Advertisement

Source attributed case-control study of campylobacteriosis in New Zealand

Open AccessPublished:November 19, 2020DOI:https://doi.org/10.1016/j.ijid.2020.11.167

      Highlights

      • Reduction in human campylobacteriosis in New Zealand since 2008 has been relatively small.
      • A year-long source-attributed case-control study of notified human cases was conducted.
      • Most cases were infected with strains attributed to a poultry source using whole genome sequencing.
      • Risk factors related to the preparation and consumption of poultry were significant.

      Abstract

      Background

      Following an initial reduction in human campylobacteriosis in New Zealand after the implementation of poultry food chain-focused interventions during 2006–2008, further decline has been relatively small.
      We report a year-long study of notified campylobacteriosis cases, incorporating a case control study combined with a source attribution study. The purpose was to generate up-to-date evidence on the relative contributions of different sources of campylobacteriosis in New Zealand.

      Methods

      The study approach included:
      • A case-control study of notified cases (aged six months or more) sampled in a major urban centre (Auckland, every second case) and a mixed urban/rural area (Manawatū/Whanganui, every case), between 12 March 2018 and 11 March 2019.
      • Source attribution of human campylobacteriosis cases sampled from these two regions over the study period by modelling of multilocus sequence typing data of Campylobacter jejuni and C. coli isolates from faecal samples of notified human cases and relevant sources (poultry, cattle, sheep).

      Results

      Most cases (84%) were infected with strains attributed to a poultry source, while 14% were attributed to a cattle source. Approximately 90% of urban campylobacteriosis cases were attributed to poultry sources, compared to almost 75% of rural cases.
      Poultry consumption per se was not identified as a significant risk factor. However specific risk factors related to poultry meat preparation and consumption did result in statistically significantly elevated odds ratios.

      Conclusions

      The overall findings combining source attribution and analysis of specific risk factors indicate that poultry meat remains a dominant pathway for exposure and infection.

      Keywords

      Introduction

      Campylobacteriosis is the most commonly notified enteric disease in New Zealand. Following a reduction of approximately 50% in human campylobacteriosis after the implementation of poultry food chain-focused interventions during 2006–2008, further decline has been relatively small (
      • Pattis I.
      • Cressey P.
      • Lopez L.
      • Horn B.
      • Soboleva T.
      Annual report concerning foodborne disease in New Zealand 2018: Ministry for Primary Industries.
      ,
      • Sears A.
      • Baker M.G.
      • Wilson N.
      • Marshall J.
      • Muellner P.
      • Campbell D.M.
      • et al.
      Marked campylobacteriosis decline after interventions aimed at poultry, New Zealand.
      Sears et al., 2011). Sentinel source attribution work has demonstrated poultry to be the major reservoir for human infections, with ruminants also being an important reservoir (
      • Mullner P.
      • Jones G.
      • Noble A.
      • Spencer S.E.
      • Hathaway S.
      • French N.P.
      Source attribution of food-borne zoonoses in New Zealand: a modified Hald model.
      ;
      • Nohra A.
      • Grinberg A.
      • Midwinter A.C.
      • Marshall J.C.
      • Collins-Emerson J.M.
      • French N.P.
      Molecular epidemiology of Campylobacter coli strains isolated from different sources in New Zealand between 2005 and 2014.
      ). The most recent case-control study of campylobacteriosis in New Zealand was conducted in 1994–1995 (
      • Eberhart-Phillips J.
      • Walker N.
      • Garrett N.
      • Bell D.
      • Sinclair D.
      • Rainger W.
      • et al.
      Campylobacteriosis in New Zealand: results of a case control study.
      ).
      We report a year-long study of notified human cases of campylobacteriosis, incorporating a case control study combined with a source attribution study based on multilocus sequence typing (MLST) as determined by whole genome sequencing (WGS). The relative contributions of different reservoirs and exposure pathways to the burden of human campylobacteriosis was estimated (
      • Mughini Gras L.
      • Smid J.H.
      • Wagenaar J.A.
      • de Boer A.G.
      • Havelaar A.H.
      • Friesema L.H.
      • et al.
      Risk factors for campylobacteriosis of chicken, ruminant, and environmental origin: a combined case-control and source attribution analysis.
      ,
      • Mughini Gras L.
      • Smid J.H.
      • Wagenaar J.A.
      • Koene M.J.
      • Havelaar A.H.
      • Friesema I.H.
      • et al.
      Increased risk for Campylobacter jejuni and C. coli infection of pet origin in dog owners and evidence for genetic association between strains causing infection in humans and their pets.
      ,
      • Wagenaar J.A.
      • French N.P.
      • Havelaar A.H.
      Preventing Campylobacter at the source: why is it so difficult?.
      ). The study’s purpose was to generate up-to-date evidence on the relative contributions of different sources of campylobacteriosis in New Zealand to inform a refreshed New Zealand Food Safety risk management strategy and action plan.

      Methods

      The study approach included:
      • A case-control study of notified campylobacteriosis cases (aged six months and over) sampled in a major urban centre (Auckland, every second case) and a mixed urban/rural area (Manawatū/Whanganui, every case), between 12 March 2018 and 11 March 2019 (the study period). The study areas cover almost 40% of the New Zealand population of approximately five million.
      • Source attribution of human campylobacteriosis cases sampled from these two regions over the period. This was done by WGS of Campylobacter jejuni and C. coli isolates from faecal samples of notified human cases and from relevant sources (poultry, cattle, sheep), using modelling to assign a probable source for each case.
      Members of the New Zealand Health Survey (NZHS) cohort, who had given consent to be approached to participate in further studies, were the control population. The NZHS is an annual survey of individuals from approximately 14,000 randomly-selected households ().
      Eligible cases were those who:
      • Had a recent acute illness with symptoms of diarrhoea, abdominal pain and/or cramps, which may have been accompanied by fever and blood in the stool, or who had a medical reason for a clinical sample being submitted.
      • Had a Campylobacter jejuni or C. coli infection culture confirmed by the Institute of Environmental Science and Research (ESR) Public Health Laboratory (PHL).
      • Had not been detected by case finding in an enteric disease outbreak investigation.
      • Had completed a case interview.
      Cases were excluded from the study if, during interview, it was determined that the case:
      • Was asymptomatic.
      • Was unable to recall onset date of symptoms.
      • Had onset of symptoms(s) more than 14 days prior to date of notification.
      • Was not able to communicate (personally or through proxy if child) in English.
      • Was resident in an institution or equivalent (i.e. did not have access to a personal phone) at date of notification.
      • Had another episode of gastrointestinal illness in the month prior to the current illness.
      • Had travelled overseas during the incubation period (seven days prior to illness onset).
      From the NZHS cohort, controls were defined as a person who was resident in the two regions during the study period who:
      • Had not been notified with campylobacteriosis in the last 30 days.
      • Had not had a recent (within last 30 days) acute illness with symptoms of diarrhoea, abdominal pain and/or cramps, which may be accompanied by fever and blood in the stool.
      • Was able to communicate (personally or through proxy if child) in English.
      • Was not resident in an institution or equivalent (i.e. did not have access to a personal phone) at date of notification.
      • Was not part of an enteric disease outbreak already identified at date of selection.

      Case and control recruitment

      All cases of campylobacteriosis were contacted by telephone by public health staff. All eligible notified cases of campylobacteriosis (or their proxy reporters) within the regions were informed of the purpose of the study and its details. Cases were asked if they wished to contribute anonymised information to the study. If they gave verbal consent, a questionnaire was administered over the telephone by the public health employee. This gathered both routine public health investigation information and more detailed risk factor and exposure data (see questionnaire in Supplementary material). The pre-tested questionnaire covered consent, eligibility, demographics, illness, travel, activities, exposures including animal contact, drinking water and recreational water contact, food consumption, preparation and kitchen hygiene. Personal identifiers were collected at this stage, as the questionnaire responses fulfilled both public health and study purposes. Cases who chose not to participate in the study were asked to co-operate with routine public health investigation activities.
      Interviewers experienced in public health research recruited, by telephone, participants fulfilling the control criteria from the NZHS cohort and carried out interviews with them. Verbal consent was obtained using the same study description information as for cases. The same questionnaire, minus personal identifiers and current campylobacteriosis illness specific questions, was subjected to cognitive testing (in a trial eight respondents explained how they determined their responses), and then administered by telephone by trained interviewers.
      For the source-attributed case control study a 1:1 case:control ratio with 600 cases and 600 controls was planned and power-based sample sizes calculated for a range of attribution scenarios. If all cases were attributed to a single source this would provide 80% power to detect an odds ratio of ∼2·4 at a significance level of 0·05 for an exposure that occurred on 2·5% of the controls. Controls were frequency matched to cases for region and age based on the distribution of cases in the 2016 calendar year.
      Ethics approval for the study was obtained from the Northern B Health and Disabilities Ethics Committee (HDEC). Cultural consultation was provided by a senior Māori researcher.

      Risk factor modelling

      Using the entire dataset of cases and controls, logistic regression analysis was conducted for each risk factor in turn, adjusting for both sex and age; and odds ratios with 95% confidence intervals were calculated. A separate analysis was conducted for the 93 cattle associated cases, compared to the full set of controls. Variable selection methods including stepwise logistic regression, least absolute shrinkage and selection operator (LASSO) regression and random forests were applied to the case-control dataset to provide a ranking of risk factor importance. Variable importance, the risk factor screening and expert opinion on the epidemiology of campylobacteriosis were used to construct two multivariable models. The first model focused on risk factors associated with poultry meat consumption and included all cases, and the second on risk factors associated with the cattle source attributed cases. Adjusted population attributable fractions with bootstrapped 95% confidence intervals on 1,000 repeated samples were estimated to assess the effect of each risk factor on the probability of being a campylobacteriosis case (
      • Bruzzi P.
      • Green S.B.
      • Byar D.P.
      • Brinton L.A.
      • Schairer C.
      Estimating the population attributable risk for multiple risk factors using case-control data.
      ). All analyses were performed in R version 3.2.2 with packages glmnet (
      • Friedman J.
      • Hastie T.
      • Tibshirani R.
      Regularization paths for generalized linear models via coordinate descent.
      ), party (
      • Hothorn T.
      • Buhlmann P.
      • Dudoit S.
      • Molinaro A.
      • van der Laan M.J.
      Survival ensembles.
      ,
      • Strobl C.
      • Boulesteix A.L.
      • Zeileis A.
      • Hothorn T.
      Bias in random forest variable importance measures: illustrations, sources and a solution.
      ,
      • Strobl C.
      • Boulesteix A.L.
      • Kneib T.
      • Augustin T.
      • Zeileis A.
      Conditional variable importance for random forests.
      ) and randomForest (
      • Liaw A.
      • Wiener M.
      Classification and Regression by randomForest.
      ). Further details are given in the Supplementary material.

      Campylobacter isolates, whole genome sequencing, and source attribution analysis

      C. jejuni and C. coli isolates from notified human cases were obtained from samples (either swabs from culture plates or faecal subsamples) provided by the clinical diagnostic laboratories serving the two regions. Sample processing by the ESR PHL provided purified isolates with confirmed speciation.
      Concurrently with the 12-month case-control and human isolate parts of the study, a sampling programme was conducted to obtain Campylobacter spp. isolates to allow source assignment. Poultry and ruminant animals (dairy and beef cattle, sheep) samples or isolates were obtained monthly from processing plants delivering product to the same regions. Faecal samples were taken from cattle and sheep at the start of the processing line and sent to ESR PHL for processing, while poultry isolates were obtained from laboratories conducting statutory monitoring programmes for the industry.
      Isolates were obtained through standardised sample processing protocols. Purified and confirmed C. jejuni and C. coli isolates were extracted to provide DNA for WGS. Further details of the whole genome sequencing and bioinformatic analyses are provided in the Supplementary material.
      Core genome (cgMLST), whole genome (wgMLST), ribosomal (rMLST) and seven-gene MLST profiles were assigned based on WGS data. The cgMLST allele profiles were plotted as a Minimum Spanning Tree using the MSTree V2 algorithm in GrapeTree (
      • Zhou Z.
      • Alikhan N.F.
      • Sergeant M.J.
      • Luhmann N.
      • Vaz C.
      • Francisco A.P.
      • et al.
      GrapeTree: visualization of core genomic relationships among 100,000 bacterial pathogens.
      ). Source attribution for human cases was compared for different gene sets using the asymmetric island model via the R package islandR, for three main sources (poultry, cattle, sheep) (

      Marshall J. 2020. https://github.com/jmarshallnz/islandR. [Accessed 20 August 2019].

      ). Attribution using seven-gene MLST profiles was extended to include the covariates location (Auckland, Manawatū/Whanganui) and season, with each allowed to vary by rurality using the 2015 designation of a seven level scale from highly rural to highly urban areas based on the residential address of the case (). The addition of covariates was performed using the method of Liao (
      • Liao S.J.
      • Marshall J.
      • Hazelton M.L.
      • French N.P.
      Extending statistical models for source attribution of zoonotic diseases: a study of campylobacteriosis.
      ).
      Assignment of human cases to sources for source-attributed case-control analyses was done by attributing each human case to the most likely source, based on the posterior probability of source given the seven-gene MLST sequence type with attribution performed without covariates. Human cases were attributed to a source where that source was dominant in at least two thirds of all posterior source probabilities. Those cases for which the attribution results were mixed (i.e. the dominant source appeared in fewer than two thirds of samples) were excluded from source-attributed analyses.
      In addition to seven-gene MLST, a sensitivity analysis was constructed using 50-gene rMLST profiles and 13 genes from the core genome as identified in Thépault et al., with comparisons of the main attribution results and results from cattle-attributed case control models (
      • Thépault A.
      • Meric G.
      • Rivoal K.
      • Pascoe B.
      • Mageiros L.
      • Touzain F.
      • et al.
      Genome-wide identification of host-segregating epidemiological markers for source attribution in Campylobacter jejuni.
      ).
      Further details are included in the Supplementary material. The sequence reads generated from this study have been deposited in the NCBI Sequence Read Archive under BioProject accession number PRJNA675916.

      Role of the funding source

      Employees of the funder had a role in the study design, data interpretation, and writing of the report, and the decision to submit the paper for publication. All authors had final responsibility for the decision to submit for publication.

      Results

      Case and control recruitment, case and source isolates

      The number of cases who were interviewed, and for which a clinical isolate was obtained, was 666 (Auckland 445, Manawatū/Whanganui 221). As targeted, 600 controls were recruited and interviewed. The targeted age distribution for controls, based on notified cases in 2016, matched the age distribution of cases over the study period. Demographics of cases and controls, and urban rural assignments, are given in Supplementary material Tables S1 and S2, respectively.
      The target for each type of source sample over the study period was 150, reasonably spread across all quarters in the collection. At the end of the study the target of 150 was exceeded for all source types. Total isolates recovered across the study were poultry (209), sheep (194) and cattle (172).
      Most isolates from both humans and sources were C. jejuni (human: 638/666 (95·8%), cattle: 140/172 (81·4%), sheep: 126/194 (64·9%), poultry: 176/209 (84·2%)). The remainder were C. coli. The small proportion of C. coli isolates precluded separate analyses by species.

      Risk factor analyses

      Table 1shows the statistically significant (p < 0·05) adjusted odds ratios for risk factors estimated from the entire dataset (n = 666 cases and n = 600 controls). Poultry consumption per se was not identified as a significant risk factor. However specific risk factors related to the preparation and consumption of poultry did result in statistically significantly elevated odds ratios, e.g. consumption of undercooked chicken, consumption of chicken outside the home. Thus, the overall findings combining source attribution and an analysis of specific risk factors indicate that poultry meat remains a dominant pathway for exposure and infection.
      Table 1Univariate risk factor analysis, comparing all cases and controls in a source-attributed case-control study of campylobacteriosis in New Zealand.
      Odds ratios for questions/risk factors included in this table are statistically significantly greater than one (p<0.05). Odds ratios adjusted for both sex and age. Results for all questions/risk factors are given in Supplementary material Table S3.
      Question
      Odds ratios for questions/risk factors included in this table are statistically significantly greater than one (p<0.05). Odds ratios adjusted for both sex and age. Results for all questions/risk factors are given in Supplementary material Table S3.
      Cases (N = 666)Controls (N = 600)Adjusted odds ratio95% confidence interval
      No, number (%)Yes, number (%)No, number (%)Yes, number (%)
      Gastric treatment in 4 weeks prior556/664 (83·7)101/664 (15·2)540/600 (90)60/600 (10)1·84(1·29,2·65)
      Chronic Illness?619/663 (93·4)36/663 (5·4)590/600 (98·3)10/600 (1·7)3·43(1·73,7·44)
      Contact with someone with gastro symptoms491/665 (73·8)90/665 (13·5)539/600 (89·8)49/600 (8·2)1·96(1·35,2·87)
      Close contact with someone with gastro symptoms (i·e. same household)?519/654 (79·4)51/654 (7·8)565/600 (94·2)23/600 (3·8)2·28(1·38,3·87)
      Did you eat meat?37/666 (5·6)617/666 (92·6)57/600 (9·5)543/600 (90·5)1·75(1·13,2·75)
      Did you eat chicken which was pink on the inside?413/442 (93·4)29/442 (6·6)471/482 (97·7)11/482 (2·3)3·21(1·6,6·87)
      Did you eat poultry outside the home?390/652 (59·8)262/652 (40·2)412/600 (68·7)188/600 (31·3)1·57(1·23,2·00)
      Did you eat chicken from…
       Café/restaurant92/652 (14·1)62/600 (10·3)1·59(1·12,2·27)
       Catered event8/652 (1.2)1/600 (0·2)6·09(1·08,114·5)
       Somewhere else39/652 (6·0)14/600 (2·3)2·67(1·46,5·19)
      Did you cook beef using a grill85/595 (14·3)57/598 (9·5)1·62(1·14,2·33)
      Did you cook lamb using a grill42/666 (6·3)19/600 (3·2)2·12(1·22,3·79)
      Did you eat beef from a social BBQ16/652 (2·5)4/600 (0·7)3·22(1·16,11·36)
      Did you eat pork from somewhere else16/652 (2·5)5/600 (0·8)2·90(1·11,8·99)
      Did you drink raw milk?
      Milk27/661 (4·1)10/592 (1·7)2·11(1·02,4·69)
      Raw meat or poultry handled in household in the last 7 days…·
       Poultry meat234/666 (35·1)173/600 (28·8)1·59(1·23,2·06)
       Other meat211/666 (31·7)161/600 (26·8)1·38(1·07,1·80)
       Someone else handled poultry meat290/666 (43·5)202/600 (33·7)1·43(1·12,1·83)
      Did you handle kitchen items which had contact with raw meats or poultry?317/662 (47·9)306/662 (46·2)389/595 (65·4)200/595 (33·6)2·19(1·7,2·82)
      Did you handle the following which had contact with raw meats or poultry?
       Knives294/623 (47·2)173/589 (29·4)2·58(1·99,3·35)
       Chopping boards290/623 (46·6)177/589 (30·1)2·41(1·86,3·14)
       Dish cloths236/623 (37·9)102/589 (17·3)3·22(2·43,4·3)
      Do you live or work on a farm or lifestyle block?492/663 (74·2)170/663 (25·6)568/600 (94·7)32/600 (5·3)6·16(4·13,9·45)
      Home/work status?
       Live on a farm/lifestyle block496/664 (74·7)123/664 (18·5)568/600 (94·719/600 (3·2)7·29(4·49,12·49)
       Work on a farm/lifestyle block16/664 (2·4)1/600 (0·2)18·34(3·65,333·68)
       Live and work on a farm/lifestyle block23/664 (3·5)11/600 (1·8)2·52(1·21,5·55)
      Do you work directly with animals?62/531 (11·7)17/485 (3·5)3·71(2·12,6·81)
      Animals you have contact with at home or work…
       Dairy38/662 (5·7)11/600 (1·8)2·81(1·44,5·93)
       Meat production61/662 (9·2)20/600 (3·3)2·89(1·7,5·09)
       Poultry44/662 (6·7)17/600 (3·3)2·12(1·19,3·93)
       Other67/662 (10·1)17/600 (3·3)3·70(2·18,6·6)
      Are you in contact with the following dairy cows at home/work?34/662 (5·1)9/600 (1·5)3·02(1·47,6·86)
      Are you in contact with the following meat production animals at home/work?
       Cattle37/662 (5·6)11/600 (1·8)3·09(1·58,6·54)
       Sheep38/662 (5·7)11/600 (1·8)2·95(1·5,6·26)
      Are you in contact with domestic poultry at home/work?41/662 (6·2)16/600 (2·7)2·13(1·18,4·01)
      Did you have contact with…
       Chicken85/666 (12·8)21/600 (3·5)3·73(2·29,6·31)
       Cattle72/666 (10·8)18/600 (3·0)3·60(2·11,6·43)
       Sheep62/666 (9·3)18/600 (3·0)3·05(1·79,5·43)
      Did you have contact with pets at home?341/666 (51·2)245/600 (40·8)1·57(1·25,1·98)
      Did you have contact with chickens …
       At home64/666 (9·6)17/600 (2·8)3·44(2·01,6·19)
       Outside the home21/666 (3·2)4/600 (0·7)3·96(1·46,13·86)
      Did you have contact with cattle…
       At home48/666 (7·2)10/600 (1·7)4·24(2·17,9·12)
       Outside the home26/666 (3·9)8/600 (1·3)2·34(1·07,5·67)
      Did you have occupational contact with cattle?640/665 (96·2)25/665 (3·8)593/600 (98·8)7/600 (1·2)2·91(1·28,7·5)
      Did you have contact with sheep…
       At home44/666 (6·6)12/600 (2·0)3·32(1·77,6·69)
       Outside the home20/666 (3·0)6/600 (1·0)2·41(1·00,6·75)
      Did you have occupational contact with sheep?646/666 (97·0)20/666 (3·0)595/600 (99·2)5/600 (0·8)3·09(1·21,9·48)
      Did you have contact with other animals at home39/666 (5·9)19/500 (3·2)2·03(1·16, 3·66)
      Does the water to your home come from…
       Private bore/spring water28/666 (4·2)7/600 (1·2)3·33(1·5,8·46)
       Roof/rain water172/666 (25·8)43/600 (7·2)4·44(3·11,6·44)
      Did you drink water from the following untreated supplies you wouldn't normally?
       Roof/rain water36/642 (5·6)5/598 (0·8)6·62(2·8,19·52)
      a Odds ratios for questions/risk factors included in this table are statistically significantly greater than one (p < 0.05). Odds ratios adjusted for both sex and age. Results for all questions/risk factors are given in Supplementary material Table S3.
      Direct exposures to live poultry with significantly elevated odds ratios included: contact with domestic poultry at home or work (reported by 6% of cases), contact with chickens at home (9% of cases) or outside the home (3% of cases). Thus, while direct contact with poultry presents a significantly elevated risk, it affects only a small proportion of the poultry attributed cases.
      The complete risk factor results are available in Supplementary material Table S3. A separate risk factor analysis using only poultry associated cases provided very similar results to that for all cases.
      The importance of non-poultry pathways was explored using the 93 cattle attributed cases, for whom these other pathways were expected to be important. Table 2compares the adjusted odds ratios for cattle associated cases (n = 93) with all controls, for variables identified as significant in both the full dataset and the cattle-associated dataset. The cattle associated adjusted ORs in bold are at least twice those in the all-case dataset. The results indicate that living or working on a farm is the most important risk for these cases.
      Table 2Comparison of the estimated univariate adjusted statistically significant odds ratios for cases assigned as cattle source (n = 93) compared to all cases. Odds ratios in bold are at least twice those in the all-case dataset.
      QuestionAdjusted odds ratio all cases95% confidence intervalAdjusted odds ratio cattle associated cases95% confidence intervalRatio of ORs cattle:all cases
      Gastric treatment in 4 weeks prior1·84(1·29,2·65)2·12(1·08,4·02)1·2
      Gastro treatment taken?
       Omeprazole4·14(2·29,7·96)5·13(1·99, 12·12)1·2
      Chronic Illness?3·43(1·73,7·44)6·96(2·71, 18·06)2·0
      Type of chronic illness?
       Other
      Excludes Crohn’s diseases, Irritable bowel syndrome, Ulcerative colitis, and Stomach or oesophagus problems.
      3·83(1·38,13·52)4·91(1·04, 23·12)1·3
      Did you eat chicken which was brought frozen then thawed1·76(1·3,2·37)2·11(1·19, 3·68)1·2
      Did you drink raw milk2·11(1·02,4·69)7·01(2·70, 18·44)3·3
      Raw meat or poultry handled in household in the last 7 days….
       I handled poultry meat1·59(1·23,2·06)1·75(1·04,2·94)1·1
      Did you handle kitchen items which had contact with raw meats or poultry?2·19(1·7,2·82)2·93(1·75, 5·00)1·3
      Did you handle the following which had contact with raw meats or poultry?
       Knives2·58(1·99,3·35)3·46(2·03,6·02)1·3
       Chopping boards2·41(1·86,3·14)2·95(1·73,5·10)1·2
       Dish cloths3·22(2·43,4·3)3·39(1·99, 5·77)1·1
      Do you live or work on a farm or lifestyle block?6·16(4·13,9·45)10·8(5·99, 19·73)1·8
      Home/work status?
       Don’t live or work on a farmReference (1·0)
       Live on a farm/lifestyle block7·29(4·49,12·49)10·91(5·34, 22·72)1·5
       Work on a farm/lifestyle block18·34(3·65,333·68)46·58(6·88, 926·91)2·5
       Live and work on a farm/lifestyle block2·52(1·21,5·55)6·93(2·45, 19·02)2·8
      Animals you have contact with at home or work…
       Dairy2·81(1·44,5·93)6·1(2·47, 15·30)2·2
       Meat production2·89(1·7,5·09)3·48(1·52, 7·77)1·2
       Other3·7(2·18,6·6)6·34(2·93, 13·73)1·7
      Are you in contact with dairy cows at home/work?3·02(1·47,6·86)5·42(1·98, 14·93)1·8
      Are you in contact with the following meat production animals at home/work?
       Beef3·09(1·58,6·54)4·58(12·67, 12·33)1·5
       Sheep2·95(1·5,6·26)2·89(1·03, 7·84)1·0
      Did you have contact with…
       Chicken3·73(2·29,6·31)3·59(1·62, 7·73)1·0
       Cattle3·6(2·11,6·43)4·27(1·15, 15·28)1·2
       Sheep3·05(1·79,5·43)9·76(4·65, 21·03)3·2
      Did you have contact with chickens at home3·44(2·01,6·19)3·67(1·53, 8·47)1·1
      Did you have contact with cattle…
       At home4·24(2·17,9·12)8·82(3·57, 22·86)2·1
       Outside the home2·34(1·07,5·67)5·86(2·05, 17·07)2·5
      Did you have occupational contact with cattle?2·91(1·28,7·5)11·59(4·37, 33·42)4·0
      Did you have contact with sheep…
       At home3·32(1·77,6·69)3·83(1·39, 10·12)1·2
       Outside the home2·41(1,6·75)6·02(1·69, 21·49)2·5
      Did you have occupational contact with sheep?3·09(1·21,9·48)9·24(2·91, 32·60)3·0
      Does the water to your home come from…
       Private bore/spring water3·33(1·5,8·46)6·87(2·38, 20·86)2·1
       Roof/rain water4·44(3·11,6·44)4·46(2·46, 8·01)1·0
      Is the private bore water to the home treated or filtered?5·13(1·7,22·22)8·44(1·67, 47·50)1·6
      Is the roof/rain water to the home treated or filtered?5·85(3·71,9·62)6·39(3·08,13·22)1·1
      Did you drink roof/rain water to the home which was filtered6·54(4,11·3)6·93(3·14, 15·29)1·1
      Did you drink water from the following untreated supplies you wouldn't normally?
       Roof/rain water6·62(2·8,19·52)7(2·03, 24·39)1·1
      Did you have contact with compost?2·36(1·77,3·15)3·5(2·11, 5·79)1·5
      a Excludes Crohn’s diseases, Irritable bowel syndrome, Ulcerative colitis, and Stomach or oesophagus problems.
      A multivariable model focusing on factors associated with poultry meat consumption, alongside non-food risk factors, was created using data from all source attributed cases and controls. Table 3shows risk factors with elevated odds ratios and statistically significant population attributable fraction (PAF) values, derived from this model. A number of poultry meat handling factors are significant, which supports the conclusion that poultry meat preparation and consumption is responsible for a high proportion of the cases attributed to poultry as a source. Using separate chopping boards for meat and other foods in this model was a significant risk factor, which seems counterintuitive. It is possible that the use of separate chopping boards may be more common for people who have a higher exposure to contaminated raw poultry. Therefore, this result could be due to residual confounding rather than a causal association.
      Table 3Multivariable model from all cases including factors associated with the preparation, handling and consumption of poultry.
      Risk factorAORAOR ratio 95% CIPAFPAF 95% CI
      Gastro treatment Losec or Omeprazole2·631·27, 5·640·060·02, 0·14
      Chronic illness4·631·35, 17·880·060·01, 0·23
      Contact with a person with gastro symptoms2·11·12, 3·960·080·01, 0·18
      Living or working on a livestock farm4·932·14, 11·840·160·07, 0·39
      Raw milk consumption (urban cases only)4·811·11, 26·290·030·00, 0·1
      Consumed chicken pink on the inside3·081·21, 8·240·050·01, 0·15
      Consumed poultry outside the home1·911·19, 3·060·160·06, 0·33
      Consumed other food such as salads, vegetables or fruit, as part of a meal that did not contain meat1·561·02, 2·390·270, 0·51
      Someone else in the home handled raw poultry meat2·11·36, 3·260·270·13, 0·47
      Handled items in contact with raw poultry or other meat2·61·61, 4·230·360·18, 0·56
      Uses separate chopping boards for foods other than meat1·721·16, 2·580·290·1, 0·52
      Water source consumed not at home (roof/rain water)5·21·21, 28·790·030, 0·13
      AOR: adjusted odds ratio, PAF: population attributable fraction, 95%CI: 95th percentile confidence interval.
      Table 4 shows risk factors with elevated odds ratios and statistically significant population attributable fraction (PAF) values, derived from a multivariable model for factors associated with the cattle associated cases (n = 93). This model supported the finding that animal contact and living or working on a farm/lifestyle block explained a significant proportion (approximately 30%) of cases. While raw milk consumption was a significant risk factor for cattle-attributed urban cases, its consumption was reported by only a small proportion (4·1%) of cases, predominantly in the Manawatū/Whanganui region (data not shown). In contrast, risk factors associated with consumption of beef were not significant after adjusting for confounding.
      Table 4Multivariable model for 93 cattle assigned cases in a source-attributed case control study of campylobacteriosis in New Zealand.
      Risk factorAORAOR 95%CIPAFPAF 95%CI
      Gastro treatment Losec or Omeprazole2·971·2, 6·870·070·01, 0·15
      Chronic illness4·081·31, 12·410·050·00, 0·13
      Raw milk consumption (rural cases)0·650·11, 3·530−0·01, 0·02
      Raw milk consumption (urban cases)6·631·72, 26·170·050·00, 0·12
      Living or working on a farm7·192·95, 18·080·250·09, 0·52
      Contact with compost0·960·43, 1·97−0·01−0·09, 0·14
      Physical contact with animals (chickens)1·050·34, 3·040−0·03, 0·09
      Physical contact with animals (pigs)1·230·2, 6·550−0·01, 0·07
      Physical contact with animals (cattle)2·340·93, 5·030·040·00, 0·14
      Physical contact with animals (sheep)1·170·4, 3·250·01−0·03, 0·07
      AOR: adjusted odds ratio, PAF: population attributable fraction, 95%CI: 95th percentile confidence interval.

      Whole genome sequencing and C. jejuni and C. coli population structure

      The cgMLST population structure of C. jejuni and C. coli isolates from human cases, poultry, cattle and sheep is shown in Figure 1. Many of the globally common lineages were present, including those belonging to seven-gene MLST (ST) types ST45, ST48, ST50, ST21 and ST61. In addition, STs strongly associated with New Zealand, and found rarely elsewhere, were common in this dataset, including ST474 and ST6964 (Supplementary material Table S4) (
      • Mullner O.P.
      • Spencer S.E.
      • Wilson D.J.
      • Jones G.
      • Noble A.D.
      • Midwinter A.C.
      • et al.
      Assigning the source of human campylobacteriosis in New Zealand: a comparative genetic and epidemiological approach.
      ,
      • French N.P.
      • Zhang J.
      • Carter G.P.
      • Midwinter A.C.
      • Biggs P.J.
      • Williamson D.A.
      • et al.
      Genomic analysis of fluoroquinolone- and tetracycline-resistant Campylobacter jejuni sequence type 6964 in humans and poultry, New Zealand, 2014-2016.
      French et al., 2019).
      Figure 1
      Figure 1Minimum spanning tree showing the core genome multilocus sequence type (cgMLST, using 1343 genes) relationship between Campylobacter jejuni and C. coli human clinical isolates (N = 666) and isolates from cattle (N = 172) sheep (194) and poultry (209). The left plot (A) is coloured by sequence type (ST) and the right plot (B) is coloured by host source. The tree is plotted using MSTree V2 in GrapeTree and branch lengths are on a log scale (
      • Zhou Z.
      • Alikhan N.F.
      • Sergeant M.J.
      • Luhmann N.
      • Vaz C.
      • Francisco A.P.
      • et al.
      GrapeTree: visualization of core genomic relationships among 100,000 bacterial pathogens.
      ). Note there are two distinct clades of ST50.

      Source attribution modelling

      STs were able to be assigned for 656 of the 666 cases for which both an interview and an isolate were obtained. The dataset of poultry, sheep and cattle isolates was used to assign a source for the 656 strains identified from the human isolates. Table 5 shows the high-level attribution for the entire dataset of 656 cases, based on the asymmetric island model where cases were attributed to a source when that source was dominant in at least two thirds of all posterior samples.
      Table 5Source assignment for cases in a source-attributed case control study of campylobacteriosis in New Zealand.
      Source assignmentNumber of human case isolates
      Poultry553 (84%)
      Sheep0 (0%)
      Cattle93 (14%)
      Not assigned
      Reason for non assignment is a lack of a clear source attribution result in the model.
      10 (2%)
      Total656
      a Reason for non assignment is a lack of a clear source attribution result in the model.
      The pattern of source assignments for urban and rural cases are different, with approximately 90% of urban campylobacteriosis cases attributed to poultry sources, compared to almost 75% of rural cases (Figure 2). Almost all the remaining cases were attributed to cattle as a source. Source attribution modelling at the population level attributes a small proportion of rural cases to sheep, although no individual cases met the conditions for assignment to sheep as a source. Direct contact with sheep, either through work or domestically, does have an elevated odds ratio in the univariable risk factor analysis, thus remaining a possible transmission pathway.
      Figure 2
      Figure 2Attribution by urban/rural status (2015) in a source-attributed case control study of campylobacteriosis in New Zealand.
      Note: The median is posterior median, quartiles are posterior quartiles, and lower/upper tails are 2·5th and 97·5th percentiles (i.e. 95% credible interval).
      When the cases were stratified by location using modelling with covariates it was evident that there were a relatively higher proportion of cases attributable to cattle in the mixed urban/rural Manawatū/Whanganui region compared to Auckland (Figure 3). There are clear seasonal differences in attribution, with a higher proportion of cases attributable to cattle in autumn and winter compared to spring and summer (Figure 4).
      Figure 3
      Figure 3Source assignment for cases in Auckland and Manawatu in a source-attributed case control study of campylobacteriosis in New Zealand.
      Note: The median is posterior median, quartiles are posterior quartiles, and lower/upper tails are 2·5th and 97·5th percentiles (i.e. 95% credible interval).
      Figure 4
      Figure 4Source assignment by season in a source-attributed case control study of campylobacteriosis in New Zealand.
      Note: The median is posterior median, quartiles are posterior quartiles, and lower/upper tails are 2·5th and 97·5th percentiles (i.e. 95% credible interval).
      The results of using seven-gene MLST, 50-gene rMLST, or 13 genes from the core genome for attribution are shown in Figure 5, Figure 6. Figure 5 shows how attribution of each human case alters using these three different gene sets, and Figure 6 shows how odds ratios from the cattle-attributed case control study would change under each gene set.
      Figure 5
      Figure 5Source assignment of cases based on three different gene sets: MLST, rMLST, or Thépault et al. in a source-attributed case control study of campylobacteriosis in New Zealand (
      • Thépault A.
      • Meric G.
      • Rivoal K.
      • Pascoe B.
      • Mageiros L.
      • Touzain F.
      • et al.
      Genome-wide identification of host-segregating epidemiological markers for source attribution in Campylobacter jejuni.
      ).
      Figure 6
      Figure 6Odds ratios from cattle-assigned case control study where cases are assigned using MLST, rMLST or Thépault et al. in a source-attributed case control study of campylobacteriosis in New Zealand (
      • Thépault A.
      • Meric G.
      • Rivoal K.
      • Pascoe B.
      • Mageiros L.
      • Touzain F.
      • et al.
      Genome-wide identification of host-segregating epidemiological markers for source attribution in Campylobacter jejuni.
      ).

      Discussion

      The results from this study provide comprehensive information on the transmission pathways and source attribution for campylobacteriosis in New Zealand. The majority of cases were infected with strains attributed to a poultry source, and poultry meat consumption was the dominant pathway.
      Although poultry consumption per se was not identified as a significant risk factor, poultry meat has been identified as a significant pathway for campylobacteriosis in New Zealand, and the approximately 50% decline in the notified incidence of campylobacteriosis in 2006-2008 has been linked to interventions in broiler meat processing (
      • Sears A.
      • Baker M.G.
      • Wilson N.
      • Marshall J.
      • Muellner P.
      • Campbell D.M.
      • et al.
      Marked campylobacteriosis decline after interventions aimed at poultry, New Zealand.
      ). Poultry is the most commonly consumed meat type in New Zealand (). Greater than 80% of people involved in the current study as cases or controls had consumed poultry within the previous seven days, and this affects the ability to detect consumption as a risk factor.
      The use of the different gene sets (rMLST or the core genome set from Thépault et al.) for assigning human cases to sources shows that the majority of cases remain attributed to the same source as they were attributed using seven-gene MLST (Figure 5). Compared to seven-gene MLST, rMLST assignments feature additional unknown cases (those with no clear source) that were attributed as cattle or poultry under the seven-gene MLST assignment, but generally similar proportions of poultry and cattle cases were attributed to these sources using rMLST. The Thépault core genome gene set assignment differs more strongly, doubling the number of cattle attributed cases with cases being reattributed from poultry and unknown groups under seven-gene MLST. However, this makes little difference to the overall conclusion of risk factors that contribute to cattle associated cases, with the source-attributed logistic regression results being largely consistent across cases attributed using all three gene sets (Figure 6), where risks include consumption of raw milk in urban cattle attributed cases, chronic illness, treatment of gastroenteritis with proton-pump inhibitors and living or working on a farm.
      All cases were eligible for inclusion in this study by being notified to the human health notifiable disease surveillance system, while controls were eligible by being members of the NZHS cohort. We maximised frequency matching of controls to cases by each of the two study regions and age in order to reduce the impact of confounding by these variables. Residual confounding resulting from imperfect matching was considered by adjusting for these variables in multivariable models.
      The decision by each case whether to present to a general practitioner may have resulted in selection bias of total cases in the community. It is generally assumed that cases presenting to general practice will be those that are more severely affected by the infection and therefore these cases may not be representative of the wider set of campylobacteriosis cases. Selection bias may also have been introduced as a result of non-participation in the recruitment of both cases and controls, however the proportion of cases refusing to participate was relatively low (8% in Auckland and 12% in Manawatū/Whanganui), while 12% of controls refused to participate. The impact of bias was minimised by using a standardised approach to interviewing both cases and controls.
      Some degree of recall bias in relation to food consumption is inevitable in such a study. This was minimised by using a study design that limited the time from illness notification to interview to 48 h or less, and asking questions relating to the recent past (seven days for food consumption); therefore, any recall bias is likely to be similar in cases and controls.
      Source isolates were obtained from samples provided by meat and poultry slaughterhouses that were likely to constitute the main supply to the study regions. However, meat and poultry distribution pathways will always be complex within a country and therefore the source samples will not have been fully representative of the food supply in the Auckland and Manawatū/Whanganui regions. Given likely sources described in the scientific literature, and the practicalities and costs of sampling, the three most likely animal sources were prioritised to obtain isolates for model attribution. Other potential sources, such as pork or pets, were considered unlikely to be major contributors for attribution, or presented sampling difficulties.
      A systematic review of source attribution of human campylobacteriosis using MLST has shown that the proportion attributed to poultry is consistently high across developed countries (
      • Cody A.J.
      • Maiden M.C.
      • Strachan N.J.
      • McCarthy N.D.
      A systematic review of source attribution of human campylobacteriosis using multilocus sequence typing.
      ). This study is consistent with that review, and attributes an even greater proportion of 84% of cases to a poultry source using the same methodology. The ability of this study to separate urban and rural cases provides more detail on the importance of poultry as a source; approximately 90% of urban campylobacteriosis cases were attributed to poultry sources compared with fewer than 75% of rural cases (in 2006 14% of the New Zealand population lived in areas classified as rural) (
      • Massey University
      Environmental health indicators New Zealand: urban rural profile.
      ).
      This study has shown that the principal pathway for exposure leading to infection with Campylobacter in New Zealand remains poultry meat. This source is somewhat diminished for the rural population (14% of New Zealanders) where infection from direct contact with farm animals, consumption of raw milk and using private bore/spring water or rainwater as a drinking water supply all make a modest contribution to infection risk. Having a chronic illness and taking certain treatments for gastrointestinal problems are distinct risk factors, presumably through increasing susceptibility to infection and clinical illness for any given level of exposure to this bacterium (
      • Esan O.B.
      • Pearce M.
      • van Hecke O.
      • Roberts N.
      • Collins D.R.J.
      • Violato M.
      • et al.
      Factors associated with sequelae of Campylobacter and non-typhoidal salmonella infections: a systematic review.
      ). Although there may be differences in results for rural populations in different areas of the country, due to varying farming types, the results for the urban population studied should be applicable to urban dwellers across the whole country.
      Generation of strong risk-based evidence on the dominant transmission pathway for campylobacteriosis in New Zealand provides a solid platform for continuing efforts by government and industry to mitigate this important public health problem. The poultry meat food chain offers several links where intensified or new control measures can be reasonably implemented, and this study provided the impetus for New Zealand Food Safety to set a new public health improvement goal for a 20% reduction in foodborne campylobacteriosis from 2020 to 2025 ().

      Contributors

      DC NF RL Principal Investigators, AM DC JS MB NF PC PS RL SH Designed the study, JB DC EA NF PC RL SH Managed the project, AM JB DW EA JS JM PC RL SP Collected the data, AM JB BH DW EA JM NF PC RL SP Analysed the data, RL Prepared the first draft, All authors reviewed the manuscript, RL Finalised the manuscript of the basis of comments from other authors and reviewer feedback.

      Declaration of interests

      DC reports personal fees from Ministry for Primary Industries, outside the submitted work; NF reports grants from Ministry for Primary Industries, during the conduct of the study; grants from New Zealand Food Safety Science and Research Centre, outside the submitted work; RL reports grants from New Zealand Ministry for Primary Industries during the conduct of the study, grants from New Zealand Ministry of Health, grants from New Zealand Food Safety Science and Research Centre, outside the submitted work; PC reports grants from New Zealand Ministry for Primary Industries during the conduct of the study; grants from New Zealand Ministry for Primary Industries, grants from New Zealand Ministry of Health, grants from New Zealand Food Safety Science Research Centre, outside the submitted work; EA reports grants from New Zealand Ministry for Primary Industries, during the conduct of the study.

      Funding

      This study was funded by the N ew Zealand Ministry for Primary Industries .
      Ethics approval for the study was obtained from the Northern B Health and Disabilities Ethics Committee (HDEC).

      Acknowledgments

      We thank Elaine Taylor (Ministry for Primary Industries) for her contribution to completion of the study: Dr. Ahmed Fayaz, Massey University, for assistance with whole genome sequencing and data management; and Lynn Rogers and Rukhshana Akhter, mEpiLab, Massey University, for library preparations.

      Appendix A. Supplementary data

      The following is Supplementary data to this article:

      References

        • Bruzzi P.
        • Green S.B.
        • Byar D.P.
        • Brinton L.A.
        • Schairer C.
        Estimating the population attributable risk for multiple risk factors using case-control data.
        Am J Epidemiol. 1985; 122: 904-914
        • Cody A.J.
        • Maiden M.C.
        • Strachan N.J.
        • McCarthy N.D.
        A systematic review of source attribution of human campylobacteriosis using multilocus sequence typing.
        Euro Surveill. 2019; 24
        • Eberhart-Phillips J.
        • Walker N.
        • Garrett N.
        • Bell D.
        • Sinclair D.
        • Rainger W.
        • et al.
        Campylobacteriosis in New Zealand: results of a case control study.
        J Epidemiol Community Health. 1997; 51: 686-691
        • Esan O.B.
        • Pearce M.
        • van Hecke O.
        • Roberts N.
        • Collins D.R.J.
        • Violato M.
        • et al.
        Factors associated with sequelae of Campylobacter and non-typhoidal salmonella infections: a systematic review.
        EBioMedicine. 2017; 15: 100-111
        • French N.P.
        • Zhang J.
        • Carter G.P.
        • Midwinter A.C.
        • Biggs P.J.
        • Williamson D.A.
        • et al.
        Genomic analysis of fluoroquinolone- and tetracycline-resistant Campylobacter jejuni sequence type 6964 in humans and poultry, New Zealand, 2014-2016.
        Emerg Infect Dis. 2019; 25: 2226-2234
        • Friedman J.
        • Hastie T.
        • Tibshirani R.
        Regularization paths for generalized linear models via coordinate descent.
        J Stat Softw. 2010; 33: 1-22
        • Hothorn T.
        • Buhlmann P.
        • Dudoit S.
        • Molinaro A.
        • van der Laan M.J.
        Survival ensembles.
        Biostatistics. 2006; 7: 355-373
        • Liao S.J.
        • Marshall J.
        • Hazelton M.L.
        • French N.P.
        Extending statistical models for source attribution of zoonotic diseases: a study of campylobacteriosis.
        J R Soc Interface. 2019; 1620180534https://doi.org/10.1098/rsif.2018.0534
        • Liaw A.
        • Wiener M.
        Classification and Regression by randomForest.
        R News. 2002; 2: 18-22
      1. Marshall J. 2020. https://github.com/jmarshallnz/islandR. [Accessed 20 August 2019].

        • Massey University
        Environmental health indicators New Zealand: urban rural profile.
        2020
        • Ministry of Health
        New Zealand health survey.
        2019
        • Mughini Gras L.
        • Smid J.H.
        • Wagenaar J.A.
        • de Boer A.G.
        • Havelaar A.H.
        • Friesema L.H.
        • et al.
        Risk factors for campylobacteriosis of chicken, ruminant, and environmental origin: a combined case-control and source attribution analysis.
        PLoS One. 2012; 7e42599
        • Mughini Gras L.
        • Smid J.H.
        • Wagenaar J.A.
        • Koene M.J.
        • Havelaar A.H.
        • Friesema I.H.
        • et al.
        Increased risk for Campylobacter jejuni and C. coli infection of pet origin in dog owners and evidence for genetic association between strains causing infection in humans and their pets.
        Epidemiol Infect. 2013; 141: 2526-2535
        • Mullner P.
        • Jones G.
        • Noble A.
        • Spencer S.E.
        • Hathaway S.
        • French N.P.
        Source attribution of food-borne zoonoses in New Zealand: a modified Hald model.
        Risk Anal. 2009; 29: 970-984
        • Mullner O.P.
        • Spencer S.E.
        • Wilson D.J.
        • Jones G.
        • Noble A.D.
        • Midwinter A.C.
        • et al.
        Assigning the source of human campylobacteriosis in New Zealand: a comparative genetic and epidemiological approach.
        Infect Genet Evol. 2009; 9: 1311-1319
        • New Zealand Food Safety
        Managing the foodborne risk of Campylobacter.
        2020 ([Accessed 16 July 2020])
        • Nohra A.
        • Grinberg A.
        • Midwinter A.C.
        • Marshall J.C.
        • Collins-Emerson J.M.
        • French N.P.
        Molecular epidemiology of Campylobacter coli strains isolated from different sources in New Zealand between 2005 and 2014.
        Appl Environ Microbiol. 2016; 82: 4363-4370
        • Pattis I.
        • Cressey P.
        • Lopez L.
        • Horn B.
        • Soboleva T.
        Annual report concerning foodborne disease in New Zealand 2018: Ministry for Primary Industries.
        2019
        • Poultry Industry Association New Zealand (PIANZ)
        Industry facts.
        2019
        • Sears A.
        • Baker M.G.
        • Wilson N.
        • Marshall J.
        • Muellner P.
        • Campbell D.M.
        • et al.
        Marked campylobacteriosis decline after interventions aimed at poultry, New Zealand.
        Emerg Infect Dis. 2011; 17: 1007-1015
        • Statistics New Zealand
        Defining urban rural New Zealand.
        2020 ([Accessed 16 July 2020])
        • Strobl C.
        • Boulesteix A.L.
        • Zeileis A.
        • Hothorn T.
        Bias in random forest variable importance measures: illustrations, sources and a solution.
        BMC Bioinform. 2007; 8: 25
        • Strobl C.
        • Boulesteix A.L.
        • Kneib T.
        • Augustin T.
        • Zeileis A.
        Conditional variable importance for random forests.
        BMC Bioinform. 2008; 9: 307
        • Thépault A.
        • Meric G.
        • Rivoal K.
        • Pascoe B.
        • Mageiros L.
        • Touzain F.
        • et al.
        Genome-wide identification of host-segregating epidemiological markers for source attribution in Campylobacter jejuni.
        Appl Environ Microbiol. 2017; 83e03085-16https://doi.org/10.1128/AEM.03085-16
        • Wagenaar J.A.
        • French N.P.
        • Havelaar A.H.
        Preventing Campylobacter at the source: why is it so difficult?.
        Clin Infect Dis. 2013; 57: 1600-1606
        • Zhou Z.
        • Alikhan N.F.
        • Sergeant M.J.
        • Luhmann N.
        • Vaz C.
        • Francisco A.P.
        • et al.
        GrapeTree: visualization of core genomic relationships among 100,000 bacterial pathogens.
        Genome Res. 2018; 28: 1395-1404