Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Preventive Veterinary Medicine 98 (2011) 1–9                                                         Contents lists availa...
2                                       T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9of a live virus and...
T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9                                 3the studies conducted on ...
4                                      T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9    The virological ...
T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9                                                  5examined...
6                                              T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9            ...
T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9                                  7making early disease det...
8                                      T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9lower compared to no...
T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9                                                    9    fo...
Upcoming SlideShare
Loading in …5

Meta analysis on the efficacy of foot-and-mouth disease


Published on

Published in: Education
  • Be the first to comment

  • Be the first to like this

Meta analysis on the efficacy of foot-and-mouth disease

  1. 1. Preventive Veterinary Medicine 98 (2011) 1–9 Contents lists available at ScienceDirect Preventive Veterinary Medicine journal homepage: on the efficacy of foot-and-mouth diseaseemergency vaccinationT. Halasa a,∗ , A. Boklund a , S. Cox b , C. Enøe aa Technical University of Denmark, the Veterinary Institute, Bülowsvej 27, 1790 Copenhagen V, Denmarkb Institute for Animal Health, Pirbright Laboratory, Ash Road, Pirbright, Working, Surry, GU 24, 0NF, United Kingdoma r t i c l e i n f o a b s t r a c tArticle history: The objectives of this study were to provide a summary quantification of the efficacy of FMDReceived 20 November 2009 emergency vaccination based on a systematic review and a meta-analysis of available liter-Received in revised form 23 August 2010 ature, and to further discuss the suitability of this review and meta-analysis to summarizeAccepted 26 August 2010 and further interpret the results. Peer-reviewed, symposium, and unpublished studies were considered in the analysis. Clinical protection and virological protection against FMD wereKey words: used as parameters to assess the efficacy of emergency vaccination. The clinical protectionEmergency vaccine, Foot-and-mouth was estimated based on the appearance of clinical signs including FMD lesions and fever,disease, Meta-analysis while the virological protection parameter was estimated based on the outcome of labo- ratory tests that were used to diagnose FMD infection. A meta-analysis relative risk was calculated per protection parameter. Results of the meta-analyses were examined using publication bias tests. In total, 31 studies were included in the analyses, of which 26 were peer-reviewed studies, 1 was a symposium study and 4 were unpublished studies. Cattle, swine and sheep were well protected against clinical disease and FMD infection following the use of emergency vaccine. Fortunately, no significant bias that would alter the conclu- sions was encountered in the analysis. Meta-analysis can be a useful tool to summarize literature results from a systematic review of the efficacy of FMD emergency vaccination. © 2010 Elsevier B.V. All rights reserved.1. Introduction countries. Following the confirmation of FMD in any EU member state, slaughter and disinfection of infected herds, Foot-and-mouth disease (FMD) is a highly contagious as well as movement restriction and surveillance of herdsviral disease affecting ruminants and pigs, which can cause within zones around the infected herds must be applied tohuge economic damage (Cox and Barnett, 2009; Pendell et contain the outbreak as fast as possible (Council Directiveal., 2007). Although FMD remains as an endemic disease 2003/85/EC). From January 1st 1992, prophylactic vacci-in several areas of the world, elsewhere FMD free status nation against FMD was prohibited in the EU countries.has been maintained or FMD was eradicated by imposing However, after public concerns regarding the mass killingstrict legislation and/or implementing a variety of control of animals in the UK and the Netherlands during the FMDstrategies including use of vaccination (Cox and Barnett, outbreak in 2001, the OIE and the EU have revised the pol-2009). The eradication of the disease and the prohibition icy to make use of emergency vaccination in case the riskof vaccination in these countries have created a highly of an extensive outbreak is suspected (Cox et al., 2005;susceptible population of animals. This means that a poten- Parida, 2009). Emergency vaccination may be defined astial outbreak could rapidly spread within and between the the use of vaccines to control an outbreak of FMD in a country free of the disease, in which routine prophylactic vaccination is not applied (Council Directive 2003/85/EC). ∗ Corresponding author. Tel.: +45 35886286; fax: +45 35886001. Based on the literature, an ideal emergency vaccine should E-mail address: (T. Halasa). have the following characteristics: (1) contain no residues0167-5877/$ – see front matter © 2010 Elsevier B.V. All rights reserved.doi:10.1016/j.prevetmed.2010.08.005
  2. 2. 2 T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9of a live virus and have minimal side effects to newborns ously reported in a symposium paper, to prevent data setsand adults; (2) after a single application with the recom- from being included twice in the analyses.mended dose, achieve a potency dose 50 (PD50 ) of ≥6; (3) The quality of the studies was further assessed by insur-be compatible with serological tests that identify infection ing that a study: (1) met the above mentioned inclusionin vaccinated animals; (4) induce reasonably long lasting criteria; (2) provided sufficient information on animal han-immunity and provide a broad spectrum of antigenic pro- dling, vaccine application and virus challenge procedure;tection; and (5) be stable under storage once formulated; (3) provided sufficient information regarding the measure(6) provide a rapid protection after vaccination; and (7) of clinical disease and laboratory procedures to diagnosereduce the reproduction ratio (R0 ) to below 1 (Goris et al., infection; (4) provided sufficient information about the2007; Cox and Barnett, 2009). duration of the experiment, and sampling frequency; (5) The economic consequence of using emergency vac- had an epidemiologic design that would meet the objec-cination following an FMD outbreak is an essential tives of that study.consideration for livestock and livestock products export- The primary check of the studies was carried out bying countries. Their export status is highly dependent on reading the abstract of the studies. When a study was foundthe freedom of FMD, which could be delayed because of to fit our inclusion criteria, the study was read to insurevaccination (Mahul and Durand, 2000). Several experimen- a good quality and thereafter a careful reading accompa-tal studies have quantified the efficacy of FMD emergency nied with data extraction was carried out. When a studyvaccines and these were recently reviewed by Cox and lacked essential information, the authors were contactedBarnett (2009). The authors observed a considerable vari- and asked to provide the necessary information.ability between studies. This would make it difficult Unpublished data might be an important source of infor-to conduct comprehensive economic assessment of the mation and should be considered in meta-analysis studiesefficacy of emergency vaccination. It was, therefore, rec- (Dohoo et al., 2003). Therefore, several researchers andommended to conduct a meta-analysis study to quantify research groups that are involved in FMD vaccine studiesthe efficacy of FMD emergency vaccination. The outcome of were approached to provide unpublished data that fittedthis meta-analysis can be used for FMD modeling exercises the above mentioned inclusion criteria from point 2 to 6.and economic analysis to assist policy makers in national The researchers were asked to provide further informationgovernments, the EU and worldwide on efficient FMD con- as needed to assess the quality of the data. This is importanttrol strategies. to correctly pool unpublished data with published studies The objectives of this study were to provide a summary in a meta-analysis (Dohoo et al., 2003).quantification of the efficacy of FMD emergency vaccina- Because different studies differ in the design, discrep-tion based on a systematic review and a meta-analysis of ancies were expected that could affect the validity of theavailable literature, and to further discuss the suitability of meta-analysis. Data related to the study design, emergencythis review and meta-analysis to summarize and further vaccination and challenge are presented in the next sec-interpret the results. tions and in supplementary file.2. Materials and methods 2.2. Emergency vaccination and challenge2.1. Selection of published and unpublished studies The majority of the studies were performed in disease secure facilities. At the start of each study, the animals were A search was conducted on the literature related to FMD free. Frequently, animals were given a few days toFMD vaccine efficacy studies published between 1960 and adapt to the new facility before assignment to either ofthe beginning of 2009. The search was carried out using the vaccinated or control groups. In the majority of thethe key words: foot-and-mouth disease; emergency; vac- studies, the emergency vaccine application was carried outcine; efficacy; and protection in different combinations. using the intramuscular route of injection using a full doseSearches were conducted in PubMed (the National Library (Supplementary file, column VA). Thereafter, a challengeof Medicine, Bethesda, USA) and Google Scholar (Google with a live homologous FMD virus was carried out. SomeInc., CA, USA). studies examined the efficacy of emergency vaccination Studies included in the analysis had to: (1) be a research using short and long periods between vaccination and chal-or symposium paper published in English; (2) be an exper- lenge, while other used only long intervals (Supplementaryimental challenge study, which showed the efficacy of an file, footnote j). In the majority of the studies conductedFMD emergency vaccine in pigs, cattle, and/or sheep; (3) on cattle, a period of 21 days between emergency vacci-report the rate or the number of FMD infected or protected nation and challenge was used. In studies conducted onanimals and the total number of animals in at least 2 groups swine and sheep, a period of 14 days between emergency(challenged vaccinated and non-vaccinated control group); vaccination and challenge was often used. However, sev-(4) include vaccination with the full vaccine dose. Proto- eral protocols were frequently tested within these studies.cols with sub-preventive doses were not included so as Control animals were separated from the vaccinated ani-not to under-estimate the efficacy of the tested vaccine; mals. In the majority of the studies conducted on cattle,(5) include a challenge with a homologous virus to that contact with infected cattle for several days or a pig forused in the vaccine, similarly in both groups (vaccinated 1 h was used as the challenge method, while intra-dermaland non-vaccinated); (6) in case of research papers, report injection of the virus in the tongue was carried out in otherthe outcome of a new experiment or protocol not previ- studies (Supplementary file, column CR). In the majority of
  3. 3. T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9 3the studies conducted on pigs, challenge was carried out et al., 1984; Salt et al., 1995, 1998) that used only VI toby direct or less frequently indirect contact with infected confirm infection from the samples. Orsel et al. (2005)pigs for a period from 1 h to several days. In few studies, used VI and PCR to confirm infection from the sam-intra-dermal injection of the virus in the heel bulb was ples.used (Supplementary file, column CR). In the majority ofthe studies conducted on sheep, challenge with infected 2.5. Evidence of FMDpigs for 2–9 h was used (Supplementary file, column CR). 2.5.1. Clinical disease2.3. Follow-up days post-challenge Clinical signs of FMD were diagnosed in all studies. However, there was variability between studies in the con- After challenge animals were monitored for clinical sideration of which clinical signs represent clinical disease.infection over variable time periods (Supplementary file, The majority of studies conducted on cattle and pigs con-column FC) and samples for laboratory analyses and diag- sidered the presence of FMD lesions and/or fever as a signnosis were obtained (Supplementary file, column FS). In of disease (Supplementary file, column Fe). In the othermost studies conducted on cattle, 7–14 days post-challenge studies only lesions on the feet were considered to diag-was used as the follow-up period for clinical diagnosis. nose clinical infection. In studies conducted on sheep, theHowever, some studies recorded clinical disease up to the presence of FMD lesions and/or fever was considered toend of the trial. Follow up for laboratory diagnosis varied diagnose clinical infection.from 14 to 35 days post-challenge, but was mainly up to 28 In the studies that used fever as a clinical sign, discrep-days post-challenge. ancies between the studies in the consideration of fever In most studies conducted on pigs, the follow-up period were also observed. In only one study conducted usingfor clinical diagnosis was also 7–14 days post-challenge. cattle, where fever was considered a clinical sign of theHowever, in one study (Eblé et al., 2009), clinical disease disease, fever was specified to be a rectal temperature ofwas recorded up to 42 days post-challenge. Follow up for ≥40 ◦ C (Aggarwal et al., 2002). In studies conducted usinglaboratory diagnosis varied from 8 up to 42 days post- swine, 40 ◦ C was used as a cut-off value, except in one studychallenge, but was frequently up to 14 days post-challenge. (Parida et al., 2007), which used 39.5 ◦ C as a cut-off value. In studies conducted on sheep, follow up for clinical Two studies conducted with swine included fever to diag-diagnosis varied from 10 up to 14 days post-challenge, nose FMD, but did not specify a cut-off value (Doel et al.,while in one study (Barnett et al., 2004), FMD lesion record- 1994; Orsel et al., 2007a). In studies conducted using sheep,ing was carried out up to 42 days post-challenge. Follow up fever was defined as ≥40 ◦ C, except one study (Parida et al.,for laboratory diagnosis varied from 14 up to 42 days post- 2008), which used ≥39.5 ◦ C.challenge and was frequently up to 14 days post-challenge.2.4. Samples collection, storage and laboratory methods 2.5.2. Virological infection In the majority of studies, FMD infection was confirmed Sample collection, storage and laboratory methods were by laboratory diagnosis (Supplementary file, column FS).generally similar among the different studies. We will only In the current meta-analysis we defined infection baseddescribe these procedures briefly, as detailed information on the results of: (1) VI from the blood, oral, nasal and/orcan be found in the original publications. esophageal–pharyngeal fluids samples; (2) presence of Heparinised and unheparinised blood samples were antibodies against NSP; or (3) presence of viral RNA in oral,stored at −70 ◦ C, while serum samples were stored at nasal and/or esophageal–pharyngeal fluids samples diag-−20 ◦ C until processing. Nasal fluids were collected on cot- nosed using RT-PCR.ton buds and stored at −70 ◦ C. Oropharyngeal fluid wascollected using cotton swabs. For ELISA, oropharyngeal 2.6. Meta-analysis procedurefluid samples were stored at −20 ◦ C, but they were storedat −70 ◦ C for virus isolation (VI). Probang samples were 2.6.1. Outcome parameterscollected using a probang sampler on sedated animals and In the current meta-analysis we defined two mainthen the samples were stored at −70 ◦ C. parameters to determine the efficacy of FMD emergency VI or titration was carried out by inoculation of the vaccination: clinical protection and virological protection.sample onto monolayers of mammalian cells in tubes or The relative risk (RR) of clinical disease was consideredplates. Infection was determined either by staining and to represent clinical protection. For each study, the RR wascounting of plaques or by microscopic examination for calculated as the incidence risk of clinically FMD diseasedthe cytopathic effect with subsequent confirmation of animals in the vaccinated group divided by the incidenceserotype by ELISA. The presence of antibodies against the risk in the unvaccinated control group. The incidence risk ofnon-structural proteins (NSP) 3ABC was tested using com- clinical FMD was calculated for each group as the number ofmercial ELISA. The RT-PCR was used to confirm and identify clinically diseased animals divided by the total number ofinfection from the samples, mostly using the MagNA Pure® animals. The length of the follow-up period post-challengeLC kit (Roche) and a light cycler® RT-PCR system. was assumed not to affect the outcome, given that animals All studies that confirmed infection using laboratory were followed up at least 7 days. Seven days after challengetests used VI, PCR and NSP, except (Barnett et al., 2002; was assumed to be a sufficient period for clinical signs toCox et al., 1999; Donaldson and Kitching, 1989; Gibson appear.
  4. 4. 4 T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9 The virological protection was estimated based on the combined in a multivariable model using a backward step-definition of infection, which considered an animal to have wise regression method. A liberal P-value <0.3 was chosenFMD infection when it was found positive to VI, NSP or RT- for the variables to be included in a combined multivari-PCR tests from the original studies. The RR of infection was able model. In case of a significant association between theused to represent virological protection. For each study, explanatory variable and the dependent variable (RR pera RR of infection was calculated as the incidence risk of study) with a P-value <0.05 in the multivariable model, theinfection in the vaccinated group divided by the incidence variable was believed to explain the heterogeneity signifi-risk in the unvaccinated control group. The incidence risk cantly.of infection was calculated for each group (vaccinated andunvaccinated) as the number of animals that were positive 2.6.4. Publication biasto any of the 3 tests divided by the total number of animals. Because studies that result in large and interestingAs for clinical disease, we assumed that the length of the effects are more likely to be published than studies thatfollow-up period post-challenge to diagnose infection did show relatively small or no effects, the outcome could benot affect the outcome, given that animals were followed a biased body of research (Halasa et al., 2009). The publi-for at least 7 days. Seven days after challenge was assumed cation bias was assessed using funnel plots. A funnel plotto be a sufficient period for virus shedding to occur and 14 is a plot of a measure of study size (standard error) ondays for NSP antibodies to develop. the vertical axis as a function of the effect size on the From 4 studies (Eblé et al., 2004, 2007; Orsel et al., horizontal axis. Large studies appear toward the top of2005, 2007b) only results of experiments using vaccinated the graph, and tend to cluster near the mean effect value.and non-vaccinated animals challenged by direct injection Smaller studies appear toward the bottom of the graphof the virus were included in the current meta-analysis. and tend to disperse across a range of values. PublicationOnly results of experiments using vaccinated and non- bias methods included Duval and Tweedie’s fill and trimvaccinated animals that had been challenged using the method (Duval and Tweedie, 2000), Begg and Mazumdarairborne route with unvaccinated infected seeders were correlation test (Begg and Mazumdar, 1994), and Egger’sincluded in the analysis from Orsel et al. (2007a). Results of regression test (Egger et al., 1997). A significant publica-other experiments from these studies were not included. tion bias was deemed to exist when adjustment for the bias altered our conclusion or when the confidence limits of2.6.2. Statistical procedure the unadjusted and the adjusted RR did not overlap. When The RRs were pooled over the studies; separately per significant publication bias and change on the estimatedprotection parameter, and animal species, using a commer- pooled RR were detected, the number of studies necessarycial analytical package (Comprehensive meta-analysis 2.0, to reverse the pooled effect was calculated using Orwin’s2009). When the pooled RRs were calculated, classification fail-safe N method (Orwin, 1983). The study influence wason virus serotype was carried out to correct for potential examined using the one study removed method (Dohoo etdifferences between the different virus serotypes in reac- al., 2003). When significant publication bias was deemedtion to vaccination. In studies that included more than one to exist, the pooled RR was presented based on the Duvalprotocol, a combined effect was calculated per study. When and Tweedie’s fill and trim estimation after correcting forone study reported the outcome of separate challenge trials the publication bias. The interpretation of each of the abovewith different virus serotypes, we treated each challenge mentioned methods is presented in Sections 3 and 4.trail as a separate study in the meta-analysis. Because stud- It is important to mention that these methods are basedies are conducted by different people, in different areas and on statistical theory. They do not necessarily prove exis-times, which could create a heterogeneous population of tence of bias, but they do suggest it, and the confirmationstudies, a random effect model was used to estimate the of the bias should be based on biological and rational rea-pooled RR (Halasa et al., 2009). A meta-analysis was con- soning (Halasa et al., 2009).ducted when at least 4 studies were available (Halasa et al.,2009). Forest plots were used to illustrate the calculated RR 3. Resultsper study and virus serotype and the overall pooled effectof all studies in the last line of the plot. 3.1. Descriptive results2.6.3. Meta-regression The primary search identified 862 published studies of As recommended by Lean et al. (2009), a weighted relevance. Of these, 69 studies examined FMD vaccines inmeta-regression was conducted in an attempt to explain cases and control groups in cattle, swine and/or sheep.the heterogeneity between studies. Explanatory variables However, only 29 published studies described a vaccineof the heterogeneity were selected based on the litera- that would meet our criteria for an emergency vaccine. Theture and expert assessment of likely factors. The variables authors of 4 studies were contacted to identify the PD50 ofused are presented in the supplementary file. The coun- the vaccine used in their studies. Finally, 28 studies weretry of origin and animal species were not included in the identified to fit our criteria and 27 of them were included inmeta-regression, while the number of tests conducted to the analysis. One study (Orsel et al., 2007c) was excluded toconfirm infection per study was included. These variables correct for age as a natural confounder, because the studywere regressed against the RR results of each study and only included older cattle. Of these 27 studies, 10 exam-weighted by the inverse variance (Dohoo et al., 2003). The ined the efficacy of emergency vaccination using cattle; 9variables were first tested in univariable models and then studies using swine; 5 studies using sheep, and 3 studies
  5. 5. T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9 5examined emergency vaccination in more than 1 species Table 1 Pooled relative risk (RR) together with the 95% confidence interval as(Supplementary file). One of the 5 studies (Cox et al., 1999) estimated based on the corresponding studies in the meta-analysis toconducted in sheep included 3 different virus strains, and quantify the clinical and virological protection against foot-and-mouthhence each experiment was treated in the analysis and disease using emergency vaccination.referred to as a separate study (Supplementary file). All Animal species Studies included in the Pooled RR and 95%studies were peer-reviewed research papers except one and measure of meta-analysis confidence limits(Salt et al., 1995), which was a symposium paper. protection Four unpublished experiments conducted on swine Cattlewere included in the analysis; all of good quality and Clinical (Aggarwal et al., 2002; 0.13 (0.09–0.18)with experimental design similar to other included studies Brehm et al., 2008; Cox et(Supplementary file). al., 2005; Cox et al., 2006; Cox et al., 2007; Doel et al., 1994; Donaldson and3.2. Protection in cattle Kitching 1989; Goris et al., 2007, 2008; Graves et al., 1968; Mattion et al., 2004; Studies included in investigating the clinical protec- Orsel et al., 2005; Salt et al.,tion provided by emergency vaccination against FMD in 1995)cattle were homogeneous (Q-value = 16.3 with 12 degrees Virological (Cox et al., 2005, 2006, 0.71 (0.59–0.85)of freedom and a P-value = 0.178). The studies included 2007; Donaldson and Kitching 1989; Orsel et al.,virus serotypes O, A and Asia 1. In general, emergency 2005; Salt et al., 1995)vaccination protected cattle well against clinical disease Swine(Table 1). There was no significant difference in clinical Clinical (Aggarwal et al., 2002; 0.48 (0.36–0.65)protection between the different vaccine serotypes (Fig. 1). Barnett et al., 2002; Buonavoglia et al., 1998;Removing any of the studies did not alter the pooled effect Doel et al., 1994; Eblé et al.,significantly. Further investigation of the publication bias 2004, 2007, 2009; Gravesshowed signs of bias. The Begg and Mazumdar test sug- et al., 1968; Orsel et al.,gested a significant correlation between study size and 2007a,b; Parida et al.,study effect ( = −0.45 and a P-value = 0.03). Egger’s regres- 2007; Sellers and Herniman, 1974; Salt et al.,sion test suggested a significant association between study 1998; Unpublished data).size and study effect with an intercept = −1.46, a stan- Virological (Barnett et al., 2002; Eblé et 0.67 (0.51–0.87)dard error = 0.45 and a P-value = 0.01. Correcting for the al., 2004, 2007, 2009; Orselbias using the Duval and Tweedie’s trim and fill method, et al., 2007a,b; Parida et al., 2007; Salt et al., 1998)resulted in an insignificant change to the pooled RR. Sheep In the meta-analysis to examine virological protec- Clinical (Aggarwal et al., 2002; 0.31 (0.18–0.53)tion against FMD, the studies were heterogeneous, and Barnett et al., 2004; Cox etnone of the variables in the meta-regression explained al., 1999; Gibson et al.,the heterogeneity significantly. Vaccinated cattle had 0.71 1984; Orsel et al., 2007a,b; Parida et al., 2008)(0.59–0.85) times lower risk of FMD infection compared Virological (Barnett et al., 2004; Cox et 0.59 (0.44–0.80)to non-vaccinated cattle (Table 1). Publication bias tests al., 1999; Gibson et al.,indicated no significant publication bias. 1984; Orsel et al., 2007a,b; Parida et al., 2008)3.3. Protection in swine publication bias. Studies investigating the clinical protection providedby emergency vaccination against FMD in swine were 3.4. Protection in sheepheterogeneous. None of the tested variables in the meta-regression explained the heterogeneity significantly. The Studies investigating the clinical protection provided bystudies included virus serotypes O, A, C and Asia. In general, emergency vaccination against FMD in sheep were het-emergency vaccination provided good protection against erogeneous. In the meta-regression, none of the testedclinical disease in swine (Table 1). There was no signifi- variables explained the heterogeneity significantly. Thecant difference in clinical protection between the different included studies examined virus serotypes O, C and Asiavaccine serotypes. Publication bias tests did not indicate a 1. In general, emergency vaccination protected sheep wellsignificant bias that would alter the results. against clinical disease (Table 1) and no significant publi- Emergency vaccination protected swine against FMD cation bias was found.infection (Fig. 2). Vaccinated swine had 0.67 (0.51–0.87) Emergency vaccination provided virological protectiontimes lower risk of FMD infection compared to non- against FMD infection in sheep (Table 1). The studies werevaccinated swine (Table 1). The studies were heteroge- heterogeneous and none of the variables in the meta-neous, and none of the variables in the meta-regression regression explained the heterogeneity significantly. Thereexplained the heterogeneity significantly. Only Egger’s was no significant difference in protection between the dif-regression test suggested an association between study size ferent vaccine serotypes. Removing any of the studies didand study effect. All other publication bias test indicated no not alter the effect. Begg and Mazumdar correlation test
  6. 6. 6 T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9 Virus serotype Study RR Lower Upper RR and 95% CI limit limit A Brehm, 2008 0.051 0.003 0.770 A Goris, 2008 0.123 0.071 0.214 A Graves, 1968 0.075 0.016 0.352 A Mattion, 2004 0.225 0.069 0.737 PRR A 0.126 0.079 0.202 Asia 1 Salt, 1995 0.124 0.052 0.295 PRR Asia 1 0.124 0.052 0.295 O Aggarwal, 2002 0.100 0.007 1.490 O Cox, 2005 0.026 0.002 0.406 O Cox, 2006 0.026 0.002 0.406 O Cox, 2007 0.294 0.177 0.488 O Doel, 1994 0.019 0.001 0.297 O Donaldson, 1989 0.179 0.089 0.361 O Goris, 2007 0.090 0.037 0.219 O Orsel, 2005 0.100 0.015 0.664 PRR O 0.127 0.068 0.236 Overall PRR 0.126 0.089 0.178 0.01 0.1 1 10 100 Favors vaccine Favors no vaccineFig. 1. Forest plot of the relative risk (RR) of the clinical disease in vaccinated cattle for each of the 13 included studies, the pooled RR (PRR) per virusserotype and the overall PRR together with the 95% confidence interval (CI). Virus serotype Study RR Lower Upper RR and 95% CI limit limit C Salt, 1998 0.484 0.285 0.821 PRR C 0.484 0.285 0.821 O Barnett, 2002 1.000 0.818 1.223 O Orsel, 2007 0.818 0.634 1.057 O Eble, 2004 0.524 0.286 0.959 O Eble, 2007 1.000 0.833 1.200 O Eble, 2009 0.211 0.126 0.353 O Parida, 2007 1.055 0.853 1.306 PRR O 0.739 0.545 1.001 Overall PRR 0.665 0.511 0.866 0.01 0.1 1 10 100 Favors vaccine Favors no vaccineFig. 2. Forest plot of the relative risk (RR) of the FMD infection of swine for each of the 7 included studies, the pooled RR (PRR) per virus serotype and theoverall PRR together with the 95% confidence interval (CI).suggested an absence of correlation between study size signs is also important, because the virus can be excretedand study effect. This result was contradicted by Egger’s through the lesions. Preventing the appearance of lesionsregression test that suggested a strong association between might, therefore, decrease virus shedding. However, if vac-study size and study effect with an intercept = −2.84, a stan- cination induces protection against clinical signs withoutdard error = 0.54 and a P-value = 0.003. Duval and Tweedie’s virological protection, clinical disease can be concealedtrim and fill method suggested adding 3 studies to theright side of the funnel plot (Fig. 3), which would shift 0.0the effect from 0.59 (0.4–0.8) (the white diamond underthe X-axis), to 0.68 (0.5–0.99) (the black diamond under Standard Error 0.5the X-axis). Nonetheless, this adjustment would not alterthe result that emergency vaccination protected sheepagainst FMD infection significantly than non-vaccinated 1.0animals. 1.54. Discussion 2.0 The results indicate that emergency vaccination pro- -3 -2 -1 0 1 2 3tect cattle, swine and sheep against FMD clinical disease Log relative riskcompared to unvaccinated animals (Table 1). More impor-tantly, emergency vaccination provided protection against Fig. 3. Funnel plot of the logarithm pooled relative risk (RR) of 7 stud-virological infection with FMD in these species. The pur- ies (empty circles) quantifying the effect of FMD vaccination against FMDpose of an emergency vaccine is to limit the spread of infection in sheep. The dark spots are the potential missing studies accord- ing to Duval and Tweedie’s trim and fill method (if they had existed, thethe disease or in other words, decrease the reproduction pooled effect would have shifted from 0.59 (0.4–0.8); the white diamondratio to <1. Therefore, virological protection is more impor- under the X-axis, to be 0.68 (0.50–0.99); the black diamond under thetant than the clinical protection. Still, reducing the clinical X-axis).
  7. 7. T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9 7making early disease detection of vaccinated farms more In some of the studies included in the analysis, severaldifficult. periods of time between vaccination and virus challenge Despite the finding that emergency vaccination pro- were tested. A longer period between vaccination and chal-tects cattle against FMD infection, it was observed that lenge gives the immune system more time to developseveral cattle persisted as carriers. Some cattle were found neutralizing antibody titers, resulting in better protectionto excrete virus up to 553 days post-inoculation (Moonen (Doel et al., 1994; Salt et al., 1998). This could have createdet al., 2004). This will pose the risk of harboring and trans- bias in our study, because some studies used longer peri-mitting the virus for a long period, which could delay ods between vaccination and challenge compared to otherregaining FMD free status, and hence increase the eco- studies. Nevertheless, an emergency vaccine should pro-nomic damage due to the outbreak. This problem can vide rapid protection after vaccination (Cox and Barnett,be solved by culling all vaccinated animals after the epi- 2009) and hence, the use of long or short duration betweendemic. However, this might not be an option, because vaccination and challenge should not affect the efficacy ofmass culling has, in previous outbreaks, triggered public the vaccine. Because the pooled RRs of clinical and viro-concerns (Cox et al., 2005; Parida, 2009), which lead to logical protection from the current study were calculatedthe introduction of the so-called “vaccine-to-live policy”. from studies that included different periods between vac-Consequently, the OIE and EU regulations were adjusted, cination and challenge, and because it is actually unknownimplying that if a vaccine-to-live policy is used, countries when a virus could infect a herd after emergency vacci-can now regain the FMD free status 6 months instead of nation, the pooled RRs of the virological protection areone year after the last case has been detected (Council useful in simulation exercises to represent the efficacy ofDirective 2003/85/EC). Given the long period of potential the emergency vaccine. Simulation exercises are importantvirus excretion from the carrier animals, a strict post- to estimate the cost-effectiveness of FMD emergency vac-vaccination screening program is essential to find carrier cination during an outbreak and its effects on the nationalanimals and cull the infected herds to prevent subsequent economy. The results from such exercise may assist pol-outbreaks. icy makers to decide whether and how to use emergency First, we observed that sheep that had emergency vaccines during an outbreak.vaccination had 0.59 (0.4–0.8) lower risk of FMD infec- We included only studies with homologous virus chal-tion compared to non-vaccinated sheep. However, after lenge. Three published studies and two of the unpublishedthorough examination of publication bias, we noticed an experiments involved semi-homologous virus strain chal-existing trend in the studies (Fig. 3). Small studies showed lenge to the vaccine (Supplementary file, footnote l), butlarge effect, while large studies showed small effect. This were included, because the challenge virus strain and thecould be an indication of publication bias. This finding was vaccine virus antigen are categorized within the same topo-supported by other publication bias tests such as Egger’s type on the basis of their VP1 gene and are antigenicallyregression test, which has been reported to be powerful closely related (Barnett et al., 2001; Cox et al., 2006).in detecting publication bias, especially when the data is A major target of emergency vaccination is to reduceheterogeneous, such as ours (Peters et al., 2006). Correct- the reproduction ratio to <1. This means that an infecteding for publication bias by including the “missing” studies case will produce <1 secondary case during its entire infec-would move the RR toward the null effect. The “missing” tious period, which will cause the epidemic to fade out.studies could have been truly missing, conducted but not Recent advancements in statistics make it possible to mea-published, due to several reasons discussed thoroughly by sure the reduction of the reproduction ratio of FMD virusThornton and Lee (2000), Bornstein et al. (2009) and Halasa in vaccinated groups (Eblé et al., 2008). It is importantet al. (2009). More important than accepting or rejecting to quantify the effect of emergency vaccination on thethis bias was that the correction for the bias did not alter reproduction ratio of FMD virus for example by use of meta-the conclusion that there was significant protective effect analysis. Such a meta-analysis has been published, but wasof vaccinating sheep against FMD. Moreover, the confi- restricted to outcomes of 10 experiments on swine fromdence limits of the corrected and uncorrected pooled RR one project (Eblé et al., 2008). Therefore, it would be inter-overlap, making the difference between them statistically esting to estimate the reproduction ratio with and withoutinsignificant. emergency vaccination from all relevant studies and for In the current study, we defined infection to be a pos- the different species. Such a parameter would be useful foritive result based on VI, NSP or RT-PCR tests. This could modeling within herd spread in simulation models.have included bias, because some studies conducted more In the current meta-analysis, experimental data fromtests than other studies. When more tests are applied, the (Eblé et al., 2004, 2007; Orsel et al., 2005, 2007b) werechance of positive diagnosis (true or false) might become included only for vaccinated and non-vaccinated animalshigher. Nonetheless, the majority of studies applied the 3 challenged by direct injection of the virus. Moreover,tests (VI, RT-PCR and NSP), and all studies conducted VI, only experimental data using vaccinated and unvaccinatedwhich is considered the gold standard test to diagnose FMD challenged animals with the airborne route with non-infection (Paixão et al., 2008). In an outbreak situation, one vaccinated infected seeders were included in the analysispositive test result will most likely be interpreted as an FMD from Orsel et al. (2007a). This was carried out, becauseinfection and further measures will be taken according to the excluded experiments would represent the dynamicsthe EU Council Directive 2003/85/EC. Therefore, the tests of infection within a population rather than the efficacy ofdo not replace each other; likewise, they supplement each the vaccine. In the excluded experiments, the seeders wereother. vaccinated; this means that their infectiousness could be
  8. 8. 8 T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9lower compared to non-vaccinated seeders and hence dif- Conflict of interest statementferent than other studies used in the current meta-analysis.Moreover, infection of the contact animals in the excluded The authors declare that no financial or personal rela-experiments is a stochastic process, because the number tionships with other people or organizations could haveof infected animals to which a given contact animal is inappropriately influenced our is not predetermined by the experiment setup, asthe challenged seeders might be protected following vac- Acknowledgementscination. These are important differences between theseexperiments and the included studies in the current meta- The authors acknowledge Graham Belsham, Phaedraanalysis, and therefore it is necessary to exclude the results Eblé, Karin Orsel, Marvin Grubman, Canio Buonavoglia, Luisof such experiments to avoid bias. Rodriguez, Satya Parida, and Paul Barnett for the active Several attempts were made to restrict the heterogene- communication and the advice in relation to the inclusionity of studies, such as: Only including challenge studies of their studies in the analysis. We acknowledge Larry Pais-after emergency vaccination, controlled studies, challenge ley for proof reading the manuscript.studies with animals of the same age category, challenge This study was funded by the Directorate for Food, Agri-studies with a homologous virus, and inclusion of sym- culture and Fisheries, Denmark. Sarah Cox is financiallyposium studies and unpublished data. Nevertheless, the supported by Defra-UK.heterogeneity existed in most of the analyses. Includingonly studies published in English language could have been Appendix A. Supplementary dataa source of heterogeneity. Language could be a sourceof heterogeneity and bias, because non-English speaking Supplementary data associated with this articleresearchers might publish their positive results in English can be found, in the online version, at doi:10.1016/language to have more publicity, they would on the other j.prevetmed.2010.08.005.hand publish their negative results in their native language(Gregoire et al., 1995). Although attempts were made toexplain the heterogeneity using meta-regression with sev- Referenceseral potential variables, none of them was able to explain Aggarwal, N., Zhang, Z., Cox, S., Statham, R., Alexandersen, S., Kitching,it significantly. Another source of heterogeneity could have R.P., Barnett, P.V., 2002. Experimental studies with foot-and-mouthbeen the virus challenge dose. Studies that have challenged disease virus, strain O, responsible for the 2001 epidemic in the Unitedanimals by direct injection of the virus showed variability Kingdom. Vaccine 20, 2508–2515. Barnett, P.V., Samuel, A.R., Statham, R.J., 2001. The suitability of the emer-in the challenge dose (Cox and Barnett, 2009). Studies that gency foot-and-mouth disease antigens held by the internationalchallenged animals using the airborne route showed even vaccine bank within a global context. Vaccine 19, 2107–2117.more variability in the challenge procedure and challenge Barnett, P.V., Cox, S.J., Aggarwal, N., Gerber, H., McCullough, K.C., 2002. Further studies on the early protective responses of pigs followingintensity (Cox and Barnett, 2009). Unfortunately, because immunization with high potency foot-and-mouth disease vaccine.these variables varied largely between studies and given Vaccine 20, 3197–3208.the small number of studies included in some analyses, it Barnett, P.V., Keel, P., Reid, S., Armstrong, R.M., Statham, R.J., Voyce, C., Aggarwal, N., Cox, S.J., 2004. Evidence that high potency foot-and-was not possible to examine the effect of these variables mouth disease vaccine inhibits local virus replication and preventson the heterogeneity in the meta-regression. Nevertheless, the ‘carrier’ state in sheep. Vaccine 22, 1221–1232.the use of random effect model to calculate the pooled Begg, C.B., Mazumdar, M., 1994. Operating characteristics of a rank corre- lation test for publication bias. Biometrics 50, 1088–1101.RR took into account the existing heterogeneity to provide Bornstein, M., Hedges, L.V., Higgins, J.P.T., Rothstein, H.R., 2009. Introduc-the pooled RR by assuming a true RR per study instead of tion to Meta-analysis. John Wiley & Sons Ltd., United Kingdom.assuming only one true RR for all studies (Bornstein et al., Brehm, K.E., Kumar, N., Thulke, H.-H., Haas, B., 2008. High potency2009). vaccines induce protection against heterologous challenge with foot- and-mouth disease virus. Vaccine 26, 1681–1687. Buonavoglia, C., di Trani, L., Gramenzi, F., Zoletto, R., Lelli, R., Scacchia, M., 1988. Duration of immunity in swine inoculated with a monova-5. Conclusions lent foot-and-mouth disease oil emulsion vaccine. J. Vet. Med. B 35, 397–401. Comprehensive meta-analysis 2.0, 2009. Meta-analysis Manual. Biostat Emergency vaccination against FMD provided protec- Institute Inc., Englewood, NJ, USA.tion against clinical disease and against FMD infection Council Directive 2003/85/EC on community measures for the control ofin cattle, swine and sheep. For clinical protection, the foot-and-mouth disease repealing, 2003. Directive 85/511/EEC and amending directive 92/46/EEC. Official J. Eur. Union L306, 46, 22pooled RRs with the 95% confidence intervals were 0.13 November 2003.(0.09–0.18), 0.48 (0.36–0.65) and 0.31 (0.18–0.53), respec- Cox, S.J., Barnett, P.V., Dani, P., Salt, J.S., 1999. Emergency vaccination oftively. For virological protection, the pooled RRs with sheep against foot-and-mouth disease: protection against disease and reduction in contact transmission. Vaccine 17, 1858–1868.the 95% confidence intervals were 0.71 (0.59–0.85), 0.67 Cox, S.J., Voyce, C., Parida, S., Reid, S.M., Hamblin, P.A., Paton, D.J., Barnett,(0.51–0.87) and 0.59 (0.44–0.80), respectively. Fortunately, P.V., 2005. Protection against direct contact challenge following emer-no significant publication bias was identified in the differ- gency FMD vaccination of cattle and the effect on virus excretion from the oropharynx. Vaccine 23, 1106–1113.ent meta-analyses. Cox, S.J., Voyce, C., Parida, S., Reid, S.M., Hamblin, P.A., Hutchings, G., Paton, The current systematic review provided valid outcomes D.J., Barnett, P.V., 2006. Effect of emergency FMD vaccine antigen pay-that can be used in simulation models to examine the eco- load on protection, sub-clinical infection and persistence following direct contact challenge of cattle. Vaccine 24, 3184–3190.nomic consequences of applying emergency vaccination Cox, S.J., Parida, S., Voyce, C., Reid, S.M., Hamblin, P.A., Hutchings, G., Paton,during an FMD outbreak situation. D.J., Barnett, P.V., 2007. Further evaluation of higher potency vaccines
  9. 9. T. Halasa et al. / Preventive Veterinary Medicine 98 (2011) 1–9 9 for early protection of cattle against FMDV direct contact challenge. Mahul, O., Durand, B., 2000. Simulated economic consequences of foot- Vaccine 25, 40–50. and-mouth disease epidemics and their public control in France. Prev.Cox, S.J., Barnett, P.V., 2009. Experimental evaluation of foot-and-mouth Vet. Med. 47, 23–38. disease vaccines for emergency use in ruminants and pigs: a review. Mattion, N., K’onig, G., Seki, C., Smitsaart, E., Maradei, E., Robiolo, B., Duffy, Vet. Res. 40, 13–43. S., Leon, E., Piccone, M., Sadir, A., Bottini, R., Cosentino, B., Falczuk, A.,Doel, T.R., Williams, L., Barnett, P.V., 1994. Emergency vaccination against Maresca, R., Periolo, O., Bellinzoni, R., Espinoza, A., La Torre, J., Palma, foot-and-mouth disease: rate of development of immunity and its E.L., 2004. Reintroduction of foot-and-mouth disease in Argentina: implication for the carrier state. Vaccine 12, 592–600. characterisation of the isolates and development of tools for the con-Donaldson, A.I., Kitching, R.P., 1989. Transmission of foot-and-mouth dis- trol and eradication of the disease. Vaccine 22, 4149–4162. ease by vaccinated cattle following natural challenge. Res. Vet. Sci. 46, Moonen, P., Jacobs, L., Crienen, A., Dekker, A., 2004. Detection of carriers of 9–14. foot and mouth disease virus among vaccinated cattle. Vet. Microbiol.Dohoo, I.R., Martin, S.W., Stryhn, H., 2003. Veterinary Epidemiologi- 103, 151–160. cal Research. Atlantic Veterinary College Inc., Prince Edward Island, Orsel, K., Dekker, A., Bouma, A., Stegeman, J.A., de Jong, M.C.M., 2005. Vac- Canada. cination against foot-and-mouth disease reduces virus transmissionDuval, S., Tweedie, R., 2000. A non-parametric “trim and fill” method of in groups of calves. Vaccine 23, 4887–4894. accounting for publication bias in meta-analysis. J. Am. Stat. Assoc. 95, Orsel, K., de Jong, M.C.M., Bouma, A., Stegeman, J.A., Dekker, A., 2007a. 89–98. Foot-and-mouth disease virus transmission among vaccinated pigsEblé, P.L., Bouma, A., de Bruin, M.G.M., van Hemert-Kluitenberg, F., Van after exposure to virus shedding pigs. Vaccine 25, 6381–6391. Oirschot, J.T., Dekker, A., 2004. Vaccination of pigs two weeks before Orsel, K., Dekker, A., Bouma, A., Stegeman, J.A., de Jong, M.C.M., 2007b. infection significantly reduces transmission of foot-and-mouth dis- Quantification of foot-and-mouth disease virus excretion and trans- ease virus. Vaccine 22, 1372–1374. mission within groups of lambs with and without vaccination. VaccineEblé, P.L., Bouma, A., Weerdmeester, J.A., Stegeman, J.A., Dekker, A., 2007. 25, 2673–2679. Serological and mucosal immune responses after vaccination and Orsel, K., de Jong, M.C.M., Bouma, A., Stegeman, J.A., Dekker, A., 2007c. The infection with FMDV in pigs. Vaccine 25, 1043–1054. effect of vaccination on foot-and-mouth disease virus transmissionEblé, P.L., Koeijer, A.A., de Jong, M.C.M., Engel, B., Dekker, A., 2008. A among dairy cows. Vaccine 25, 327–335. meta-analysis quantifying transmission parameters of FMD strain O Orwin, R., 1983. A fail-safe N for effect size in meta-analysis. J. Educ. Stat. Taiwan among non-vaccinated and vaccinated pigs. Prev. Vet. Med. 8, 157–159. 83, 98–106. Paixão, T.A., Carvalho Neta, A.V., Paiva, N.O., Reis, J.R., Barbosa, M.S., Serra,Eblé, P.L., Weerdmeester, K., van Hemert-Kluitenberg, F., Dekker, A., 2009. C.V., Silva, R.R., Beckham, T.R., Martin, B.M., Clarke, N.P., Adams, L.G., Intradermal vaccination of pigs against FMD with 1/10 dose results in Santos, R.L., 2008. Diagnosis of foot-and-mouth disease by real time comparable vaccine efficacy as intramuscular vaccination with a full reverse transcription polymerase chain reaction under field condi- dose. Vaccine 27, 1272–1278. tions in Brazil. BMC Vet. Res. 4, 53–59.Egger, M., Davey Smith, G., Schneider, M., Minder, C., 1997. Bias in meta- Parida, S., Fleming, L., Oh, Y., Mahapatra, M., Hamblin, P., Gloster, J., Doel, analysis detected by a simple graphical test. BMJ 315, 629–634. C., Gubbins, S., Paton, D.J., 2007. Reduction of foot-and-mouth diseaseGibson, C.F., Donaldson, A.I., Ferris, N.P., 1984. Responses of sheep (FMD) virus load in nasal excretions, saliva and exhaled air of vacci- vaccinated with large doses of vaccine to challenge by airborne foot- nated pigs following direct contact challenge. Vaccine 25, 7806–7817. and-mouth disease virus. Vaccine 2, 157–161. Parida, S., Fleming, L., Oh, Y., Mahapatra, M., Hamblin, P., Gloster, J., Paton,Goris, N., Merkelbach-Peters, P., Diev, V.I., Verloo, D., Zakharov, V.M., Kraft, D.J., 2008. Emergency vaccination of sheep against foot-and-mouth H.-P., De Clercq, K., 2007. European Pharmacopoeia foot-and-mouth disease: significance and detection of subsequent sub-clinical infec- disease vaccine potency testing in cattle: between test variability and tion. Vaccine 26, 3469–3479. its consequences. Vaccine 25, 3373–3379. Parida, S., 2009. Vaccination against foot-and-mouth disease virus: strate-Goris, N., Maradeib, E., D’Aloiab, R., Fondevila, N., Mattiond, N., Perezb, gies and effectiveness. Exp. Rev. Vaccines 8, 347–365. A., Smitsaarte, E., Nauwynck, H.J., La Torred, J., Palmag, E., De Clercq, Pendell, D.L., Leatherman, J., Schroeder, T.C., Alward, G.S., 2007. The eco- K., 2008. Foot-and-mouth disease vaccine potency testing in cattle nomic impacts of a foot-and-mouth disease outbreak: a regional using homologous and heterologous challenge strains: Precision of analysis. J. Agric. Appl. Eco. 39, 19–33. the “Protection against Podal Generalisation” test. Vaccine 26, 3432– Peters, J.L., Sutton, A.J., Jones, E.R., Abrams, K.R., Rushton, L., 2006. Com- 3437. parison of two methods to detect publication bias in meta-analysis.Graves, J.H., McKercher, P.D., Farris Jr., H.E., Cowan, K.M., 1968. Early JAMA 295, 676–680. responses of cattle and swine to inactivated foot-and-mouth disease Salt, J.S., Williams, L., Statham, R., Barnett, P.V., 1995. Further studies on vaccine. Res. Vet. Sci. 9, 35–40. the rate of development of protection in cattle given emergency vac-Gregoire, G., Derderian, F., LeLorier, J., 1995. Selecting the language of the cination against FMD. Rep. Session Res. Group Standing Techn. Comm. publications included in a meta-analysis: is there a tower of Babel Eur. Comm. Contr. 17, 90–97, FMD, Modelling, Denmark, FAO, Rome, bias? J. Clin. Epidemiol. 48, 159–163. Appendix.Halasa, T., Østerås, O., van Werven, T., Hogeveen, H., Nielen, M., 2009. Salt, J.S., Barnett, P.V., Dani, P., Williams, L., 1998. Emergency vaccination of Meta-analysis on dry cow management for dairy cattle. Part 1. pigs against foot-and-mouth disease: protection against disease and Protection against new intramammary infections. J. Dairy Sci. 92, reduction in contact transmission. Vaccine 16, 746–754. 3134–3149. Sellers, R.F., Herniman, K.A.J., 1974. Early protection of pigs against foot-Lean, I.J., Rabiee, A.R., Duffield, T.F., Dohoo, I.R., 2009. Use of meta-analysis and-mouth disease. Br. Vet. J. 130, 440–445. in animal health and reproduction: methods and applications. J. Dairy Thornton, A., Lee, P., 2000. Publication bias in meta-analysis: its causes Sci. 92, 3545–3565. and consequences. J. Clin. Epidemiol. 53, 207–216.