HLA–B51/B5 and the risk of Behçet's disease: A systematic review and meta-analysis of case–control genetic association studies




To quantify by meta-analysis the genetic effect of the HLA–B5 or HLA–B51 (HLA–B51/B5) allele on the risk of developing Behçet's disease (BD) and to look for potential effect modifiers.


Relevant studies were identified using the PubMed Medline database and manual searches of the literature. Pooled odds ratios (ORs) and 95% confidence intervals (95% CIs) were calculated by using the random-effects model. Subgroup meta-analyses and meta-regression analyses were undertaken to investigate the effects of selected study-level parameters on the pooled OR. Heterogeneity was assessed using the I2 statistic. Pooled results were used to calculate population-attributable risks (PAR) for BD in relationship to HLA–B51/B5.


A total of 4,800 patients with BD and 16,289 controls from 78 independent studies (published 1975–2007) were selected. The pooled OR of HLA–B51/B5 allele carriers to develop BD compared with noncarriers was 5.78 (95% CI 5.00–6.67), with moderate between-study heterogeneity (I2 = 61%). The subgroup analyses stratifying studies by geographic locations (Eastern Asia, Middle East/North Africa, Southern Europe, Northern/Eastern Europe) yielded consistent OR ranges (5.31–7.20), with I2 ranges of 52–70%. Univariate random-effects meta-regression indicated the percentage of male BD cases (P = 0.008) as a source of heterogeneity. The PAR within the various geographic areas were estimated at 32–52%.


The strength of the association between BD and HLA–B51/B5, and its consistency across populations of various ethnicities, lends further support to this allele being a primary and causal risk determinant for BD. Variations according to sex support an interaction of this allele with BD characteristics.


Behçet's disease (BD) is a chronic vasculitis of arteries and veins of all sizes characterized by recurrent oral ulcers, genital ulcers, ocular and skin involvement, and other multisystemic features. Despite a worldwide distribution, BD clusters in an area that extends from far Eastern Asia to the Mediterranean basin (1, 2).

Several lines of evidence suggest that host genetic factors play a pivotal role in determining susceptibility to BD. In particular, it has long been known that BD is associated with the major histocompatibility complex HLA–B5 allele (3) and, more specifically, with HLA–B51 (4), the predominant split of the HLA–B5 broad antigen (5–7). While the overrepresentation of HLA–B5 or HLA–B51 (HLA–B51/B5) among individuals with BD has been abundantly replicated, substantial between-study differences were found in the strengths of this genetic association with reported risk increases that ranged between 1.3 and 16 (2, 8). It is likely that such disparities result from small sample sizes, but a true variability of the relationship between BD and HLA–B51/B5 might also exist across populations of different ethnicities or clinical subtypes of BD (2, 9).

We undertook a systemic literature review and meta-analysis of genetic association case–control studies with the aim of estimating the true risk increase for BD development associated with HLA–B51/B5 carriage and identifying clinical or methodologic factors potentially influencing the effect size of this genetic association.


Identification and selection of studies.

Case–control studies in which the HLA class I genotypes had been linked with BD were identified by electronic searches of the PubMed Medline database between January 1, 1973 and December 31, 2007. Terms used for the search contained combinations of the medical subject heading terms “Behçet's syndrome” and “HLA antigens” (or alternatively “histocompatibility”), or several combinations of text search terms including “Behçet's disease” (or “Behçet's syndrome”) with “HLA–B5” (or “HLA–B51”) and with “genetic association” (or “case–control study”). Studies were also manually searched in the reference lists of retrieved articles, in review articles on BD, in textbooks and conference proceedings, by communication with experts, and through additional Internet searches. Full publications and abstracts were allowed and no language restriction was imposed.

To be included in the meta-analysis, a study had to be designed as a case–control study and had to provide sufficient information to construct a 2 × 2 contingency table for the frequencies of HLA–B51/B5 genotypes in BD cases and controls. Anticipating the potential issue that several studies include to some extent the same cases and/or controls, we paid close attention to authors, study subjects' geographic locations, and numbers of study subjects. Unless these precisions were unequivocally stated in the cross-references of the publications, for multiple studies stemming from the same research teams, we contacted the authors asking them to notify us of such an overlap between cases and/or controls among their studies. With multiple overlapping publications, we generally selected the publication that contained the most BD cases.

Data extraction.

Using a study-specific questionnaire, 2 investigators (MdM and AM) independently extracted data from the primary studies. For articles written in languages not spoken by either of these 2 investigators, data were extracted with the help of native or fluent speakers of the respective languages. The following data were collected: authors, publication year, journal, publication type and language, study population location, genotyping method, allele genotyped (HLA–B5 and/or HLA–B51), numbers of cases and controls, definitions or classification criteria used for BD, BD sample description, control sample description, and numbers and/or percentages of HLA–B51/B5–positive cases and controls. Results were compared, and disagreements were resolved by discussion and consensus. The primary author of one study was contacted for clarification and provided additional data on the number of investigated subjects.

Statistical analysis.

For each study, we computed odds ratios (ORs) and 95% confidence intervals (95% CIs) using normal approximation to the log10 of the OR. Data from the entire study populations were pooled to compare HLA–B51/B5 frequencies between BD patients and control subjects. Pooled ORs and 95% CIs were calculated using random-effects inverse-variance models (10). Between-study heterogeneity was assessed using the DerSimonian–Laird chi-square–based Q statistics (11) and quantified using the I2 statistic (12). Wald's chi-square tests were performed to assess between-subgroup differences of pooled ORs.

Random-effects normal-logistic models were used to calculate pooled estimates for the frequency of the HLA–B51/B5 allele in BD cases and control populations. This approach was recently used to calculate the prevalence of rare adverse events in colorectal cancer screening studies (13).

Stratified analysis by subgroup was performed to search for modifiers possibly affecting the observed genetic effect and to investigate the reasons for heterogeneity among studies. The 7 prespecified categorical study-level characteristics for assessment of sources of inter-study heterogeneity were: geographic area (classified into 5 regions: Eastern Asia, Middle East/North Africa, Southern Europe, Northern/Eastern Europe, and North America); allele genotyped (HLA–B5 or HLA–B51); serologic (i.e., complement-dependent microlymphocytoxicity assays) and molecular genotyping techniques (i.e., sequence-specific oligonucleotide or primer polymerase chain reaction); classification criteria used to define BD; publication type (defined as peer-reviewed journal versus publication in conference proceedings or books); publication language (defined as English versus non-English); and ethnic matching of controls to cases (defined as specifically mentioned versus not mentioned).

To further explore heterogeneity in study findings, we used univariate random-effects inverse-variance meta-regression analyses (10). The prespecified explanatory variables included the same 7 categorical covariates used for subgroup analyses (see above), plus the following 7 prespecified continuous covariates: mean or median age at disease onset or diagnosis for BD cases; mean or median age at study time for BD cases; percentage of BD cases in men, and with ocular, large-vessel, or central nervous system disease; and year of publication.

Funnel plot analysis and the arcsine test for binary outcomes (14) were used to detect publication bias. In addition, we also assessed publication bias with the Egger's regression asymmetry test, the Begg-Mazumdar adjusted rank correlation test, and the test proposed by Peters et al (15–17). To further examine the possibility of “small study effects,” we conducted 3 sensitivity analyses restricted to the studies that included ≥30 BD cases, ≥50 BD cases, and ≥100 subjects (cases plus controls) in total.

To assess the contribution of HLA–B51/B5 to BD development in the overall population, we calculated specific population-attributable risks (PAR) for HLA–B51/B5 carriage using the following standard formula:

equation image

where p is the percentage of HLA–B51/B5 allele-carrying BD cases (18). Relative risk in this formula was approximated by the OR.

All statistics were computed with SAS software, version 9.1 (SAS Institute, Cary, NC). All statistical tests were 2-tailed and P values less than 0.05 were considered statistically significant, except publication bias tests, for which P values less than 0.10 were considered statistically significant.


Eligible studies.

Figure 1 shows the flow chart of publications identified by the literature search to those finally retained. A total of 149 publications met the eligibility criteria for inclusion in the meta-analysis. Among those, 71 publications were subsequently excluded because they were considered to contain duplicate information or overlaps in case and/or control samples. Overlapping was primarily verified on the basis of cross-references and through the information provided by the contacted authors. In some instances, these decisions were made based on the concordances of authors, study areas, years of publication, and numbers of study subjects.

Figure 1.

Flow chart of studies identified through the systematic review of the literature. BD = Behçet's disease.

Eventually, we retained 78 independent, nonoverlapping studies meeting the selection criteria (4, 19–95). These studies were published between 1975 and 2007, and included 1 electronic publication (print version became available in 2008) (26) and 19 publications written in non-English languages (20, 30, 34, 36, 38, 40–42, 46, 49, 60, 65, 70, 77, 81, 82, 88, 89, 95). The relevant data in 2 articles were based on 2 distinct geographic sites and ethnic populations (19, 53), each of which was included as a separate population in the meta-analysis. Consequently, the 78 selected studies contributed to 80 distinct study populations.

The final data set included a total of 4,800 cases and 16,289 controls. The case number did not include a subgroup of immigrant patients with BD analyzed in one study because no ethnically representative control group had been considered (40). From another study that used the 1974 Japanese BD Research Committee (JBDRC) criteria to define BD (96), we did not use the subsets of cases with “possible” and “probable” BD because we considered the diagnostic specificity of these subgroups to be too low (84). From another study that contained several familial cases, we retained only the data for the unrelated subjects with BD (31).

Study characteristics.

The principal study characteristics are shown in Figure 2. The geographic areas from which the 80 study populations had been recruited are further summarized in Table 1. For rare studies reporting results for both HLA–B5 and HLA–B51 genotypes (4, 25, 42, 80, 93), only the latter data were used for all analyses. The classification criteria used were: the International Study Group for Behçet's Disease (97), n = 36; (19–22, 26, 28, 32, 34, 43, 44, 48, 50–55, 57, 63, 64, 67, 68, 72, 74–76, 78, 81, 87–89, 91, 92, 95); 1974/1987 revised JBDRC (96, 98), n = 16 (4, 24, 29, 31, 33, 45, 47, 60–62, 70, 71, 79, 84, 85, 94); O'Duffy (99), n = 6 (25, 27, 40, 66, 69, 93); Mason and Barnes (100), n = 4 (23, 39, 42, 56); Cheng and Zhang (101), n = 1 (82); and multiple sets, n = 5 (30, 38, 46, 58, 73). For 12 study populations, no statements were made with respect to the criteria used to define BD cases (35–37, 49, 59, 65, 77, 80, 83, 86, 90). Matching criteria for controls were cited for 52 study populations and included ethnicity (n = 50), age (n = 7), and/or sex (n = 6).

Figure 2.

Forest plot of results of published studies on the association between HLA–B51/B5 and Behçet's disease (BD) (listed chronologically). Odds ratios (ORs) for the risk of HLA–B51/B5 allele carriers to develop BD compared with noncarriers. The dashed vertical line indicates the pooled value for the entire population considered. *See Materials and Methods for a list of other criteria applied. 95% CI = 95% confidence interval; E = Eastern; Serol. = serologic; JBDRC = Japanese Behçet's Disease Research Committee; N = North; M = Middle; NR = not reported; S = Southern; ISG = International Study Group; Molec. = molecular.

Table 1. Geographic locations of the study populations (n = 80)
Geographic area/countryPopulations, no.Ref.
Eastern Asia  
 China561, 81, 82, 88, 89
 Japan144, 49, 50, 60, 62, 65, 68, 70, 79, 84–86, 91, 92
 Korea547, 51, 57, 58, 72
Middle East/North Africa  
 Iran321, 64, 80
 Israel324, 31, 45
 Jordan/Palestine228, 87
 Saudi Arabia190
 Tunisia246, 48
 Turkey1319, 22, 26, 27, 37, 43, 44, 53, 55, 63, 66, 74, 83
Southern Europe  
 Greece354, 77, 94
 Italy629, 39, 71, 73, 75, 78
 Spain430, 38, 42, 67
Northern/Eastern Europe  
 Germany336, 53, 95
 Russia320, 41, 76
 UK419, 35, 59, 93
North America  

Aggregate information for one or several demographic and clinical characteristics of the BD cases (and which were given for the exact number of genotyped individuals) could be retrieved from 49 studies. These included information on mean/median age at disease onset/diagnosis (range 28–34 years) in 5 studies (28, 30, 32, 51, 54); mean/median age at study time (range 27–49.5 years) in 20 studies (22, 27, 39, 44, 46, 47, 51, 54, 55, 57, 58, 68, 69, 72, 81, 86, 88, 91, 93, 94); percentages of males (range 28.6–90.5%) in 44 studies (22, 26–28, 30–33, 36, 39, 41, 43–47, 51, 52, 54–58, 60–62, 65, 66, 68–71, 75, 81–83, 85, 86, 88, 91–95); and percentages of cases with ocular involvement (range 20–100%) in 32 studies (22, 24, 27, 29–31, 33, 35, 39, 41, 45–48, 51, 52, 55–57, 65, 66, 68, 70–73, 75, 82, 83, 88, 93, 94), with central nervous system involvement (range 2.8–50%) in 16 studies (22, 24, 30, 31, 39, 46–48, 51, 55–57, 72, 73, 75, 94) and with large-vessel involvement (range 5.6–35.7%) in 16 studies (22, 24, 27, 30, 31, 39, 41, 47, 51, 56, 57, 66, 72, 73, 75, 94).

Quantitative data synthesis.

The ORs for BD susceptibility associated with HLA–B51/B5 carriage are shown in Figure 2. Results are shown for all 80 separate case–control data sets, and the pooled result across all the data sets. The ORs for HLA–B51/B5 carriers to develop BD across the primary data sets were all >1, with values ranging from 1.18 (69) to 34.62 (75). For 2 studies with no controls having positive findings for HLA–B51/B5 (52, 75), a single positive control was added to allow OR and 95% CI computations.

The pooled OR for BD susceptibility was 5.78 (95% CI 5.00–6.67). Pooled ORs within predefined subgroups consistently yielded positive ORs with overlapping 95% CIs (Table 2). For the overall estimate (I2 = 61%), and within most of the subgroups (I2 = 38–78%), analyses showed moderate heterogeneity.

Table 2. Pooled estimates for overall and subgroup meta-analyses for HLA–B51/B5 carriage and its association with BD risk*
SubgroupsPopulations, no.Pooled prevalence for HLA–B51/B5OR (95% CI)I2 (%)PhetPcov
BD cases (95% CI)Controls (95% CI)
  • *

    BD = Behçet's disease; 95% CI = 95% confidence interval; OR = odds ratio; Phet = P values for heterogeneity statistics; Pcov = P values for significance of corresponding covariates in the pooled genetic effect (calculated by random-effects meta-regression).

  • Pooled prevalence values were calculated using random-effects normal-logistic models.

  • Two studies combined in the North American group had distinctly different ethnicities.

Overall8057.2 (53.4–60.9)18.1 (16.1–20.3)5.78 (5.00–6.67)60.60.0001 
By geographic area      0.31
 Eastern Asia2555.0 (49.8–60.1)19.6 (16.0–23.7)5.18 (4.15–6.47)52.20.001 
 Middle East/North Africa2763.5 (58.8–68.0)21.7 (18.2–25.7)6.25 (4.87–8.03)70.40.0001 
 Southern Europe1560.6 (51.9–68.7)16.8 (13.3–21.0)7.20 (4.89–10.62)57.20.003 
 Northern/Eastern Europe1139.0 (28.2–51.1)11.2 (8.1–15.3)5.31 (3.35–8.40)55.60.013 
 North America234.2 (6.0–80.8)18.0 (7.6–37.1)2.35 (0.56–9.82)57.00.13 
By genotype      0.81
 HLA–B5150  5.90 (4.87–7.16)66.80.0001 
 HLA–B530  5.64 (4.57–6.95)45.10.005 
By genotyping technique      0.07
 Serologic49  6.31 (5.23–7.60)59.30.0001 
 Molecular23  4.74 (3.71–6.05)65.10.0001 
By classification criteria      0.09
 International Study Group36  5.31 (4.38–6.43)61.30.0001 
 Japanese BD Research  Committee16  7.48 (5.25–10.64)57.90.002 
By publication type      0.45
 Peer-reviewed journal72  5.66 (4.88–6.57)58.00.0001 
 Conference proceedings/book8  7.20 (4.08–12.71)77.80.0001 
By publication language      0.16
 English61  6.12 (5.18–7.23)63.90.0001 
 Non-English19  4.66 (3.60–6.04)38.20.05 
By ethnic matching of controls      0.37
 Specifically mentioned in the  publication50  6.10 (5.05–7.36)62.50.0001 
 Not specified30  5.29 (4.25–6.60)57.80.0001 

The random-effects pooled prevalence for HLA–B51/B5 in BD cases and control populations were 57.2% (95% CI 53.4–60.9%) and 18.1% (95% CI 16.1–20.3%), respectively (Table 2). Compared with case and control populations from Eastern Asia, the Middle East/North Africa, and Southern Europe, the pooled percentages of HLA–B51/B5 carriage were 1.4–1.9-fold lower for cases and controls from Northern/Eastern Europe.

Bias and heterogeneity diagnostics.

Weights attributed to the 80 individual data sets ranged from 0.15–4.87%, suggesting that none of the individual case–control populations had an unduly large effect on the pooled OR (Figure 2).

The results of the meta-regression analyses assessing the impact of the selected categorical and continuous covariates on the genetic association are shown in Tables 2 and 3, respectively. These analyses identified only the percentages of male BD cases as significantly affecting the genetic effect (Pcovariate = 0.008). However, compared with the unadjusted model (i.e., the model that did not include any covariate), adjustment for sex did not markedly lower heterogeneity in the model (detailed result not shown).

Table 3. Univariate random-effects meta-regression assessing potential effects of selected continuous covariates in the observed pooled effect of HLA–B51/B5 on the risk of developing BD*
VariablePopulations, no.dfOR (95% CI)Pcov
  • *

    The ORs shown in the table can be interpreted as the change in the odds for any one-unit increase of the corresponding covariates. BD = Behçet's disease; OR = odds ratio; 95% CI = 95% confidence interval; Pcov = P values for significance of covariates in the pooled genetic effect; CNS = central nervous system.

BD case characteristics    
 Male sex, %4411.02 (1.004–1.03)0.008
 Age at study time, years2011.00 (0.95–1.06)0.96
 Age at onset/diagnosis, years510.95 (0.70–1.29)0.74
 Ocular involvement, %3211.01 (0.999–1.02)0.08
 CNS involvement, %1611.02 (0.99–1.05)0.14
 Large-vessel involvement, %1611.03 (0.99–1.07)0.10
Year of publication8010.99 (0.98–1.01)0.44

The funnel plot distribution had a slightly asymmetric shape (Figure 3), and the tests proposed by Egger (P = 0.02) and Begg-Mazumadar (P = 0.10) were statistically significant, raising the possibility of a publication bias. However, the arcsine test (P = 0.13) and the test proposed by Peters (P = 0.12) did not reach statistical significance. In addition, the sensitivity analyses based only on the studies that included ≥30 BD cases, ≥50 BD cases, or a total sample size of ≥100 cases and controls, showed only a minor reduction of the pooled OR at 5.53 (95% CI 4.73–6.48; 54 comparisons), 5.38 (95% CI 4.56–6.35; 39 comparisons), and 5.71 (95% CI 4.94–6.60; 66 comparisons), respectively.

Figure 3.

Funnel plot for publication bias. The dashed vertical line indicates the pooled estimate. Precision was assessed by the inverse variance method.

PAR of HLA-B51/B5 for BD.

PAR were calculated using the pooled OR obtained for the entire sample. Owing to the geographic variations of the percentages of BD cases carrying the HLA–B51/B5 allele, PAR were computed for geographic strata. The resulting PAR were 44.4% for Eastern Asia, 49.4% for Middle East/North Africa, 52.2% for Southern Europe, and 31.7% for Northern/Eastern Europe.


The pooled estimates of this meta-analysis of 4,800 cases and 16,289 controls indicate that the risk (“odds”) of HLA–B51/B5 carriers developing BD is increased by a factor of 5.78, and by a factor of 5.90 when considering only the split antigen HLA–B51. In agreement with previous observations about meta-analyses on a variety of genetic associations (102), we found that the pooled rates of HLA–B51/B5–positive BD cases varied across geographic locations, but the relative risk increases associated with this allele appeared to be fairly even for different ethnic populations. An additional important finding concerns the moderate heterogeneity, which we were able to partly explain by variations of sex distributions in the studied BD populations.

This meta-analysis was undertaken to gain further insight into the long known relationship between BD and the HLA–B51/B5 allele. The considerable range of the reported effect sizes, which became even more obvious during this investigation, has prevented accurate appraisal of the true impact this allele exerts on BD susceptibility. Another area of uncertainty was highlighted by the suggestion that this genetic association might hold true only for areas of high BD prevalence (2). Hence, the critical question persisted whether HLA–B51/B5 represented the disease-causing factor or simply reflected linkage disequilibrium with a nearby and truly causal BD gene (8, 9, 103, 104).

Evaluation of our meta-analysis findings against the strength and consistency criteria for causal inference for gene–disease associations (105, 106) adds epidemiologic support to the concept that HLA–B51/B5 itself is the BD-susceptibility allele. The likelihood that a given genetic association takes account of confounding (e.g., by linkage disequilibrium) diminishes with the increasing strengths of the association, and an unbiased OR of at least 3 to 5 was advanced as being indicative of causality (106). In addition, all 80 analyzed study populations consistently produced ORs >1, and our finding of a lack of a significant modifying effect of the ethnic background further weakens the hypothesis of linkage disequilibrium. Another criterion to infer a causal role of HLA–B51/B5 in BD is based on biologic plausibility. HLA class I molecules play a central role in the immune system and, despite the lack of direct experimental evidence for this theory, it is hypothesized that HLA–B51/B5 triggers BD by means of selective binding of pathogenic peptides (8, 9, 103, 104).

To assess the absolute impact the HLA–B51/B5 allele might have on BD development, we calculated the PAR. This computed percentage suggested that the HLA–B51/B5 allele accounts for 32–52% of BD cases within various geographic subgroups. These values are substantially higher than a previous estimate derived from a family- based genetic association study, which indicated that 19% of the genetic susceptibility to BD would be explained by this allele (107).

As is common to meta-analyses of genetic association studies, statistical heterogeneity was found (108). Consequently, all our risk estimates were calculated using random-effects models. Although random-effects models generate confidence intervals with wider boundaries, the lower 95% confidence limits still largely supported positive odds of BD susceptibility for the HLA–B51/B5 carriers for the entire sample and within the analyzed subgroups (Table 2). Therefore, we think that the underlying heterogeneity should not have major repercussions on the interpretation of the overall findings of our meta-analysis, but this situation strongly highlights the influence of modifiers on the association of BD with HLA–B51/B5.

Pertinently, our meta-regression analyses pinpointed the differing sex distributions as a significant source of between-study heterogeneity, although this result has to be interpreted keeping in mind the possibility of spurious statistical significance due to testing of multiple hypotheses. Nevertheless, this finding agrees with reports of HLA–B51/B5 carriage predominantly in male BD cases (32, 52, 88, 95), and might reflect either a sex-specific genetic effect or confounding, since both male sex and HLA–B51/B5 carriage may be markers of more severe BD (1, 2). In contrast, despite the reports that HLA–B51/B5 positivity determines the propensity for developing such manifestations (2, 9), variations of percentages of BD cases with eye, central nervous system, or large-vessel disease did not appear to be statistically significant sources of heterogeneity. However, these calculations might have been underpowered in our analyses. The observed trend toward a significant variation according to genotyping technique is counterintuitive in light of the relatively small misassignment rates reported for HLA class 1 typing with serologic compared with molecular techniques (109, 110).

The unresolved heterogeneity could also reflect biases in the reported strengths of the association between HLA–B51/B5 and BD. Indeed, graphic (Figure 3) and statistical methods indicated the possibility of a publication bias, even though sensitivity analyses suggested that the impact of “small study effects” on our overall results was not very strong. Alternatively, it can be hypothesized that the heterogeneity derives from epistatic or environmental risk determinants on the contribution of HLA–B51/B5 to BD development distributed unequally across BD populations. However, such an interaction with HLA–B51/B5 has not yet been identified for any of the other potential genetic or environmental BD-susceptibility factors (8, 9, 103, 104).

Our study has limitations. The allowance for inclusion of studies investigating the association of BD with HLA–B5 is arguable because this allelic group also recognizes the HLA–B52 split, which is not linked to BD (8, 103). Hence, as suggested by the results of the subgroup meta-analysis for the individual effects of HLA–B5 and HLA–B51, our overall pooled result likely represents an underestimation, albeit small, of the true susceptibility associated with this genetic factor. The claim of consistency of the HLA–B51/B5 effect on BD risk across ethnic groups needs to be understood in the context that this effect remains to be demonstrated for the North American continent—there were only 2 eligible studies on ethnically diverse populations from the US and Mestizo Mexicans—and for other continents for which no such data have yet become available. Furthermore, because the analyzed studies were reported over a period of over 3 decades, they could not be evaluated for their compliances with current quality standards for genetic case–control association studies (e.g., control for population stratification) (111). To address this possible shortcoming, we used publication type, publication language, and year of publication as possible surrogate indices of study quality, and we conducted a subgroup meta-analysis separating publications according to whether or not they mentioned ethnic matching of case and control samples.

This comprehensive analysis of the effect of the HLA–B51/B5 allele on the risk of developing BD further illustrates this disease–gene association as one of the tightest known for a complex human trait. Future research should be directed toward unravelling the potential interactions of this allele with other genetic and nongenetic risk factors of BD and with BD phenotype.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Mahr had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. De Menthon, Mahr.

Acquisition of data. De Menthon, Maldini, Mahr.

Analysis and interpretation of data. De Menthon, LaValley, Guillevin, Mahr.


We are indebted to the following physicians and researchers for their help in identification and selection of relevant studies: M. Accorinti, MD (Rome, Italy); Z. Alekberova, MD (Moscow, Russia); G. Azizlerli, MD (Istanbul, Turkey); T. Chajek-Shaul, MD (Jerusalem, Israel); M. Chen (Beijing, China); J. A. Correia, MD (Porto, Portugal); J. Crespo, MD (Coimbra, Portugal); F. Davatchi, MD (Tehran, Iran); M. F. Gonzalez-Escribano, PhD (Sevilla, Spain); J. Graña (Coruña, Spain); A. Gül, MD, PhD (Istanbul, Turkey); H. Houman, MD (Tunis, Tunisia); S. K. Kim, MD (Daegu, South Korea); S. Kobayashi, MD (Tokyo, Japan); T. Lehner, MD (London, UK); W. Madanat, MD (Amman, Jordan); S. Ohno, MD, PhD (Sapporo, Japan); G. Palimeris, MD (Athens, Greece); S. Perrone, MD (Padua, Italy); P. Pivetti-Pezzi, MD (Rome, Italy); C. Salvarani (Reggio Emilia, Italy); J. Vaz-Patto, MD (Lisbon, Portugal); G. Wallace, MD (Birmingham, UK); H. Yazici, MD (Istanbul, Turkey); C. C. Zouboulis, MD (Dessau, Germany). We are also grateful to Drs. R. Seror, L. Teixeira, K. Yasuda, and Y. Zhang for their help in the translation of articles, and to Ms Janet Jacobson and Dr. Andrew Wang for editorial assistance.