Background Substantial geographical differences have been observed among prevalence surveys of Behçet's disease (BD). However, the magnitude of such variations has not been fully assessed and whether some of these differences in fact reflect methodological inconsistencies is unclear.
Objectives To study by meta-analysis the effect of geographic location and differences in study methodology on the prevalence of BD.
Methods We performed a systematic literature review to identify population-based prevalence surveys of BD in PubMed MEDLINE. Studies were eligible if they provided sufficient data to recalculate the prevalence rate; for studies not reporting the denominator population, these values were back-calculated by using the reported numerators and prevalence rates. Two investigators used a standardized questionnaire to independently extract these and other relevant data such as study design, year of publication and criteria used for BD classification. Pooled prevalence rates across all studies and across pre-defined subgroups were computed with random-effects models based on negative binomial models.
Results The PubMed search yielded 230 citations. From these and additional articles found by handsearching, 33 potentially eligible articles were identified; 2 were excluded because of no numerator data. The 31 analyzed studies were published from 1974 to 2013 and included 7 studies from Turkey, 2 from other Asian countries, 8 from North Africa/Middle East, 5 from southern Europe, 7 from northern Europe and 2 from North America/Caribbean Islands. Table 1 shows the pooled prevalence rates for all studies and for subgroups of studies. Striking differences were seen for studies using sampling approaches as compared with those using census approaches. In contrast, the classification criteria used and the periods of publication appeared to have a negligible impact.
Conclusions Differences in BD prevalence rates likely reflect a combination of true geographic variation and methodological artefacts. In particular, the sampling or census study designs of surveys may strongly influence the outcome.
Disclosure of Interest None declared