University of Groningen

Disease transmission through hospital networksDonker, Tjibbe

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite fromit. Please check the document version below.

Document VersionPublisher's PDF, also known as Version of record

Publication date:2014

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):Donker, T. (2014). Disease transmission through hospital networks. Groningen: s.n.

CopyrightOther than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of theauthor(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons thenumber of authors shown on this cover page is limited to 10 maximum.

Download date: 24-09-2020

Chapter Six

Caring for the biggerpicture: Monitoring thenational spread of MRSAfrom a reference laboratory

perspective

Tjibbe Donker1,2,∗,Thijs Bosch2, Rolf Ypma2, Anja Haenen2, Marijnvan Ballegooijen2, Max Heck2, Leo Schouls2, Jacco Wallinga2, Hajo

Grundmann1,2

1 Department of Medical Microbiology, University Medical Centre Groningen, Groningen, The Netherlands2 Centre for Infectious Disease Control, National Institute for Public Health and the Environment, Bilthoven, The

Netherlands

Submitted

107

Monitoring the spread of MRSA

Abstract

Background In the Netherlands efforts to control methicillin-resistant Staphy-lococcus aureus (MRSA) in hospitals have been largely successful due to strin-gent screening of patients on admission and isolation of patients that meetthe criteria for being at high risk. However, Dutch hospitals are not free ofMRSA, and a considerable number of cases are coincidental findings: casesthat do not belong to any of the risk categories. For some cases, carriage canstill be the result from undetected nosocomial transmission, while the originof carriage for others remains unknown. Estimating the size and scale ofclusters of undetected nosocomial transmission is a public health issue andhas implications for future policies.Methods By combining routine data available for all MRSA isolates submittedbetween 2008-2011 to the Dutch Staphylococcal Reference Laboratory, weapplied a clustering algorithm that combines time and place of bacterialisolation and genetic distance of isolates to map the geo-temporal distributionof MRSA clonal lineages over the entire Dutch health care network.Results Of the 2966 isolates reported as coincidental findings, 579 werepart of geo-temporal clusters. The remaining 2387 were classified as MRSAof unknown origin (MUO). We observed marked differences in the propor-tion of isolates that belonged to geo-temporal clusters between specific MLVAclonal complexes, indicating lineage-specific transmissibility. The majority ofclustered isolates (74%) were reported from more than one laboratory.Conclusion The frequency of MRSA of unknown origin among patients withcoincidental MRSA is an indication of a largely undefined extra-institutionalbut genetically highly diverse reservoir. Efforts to understand the emergenceand spread of high risk clones require the pooling of routine epidemiologicalinformation and typing data into central databases to identify geo-temporalclusters of undetected nosocomial transmission.

108

6.1 Introduction

6.1 Introduction

Occurrence of MRSA in hospitals differs markedly between countries (EARS-Net,2011). The low levels in the Netherlands have conventionally been attributed tothe so-called Dutch search and destroy policy (Vandenbroucke-Grauls, 1996), whichstipulates that all patients who have had MRSA in the past or who are regardedat high risk of being colonised or infected are screened on admission and treatedin strict isolation until screening results become available(Infection PreventionWorking Party, 2012). Despite of these efforts, hospitals in the Netherlands arenot free of MRSA. Time and again, MRSA is isolated from hospitalised patientswithout obvious risk factors after being admitted for more than 2 days. This leadsto extensive screening of contact patients and possibly even hospital staff, whichcan be rather disruptive.

Patients with these coincidental MRSA findings pose another epidemiologicaland public health challenge. They either represent cases of primary introduction,which are regarded as MRSA of unknown origin (MUO) or are the result ofunobserved secondary spread. The first would be suggestive of a tip of the icebergwhereby the frequency with which MOUs are isolated in hospitals would be areflection of an undetermined extra-institutional reservoir, the second an indicationof hidden intra- or inter-institutional transmission chains.

Conventional hospital-based investigational epidemiology has its limitations asit can only link cases with an apparent epidemiological association, for instanceif patients had shared a room, whereas data available at national public healthinstitutes or reference laboratories contain too little information to make theseobvious epidemiological links. However, molecular typing data, and informationabout the location and time of isolation may provide sufficient detail to addresssome of the above mentioned challenges.

By combining data available at the Dutch staphylococcal reference laboratoryat the National Institute for Public Health and the Environment (RIVM) we heredescribe a genuine approach to map the distribution of MLVA types over time andspace and provide clues to the size and clonal composition of institutional andmulti-institutional MRSA clusters. We also determine the number and geneticdiversity of MUOs as an estimate of the frequency of primary hospital introductionsby patients without the obvious risk factors of MRSA carriage. Using the resultsof these analyses, we draw conclusions about the ability of the search and destroypolicy to control MRSA in hospitals using the current guidelines, and the existenceof unknown reservoirs outside of hospitals.

109

Monitoring the spread of MRSA

6.2 Methods

6.2.1 Data and algorithm

MRSA Surveillance

Primary MRSA isolates from patients admitted to Dutch hospitals are routinelysent to the National Institute for Public Health (RIVM) and the Environment by allmicrobiological laboratories in the Netherlands for reference typing. Each isolate isinvestigated by Multi Locus Variable Number of Tandem Repeat Analysis (MLVA),staphylococcal protein A (spa) typing and mecA and lukS lukF PCR as part of thestandard reference service. The hospitals are asked to complete a questionnaireabout epidemiological metadata for each isolate. This questionnaire includesquestions about the anatomical site of isolation, demographics of the patient, thereason for sampling, and if the patient belonged to one of the risk-groups definedby the Workgroup Infection Prevention (Infection Prevention Working Party, 2012).We used data collected between 2008 and 2011 (4 years), and included only theisolates for which information for the following variables was available: date ofisolation, MLVA type, postal code of patients’ residence, and hospital or health careinstitution. We excluded duplicate isolates from the same patient, same institutionand same year. We used typing data partitioned at the level of MLVA clonalcomplex (MC). Because of the large number of isolates belonging to MC398, andbecause the algorithm is computationally intensive, we only used data from 2009for the MC398 clonal complex.

Cluster Algorithm

To identify clusters of isolates with higher than expected geo-temporal occurrence inthe MRSA surveillance database, a non-parametric clustering algorithm proposedby Ypma et al (2013) was used. In short, this algorithm uses the ranked distancebetween any two samples for each of the variables, to calculate the combineddistance between them. Each isolate is assigned to its nearest neighbour, being theclosest isolate in the combined ranked distance. The resulting tree is subsequentlyrecursively split up by its weakest link, resulting in N − 1 possible clusters, whereN is the number of isolates. The combination of cluster size and the strengthof the weakest link in the cluster is then tested against 1000 random generatedtransmission trees to assess whether its weakest link was stronger than could beexpected for a cluster of the same size under random assumption.

Ranked distance

For each variable, the pairwise ranked distance is calculated. The distance betweenisolate A and B is defined as the number of isolates between them:

110

6.2 Methods

di(a, b) = |p : Di(a, p) ≤ Di(a, b) ∧Di(b, p) ≤ Di(b, a)| − 1

Where |.| denotes the number of elements of a set, Di the absolute distance inthis variable and the -1 ensures di(a, a) = 0. We obtain the full distance d betweentwo points a and b as the product of the data type specific distances:

d(a, b) = dgen(a, b) ∗ dgeo(a, b) ∗ dtime(a, b)

Permutation test

To assess the distribution of the weakest link-strength for each cluster size underrandom assumption, we performed a permutation test. For each of the variables,a random order of samples was chosen, creating random combinations of thevariables per isolate. Subsequently, all weakest link-cluster size combinations weredetermined as described above. For each cluster size, we selected the strongest linkout of the set of weakest links found for this permutation dataset. This processwas repeated on 1000 permutation datasets.

Distance measures

For each of the variables, a pairwise distance metric (Di) needs to be defined. Fordate and residence postal code the difference between the dates of isolation in days,and Euclidean distance between the centroids of the postal code areas were used.For the MLVA data, the number of VNTR loci difference between the isolates wasused (range 0-8).

As the distance measure between health-care institutions, we used the referraldistance. This is the shortest path between two hospitals through the nationalpatient referral network (Donker et al, 2010), formed by the exchange of patientsbetween hospitals. The shortest paths between all hospitals were determined usinga simulated patient referral network (See supporting information).

6.2.2 Epidemiological validation

We performed the integrated cluster algorithm using the following variables: iso-lation date, postal code of patients’ residence and MLVA type. In a subsequentanalysis, we replaced postal code of patients’ residence by position of health careinstitution within the Dutch referral network. For each MLVA clonal complex, werecorded the number of isolates that form geo-temporal clusters, the total numberof clusters, and the number of clusters that included isolates from more than onehospital.

In order to test the robustness of the algorithm, we assessed the influenceof spurious connections on the clustering by repeating the analysis on randomly

111

Monitoring the spread of MRSA

chosen subsets of the data. We selected 90% of the isolates and repeated thecluster algorithm and permutation test. This jack-knife process was repeated 1000times. The distribution of the percentage clustered isolates was compared with theobserved point-estimate, using the entire dataset.

Certain risk categories defined by the Dutch Working Group for InfectionPrevention (WIP, Infection Prevention Working Party (2012)) such as patientswho had contact with a known MRSA carrier are more likely associated withnosocomial transmission and therefore isolates from these patients are more likelyto from clusters. Conversely, if patients were screened because they had previouslybeen hospitalised outside the Netherlands, the isolates can be expected to beintroductions, and not likely to from a cluster. We compared the odds of an isolatebelonging to a cluster, stratified by WIP risk category, to test whether the resultscoincided with the expectations based on their epidemiological profile.

6.3 Results

6.3.1 MRSA Surveillance

Between 2008 and 2011, 14042 MRSA isolates were submitted to the NationalInstitute for Public Health and the Environment (RIVM for reference confirmationand typing. Of these, 2864 were duplicate isolates and 2226 lacked information aboutthe patient’s place of residence and were excluded from the analysis. Moleculartyping grouped 7720 (86.2%) isolates into five dominant MLVA clonal complexes(MC, Table 6.S2, Figure 6.1). The remaining 1232 isolates belonged to an additional11 MCs. The MC that contained most isolates (3852, 43.0%) was MC398 coincidingwith sequence type (ST)398 which represents livestock-associated MRSA in theNetherlands (de Neeling et al, 2007). Since the integrated cluster algorithm scalesexponentially with the number of isolates, inclusion of all MC398 isolates becamecomputationally too intensive and only MC398 isolates for 2009 were included intothe current study resulting in a total of 6295 isolates.

6.3.2 Clusters

Based on a novel algorithm that integrates time, place, and genetic distance, 1724isolates (27.4%) formed 155 significant clusters that were unlikely to have occurredby chance (Figure 6.2, Table 6.S3). Sixty-eight clusters (44%) included isolatesthat originated from more than one health care institution, whereby the largestcluster included isolates from 21 different hospitals, and 74% of all clustered isolateswere part of multi-institutional clusters. The proportion of isolates that formedclusters differed considerably between different MLVA clonal complexes. Only 4.4%of the MC398 isolates were part of a cluster in contrast to 46.6% in MC45 (Figure

112

6.3 Results

Figure 6.1: Isolates included in the study. Of the 14042 isolates submitted to thenational institute for public health and the environment, 5090 isolates were excluded (2864duplicates and 2226 lacking information). A further 2657 MC398 isolates were excludedto reduce computational time, resulting in a total of 6295 isolates used in the integratedalgorithm.

6.3). The dominant community and animal-associated MLVA complexes showedsignificantly lower proportions of clustered isolates than MLVA complexes typicallyassociated with HA-MRSA. Identified clusters appeared to be robust since observedpoint estimates were within the interquartile range of the jack-knife intervals forthe majority of dominant MLVA clonal complexes (Figure 6.3).

When using the position of the health care institutions within the nationalpatient referral network instead of postal code of patients’ residence, we identified1420 isolates (Table 6.S4) forming 142 clusters, whereby 47 clusters were multi-institutional. Clusters in this analysis largely overlapped with that using patients’postal code, i.e. 1120 isolates clustered in both analyses, and the proportions ofclustered isolates per MLVA clonal complex show a similar pattern in both analyses(Figure 6.S1).

113

Monitoring the spread of MRSA

Figure 6.2: Data sources for cluster detection in MRSA isolates in the Nether-lands 2008 to 2010, MLVA complex 8. A) The postal code of the patients’ homeaddresses, B) the MLVA cluster, here shown as a minimum spanning tree, and C) timeat which the sample was isolated, here shown in weekly aggregate numbers. Light graydenotes isolates outside clusters, blue, red, green, and black show the isolates present inthe four largest clusters. The other clustered isolates are shown in dark gray. Whereaseach of the sources shows weak clustering pattern, the three sources, taken together, revealconsiderably stronger clustering pattern.

6.3.3 Coincidental findings

Of the 6295 isolates, 2966 (47.1%) represented coincidental findings which consistedof isolates marked as unexpected findings and clinical isolates submitted directlyfrom laboratories unaccompanied by epidemiological information collected byinfection control nurses. Of these isolates, 579 (19.5%) were part of a cluster, whenusing the patient’s residence. When using the position of the health care institutionwithin the national patient referral network, 467 (15.7%) of these isolates wereidentified as part of a cluster. We could therefore not establish an epidemiologicalassociation for 2387 (37.9%) of all isolates which thus represent true MUOs.

114

6.3 Results

Figure 6.3: The proportion of isolates part of a cluster according to the algo-rithm, dots show the point estimate, box plots (median, IQR, 95% interval) include jack-knife estimates for 90% of the isolates. Common hospital-associated MRSA strains (MC5,MC45, MC22) show higher proportions of clustering than community-(MC8) or livestock-associated (MC398) MRSA.

6.3.4 Epidemiological validation

For 1757 (27.9%) isolates epidemiological metadata including the patient’s WIP riskcategories were known. The most frequent risk category was “contact with farmanimals” which coincided with the isolation of an MC398 strain. Among the otherMCs the most common risk categories were “admission from a foreign hospital” and“admitted to hospital with known MRSA problem” (See table 6.S2). The resultsfrom the cluster analysis were largely supported by this additional information.Of all isolates from patients who were admitted after hospitalisation abroad, only8% clustered, while 72% of the isolates from patient who shared a room withknown MRSA carriers clustered. Moreover, most isolates (59%) identified as part

115

Monitoring the spread of MRSA

of contact screening were part of a cluster (Figure 6.4). Also isolates containing thePanton-Valentine Leukocidin (PVL) gene, commonly associated with communityacquisition(Otter & French, 2010; Francis et al, 2005; Zetola & Francis, 2005)and mainly representing coincidental findings in our sample (60%), only clusteredoccasionally (Figure 6.5). MC8 harboured the largest proportion (40.5%) of thePVL-positive isolates. Among clustered MC8 isolates, only 12% were PVL-positiveversus 50% of those that did not cluster (p<0.0001).

Figure 6.4: The odds of belonging to a cluster of isolates differ considerablybetween epidemiological risk categories. The results of the clustering algorithmoverlap with expectations based on risk group. The coincidental findings more often falloutside the clusters, while isolates obtained as part of contact tracing during outbreaks,were more often found within clusters. Category 2b* is a health care worker (HCW)specific risk group.

Figure 6.5: The proportion of isolates positive for Panton-Valentine Leukocidin(PVL) is higher amongst cases outside the clusters. Black bars show clustered isolates,grey bars show non-clustered isolates. A) The proportion of PVL+ cases for the largestMLVA clonal complexes (MCs), and B) for isolates obtained as part of contact tracingduring outbreaks, those isolated within 48 hours after admission (community onset), andcoincidental. C) In absolute number of isolates, most PVL-positive isolates are foundamong coincidental findings outside clusters.

116

6.4 Discussion

6.4 Discussion

Stringent screening policies are implemented in the Netherlands for patients whobelong to risk categories. These risk categories identify many MRSA carrierson admission to Dutch hospitals. However, a considerable number of MRSAisolates submitted to the National Institute for Public Health and the Environment(RIVM) are isolated from patients who do not belong to any of the risk categoriesand represent coincidental findings. The corresponding isolates are often justmarked as unexpected findings, or submitted to the reference laboratory withoutaccompanying epidemiological information. Some of these coincidental isolates arethe result of unobserved transmission while others were introduced from hithertoundefined reservoirs. Isolates of the last group can be regarded as “MRSA ofUnknown Origin” (MUO) (Lekkerkerk et al, 2012).

Utilising an integrated clustering algorithm (Ypma et al, 2013), we were able tomap the distribution of MLVA types over time and space across the entire Dutchhealth care system, and thus determine the size and clonal composition of clustersof related isolates. This allowed us to infer what proportion of isolates belonging toany specific lineage is typically associated with intra- and inter-hospital transmissionand determine the number and genetic diversity of MUOs as an estimate of thefrequency of primary introductions and the size of extramural reservoirs.

The algorithm assigned 579 of the 2966 (19.5%) coincidental isolates to clusters,indicating a close epidemiological association, probably attributable to transmission.This finding illustrates that local hospital epidemiologists are unable to identifyall nosocomial transmission events especially if transmission chains extend overdifferent health care institutions. The remaining coincidental isolates, whichrepresented almost 40% of all included submissions, showed no epidemiologicalassociations and represented true MUOs. Their exact reservoir remains elusive,although the high proportion of PVL-positive isolates among the MUOs in MLVAclonal complex 8 (MC8) clearly suggests a community origin (Otter & French, 2010;Francis et al, 2005; Zetola & Francis, 2005).

The shortcomings in the epidemiological data are a reflection of how conventionalepidemiological methods rely heavily on data collected on-site by infection controlstaff. Although the data collection for a single case may appear rather manageable,the cumulative effort for all cases becomes daunting given the current constraints onhuman personnel and economic resources in health care institutions. This reducesthe sensitivity of local investigational epidemiology because cases will be missedand direct links disappear. The autonomous nature of the clustering algorithm istherefore one of its key advantages. The algorithm only requires the most basicinformation: time, place and genomic profile and can suggest epidemiologicalassociations in an unbiased manner.

We tested the results for epidemiological consistency using the available meta-

117

Monitoring the spread of MRSA

data for a subset of isolates and found a high correlation between the WIP riskgroups and the degree of clustering as suggested by the algorithm. Isolates thatwere generally considered to be introduced by patients after a period of hospitalisa-tion abroad were unlikely to form clusters (OR = 0.22), while isolates obtained aspart of contact tracing during outbreaks, were more often found within clusters(OR = 4.38). However, the sensitivity of clustering algorithms always depends onthe sampling density. If fewer isolates are collected, it will become harder for thealgorithm to distinguish outbreaks from the independent introductions, becausefewer isolates from the outbreak will be available. This would mainly apply toshort transmission chains, larger events will still be detected, because the chanceof finding two related cases increases with the number of cases in the chain. It istherefore possible that some of the isolates assigned as independent introductionswere still part of small, but largely unobserved, transmissions, which have beenmissed.

The algorithm tries to find clusters of cases without prior knowledge of thestructure of the data. Consequently, some of the links within a cluster may be weakif two of the three epidemiological parameters (location, time and genetic distance)suggest a non-random association. For instance, a cluster can include isolates basedon spatial and temporal proximity that for evolutionary reasons would be deemedmore peripheral. Given these potential shortcomings in specificity we expect thatat least 40% of all incident MRSA cases in Dutch health care institution between2008 and 2011 were MRSA of unknown origin. Intriguingly, more of these isolatescould be assigned to clusters when patients’ residential postal codes were used as aproxy for location rather than the position within the national referral networkof the patient’s health care institution. This may be an indication of communitytransmission and suggests that MUOs recovered from admitted patients indeedrepresent a sample of MRSA co-circulating among non-hospitalised individuals inDutch communities.

Using a cluster algorithm to identify clusters of related isolates delivers novelinsights in the dynamics of MRSA at national level that can aid the design ofnovel intervention strategies and guidelines. Firstly, many of the clusters connectmultiple health-care institutions, which would have remained undetected if onlythe isolates were typed locally and isolates not be made available to referencelaboratories. These multi-institutional clusters are likely caused by referred MRSA-positive patients between hospitals. It shows how the occurrence of nosocomialpathogens in hospitals cannot be viewed as events in single, independent units.Any intervention strategy or surveillance system adds value if it is organised atregional or national level.

Secondly, the proportion of clustered isolates reflects the propensity of a cloneto spread through the hospital population, as nosocomial transmission will result inclustered isolates. The isolates belonging to MC398, representing ST398, show very

118

6.4 Discussion

few clusters. These livestock-associated MRSA isolates are considered to spreadeasily within and between animal herds (Van Loo et al, 2007), but apparentlyonly spread through the human population on a very limited scale. Most ofthe MC398 isolates would therefore be the result of independent introductionspresumably from the animal reservoir. This same pattern is seen in MRSA strainsassociated with community acquisition. Novel intervention strategies will requirerapid identification at clonal level to take the differential transmissibility intoaccount and inform infection control measures on the basis of epidemic potentialof individual clonal lineages of MRSA.

The results also hold implications for the current guideline in the Netherlands.The search and destroy policy is able to prevent a significant part of the MRSAtransmission in Dutch hospitals by identifying patients at risk. In that way, thepolicy contributes to the low MRSA prevalence in the Dutch hospitals, and shouldtherefore be maintained. However, this will become an increasingly difficult task,as we expect the number of MUOs to rise. The suggested community reservoir ofthese MUOs calls for a more detailed risk analysis to improve the WIP categoriesand maintain the low MRSA prevalence in Dutch hospitals.

Acknowledgments

This study was funded by the Strategic Research Fund of the National Institute forPublic Health and the Environment (RIVM), the Dutch ministry of Health, Welfareand Sport, and the United Kingdom Clinical Research Collaboration (UKCRC)Translational Infection Research Initiative. The authors thank Mariano Ciccolinifor constructive comments and Laurens Zwakhals for useful help with respect tothe GIS analysis.

119

Monitoring the spread of MRSA

6.5 Appendix

Simulating the hospital network

In order to measure the distance between hospitals, we simulated the nationalpatient referral network of the Netherlands. The general structure of this network(Donker et al, 2010) is known, however, the exact position of each of the hospitalsis not known, because they are only known by their semi-anonymous identifier. Wetherefore calculated the pairwise distance between the coordinates of all hospitals,measured on the Dutch national grid. These distances (Dij) were then used tocalculate the link strength between hospitals:

Lij = wij/Dij

Where wij is a weight factor (See table 6.S1) to compensate for the disassortativemixing between hospital categories. The resulting metric is similar to the patientflow between hospitals, a high value indicates hospitals close together, a low valuehospitals further apart. It is, however, not a direct reflection of the number ofexchanged patients, but the resulting network has a similar structure to the originalmeasured one (Figure 6.S2).

Table 6.S1: To simulate the patient referral network, the geographical distance betweenhospitals was compensated by a weight depending on the types of hospitals, to accountfor the disassortative mixing between hospitals in the network (i.e. general hospitals tendto refer to teaching or university hospitals, not to other general hospital).

Gen

eral

Teac

hing

Uni

vers

ity

General hospital 1 3 5Teaching hospital 3 1 7University hospital 5 7 1

120

6.5 Appendix

Tab

le6.S2

:The

numbe

rof

isolates,

split

upby

MLV

Aclon

alcomplex

(MC)an

dep

idem

iologicalinformation,

such

astheris

kgrou

psde

fined

bytheWorkg

roup

InfectionPrevention(W

IP). All*

MC398*

MC5

MC8

MC45

MC22

MC30

MC621

MC80

MC1

MC482

MC2

MC88

MC435

MC7

MC2236

MC632

All

6295

1195

1416

1292

611

549

349

211

201

137

9063

5146

3127

26W

IPIn

trod

uctio

ns10

6450

715

812

249

8626

4116

207

124

39

22

WIP

Tran

smis

sion

522

1320

710

964

6913

71

82

202

12

04

Coi

ncid

enta

lcas

es29

6643

465

968

925

323

822

010

013

574

5410

2830

1514

13C

onta

ctTr

acin

g83

621

198

215

174

8138

2118

1611

187

33

75

Con

tact

with

Farm

Animals

543

497

128

74

44

31

20

00

01

0Adm

itted

from

foreignho

spita

l38

97

109

8737

8019

613

93

123

30

01

Ado

pted

child

112

229

252

22

310

90

01

09

00

>2mon

thsagoin

foreignho

spita

l22

111

23

01

00

11

00

00

11

Foreigndialysis

patie

nt3

02

00

00

00

01

00

00

00

Protected

contactwith

MRSA

carrier

212

46

17

00

00

01

00

00

0Unp

rotected

contactwith

MRSA

carrier

194

577

3430

2110

21

20

52

10

04

Adm

itted

toho

spita

lwith

know

nMRSA

problem

178

581

4814

182

10

51

30

00

00

Shared

room

with

MRSA

patie

nt15

24

5527

2223

14

01

112

00

20

0Kno

wnMRSA

carrier

149

2046

2911

1213

56

21

01

00

21

Kno

wnHCW

MRSA

carrier

50

02

11

00

00

00

01

00

0

121

Monitoring the spread of MRSA

Tab

le6.S3

:The

numbe

rof

isolates

assign

edto

aclusterusingthepo

stal

code

ofthepa

tientsreside

ntiala

ddress

aslocation

data,

split

upby

MLV

Aclon

alcomplex

(MC)an

dep

idem

iologicalinformation,

such

astherisk

grou

psde

fined

bytheWorkg

roup

Infection

Prevention(W

IP).

All*

MC398*

MC5

MC8

MC45

MC22

MC30

MC621

MC80

MC1

MC482

MC2

MC88

MC435

MC7

MC2236

MC632

All

1724

5257

135

628

519

335

4937

3736

2121

133

105

WIP

Intr

oduc

tions

9433

2312

89

13

01

11

00

01

1W

IPTr

ansm

issi

on34

21

145

7337

574

70

61

80

12

00

Coi

ncid

enta

lcas

es57

910

194

109

9455

916

2821

171

106

17

1C

onta

ctTr

acin

g49

11

139

122

113

5612

105

54

117

10

23

Con

tact

with

Farm

Animals

4633

53

30

02

00

00

00

00

0Adm

itted

from

foreignho

spita

l31

012

44

91

00

00

10

00

00

Ado

pted

child

110

45

00

01

01

00

00

00

0>2mon

thsagoin

foreignho

spita

l7

04

01

00

00

00

00

00

11

Foreigndialysis

patie

nt2

01

00

00

00

01

00

00

00

Protected

contactwith

MRSA

carrier

101

14

04

00

00

00

00

00

0Unp

rotected

contactwith

MRSA

carrier

115

049

2414

183

20

10

30

10

00

Adm

itted

toho

spita

lwith

know

nMRSA

problem

121

061

3110

131

10

40

00

00

00

Shared

room

with

MRSA

patie

nt11

00

4218

1522

04

01

15

00

20

0Kno

wnMRSA

carrier

311

154

52

01

10

10

10

00

0Kno

wnHCW

MRSA

carrier

20

02

00

00

00

00

00

00

0

122

6.5 Appendix

Tab

le6.S4

:The

numbe

rof

isolates

assign

edto

aclusterusingthepo

sitio

nof

thesend

ingho

spita

llab

oratoryin

thepa

tient

referral

netw

orkas

locatio

nda

ta,s

plit

upby

MLV

Aclon

alcomplex

(MC)an

dep

idem

iologicalinformation,

such

astheris

kgrou

psde

fined

bytheWorkg

roup

InfectionPrevention(W

IP).

All*

MC398*

MC5

MC8

MC45

MC22

MC30

MC621

MC80

MC1

MC482

MC2

MC88

MC435

MC7

MC2236

MC632

All

1420

446

231

726

516

239

2424

3528

1221

183

06

WIP

Intr

oduc

tions

681

169

1213

22

02

25

12

00

1W

IPTr

ansm

issi

on27

00

114

6023

525

60

51

10

12

00

Coi

ncid

enta

lcas

es46

71

161

9284

3612

818

1912

49

81

02

Con

tact

Trac

ing

528

212

312

912

149

105

53

40

71

00

3Con

tact

with

Farm

Animals

151

42

41

12

00

00

00

00

0Adm

itted

from

foreignho

spita

l46

010

48

121

00

12

51

20

00

Ado

pted

child

60

32

00

00

01

00

00

00

0>2mon

thsagoin

foreignho

spita

l2

00

10

00

00

00

00

00

01

Foreigndialysis

patie

nt1

01

00

00

00

00

00

00

00

Protected

contactwith

MRSA

carrier

110

14

06

00

00

00

00

00

0Unp

rotected

contactwith

MRSA

carrier

850

3718

716

31

01

01

01

00

3Adm

itted

toho

spita

lwith

know

nMRSA

problem

981

4726

89

21

04

00

00

00

0Sh

ared

room

with

MRSA

patie

nt90

036

1610

210

40

01

00

02

00

Kno

wnMRSA

carrier

220

122

52

00

00

00

10

00

0Kno

wnHCW

MRSA

carrier

20

02

00

00

00

00

00

00

0

123

Monitoring the spread of MRSA

Figure 6.S1: The proportion of isolates part of a cluster according to the algo-rithm. The results of the analysis using the postal code are shown in black, results fromthe analysis using the patient referral network in green. Using the network as distancemeasure generally delivers less and smaller clusters.

124

6.5 Appendix

Figure 6.S2: The structure of the patient referral network shown as a minimumspanning tree. A) The patient referral network constructed from the Dutch MedicalRegistry 2004. B) The network reconstructed using geographical distances betweenhospitals, weighted by a hospital category-specific factor. Red squares denote universityhospitals, blue triangles teaching hospitals, and black dots general hospitals. The structureof both networks is more or less the same, with groups of general hospitals around universityhospitals, and a back-bone structure consisting of mostly University and teaching hospitals.

125