorable variance sothat it might be improved. Alternatively, quality improve-ment initiatives that incorporate evidence and consensus maythen be implemented to incite practice change with asubsequent study of its impact across the network to achievereal and sustainable outcome improvement for thesevulnerable infants.

Acknowledgment

The authors thank the Canadian Institute for HealthResearch team in maternal infant care and the Ontario Ministryof Health and Long-Term Care for their organizational support.

References

[1] Cust AE, Darlow BA, Donoghue DA, et al. Outcomes for high riskNew Zealand newborn infants in 1998-1999: a population based,national study. Arch Dis Child Fetal Neonatal Ed 2003;88:F15-22.

[2] Lee SK, McMillan DD, Ohlsson A, et al. Variations in practice andoutcomes in the Canadian NICU network: 1996-1997. Pediatrics2000;106:1070-9.

[3] Horbar JD, Rogowski J, Plsek PE, et al. Collaborative qualityimprovement for neonatal intensive care. NIC/Q Project Investigatorsof the Vermont Oxford Network. Pediatrics 2001;107:14-22.

[4] Skarsgard ED, Blair GK, Lee SK. Toward evidence-based bestpractices in neonatal surgical care-I: the Canadian NICU Network.J Pediatr Surg 2003;38:672-7.

[5] Skarsgard ED, Claydon J, Bouchard S, et al. Canadian PediatricSurgical Network: a population-based pediatric surgery network anddatabase for analyzing surgical birth defects. The first 100 cases ofgastroschisis. J Pediatr Surg 2008;43:30-4.

[6] Richardson DK, Corcoran JD, Escobar GJ, et al. SNAP-II andSNAPPE-II: simplified newborn illness severity and mortality riskscores. J Pediatr 2001;138:92-100.

[7] Skarsgard ED, MacNab YC, Qiu Z, et al. SNAP-II predicts mortalityamong infants with congenital diaphragmatic hernia. J Perinatol2005;25:315-9.

[8] Mills JA, Lin Y, Macnab YC, et al. Perinatal predictors of outcome ingastroschisis. J Perinatol 2010;30:809-13 [Epub ahead of print].

[9] Moss RL, Dimmitt RA, Barnhart DC, et al. Laparotomy versusperitoneal drainage for necrotizing enterocolitis and perforation.N Engl J Med 2006;354:2225-34.

[10] Rees CM, Eaton S, Kiely EM, et al. Peritoneal drainage or laparotomyfor neonatal bowel perforation? A randomized controlled trial. AnnSurg 2008;248:44-51.

[11] Dillon P, Hammermeister K, Morrato E, et al. Developing a NSQIPmodule to measure outcomes in children's surgical care: opportunityand challenge. Semin Pediatr Surg 2008;17:131-40.

[12] Arnold MA, Chang DC, Nabaweesi R, et al. Risk stratification of 4344patients with gastroschisis into simple and complex categories.J Pediatr Surg 2007;42:1520-5.

[13] Chang D, Salazar-Osuna J, Choo S, et al. Benchmarking the quality ofcare of infants with low-risk stratification using a novel riskstratification index. Surgery 2010;147:766-71.

[14] http://www.epiq.ca.[15] Lee SK, Aziz K, Singhal N, et al. Improving the quality of care for

infants: a cluster randomized controlled trial. CMAJ 2009;181:469-76.[16] Boutros J, Regier M, Skarsgard ED, et al. Is timing everything? The

influence of gestational age, birth weight, route, and intent of deliveryon outcome in gastroschisis. J Pediatr Surg 2009;44:912-7.

[17] Frenckner BP, Lally PA, Hintz SR, et al. Prenatal diagnosis ofcongenital diaphragmatic hernia: how should the babies be delivered?J Pediatr Surg 2007;42:1533-8.

[18] Safavi A, Lin Y, Skarsgard ED, et al. Perinatal management ofcongenital diaphragmatic hernia (CDH): when and how should babiesbe delivered? J Pediatr Surg 2010;45:2334-9.

[19] Grushka JR, Laberge JM, Puligandla P, et al. Effect of hospital casevolume on outcome in congenital diaphragmatic hernia: the experienceof the Canadian Pediatric Surgery Network. J Pediatr Surg 2009;44:873-6.