The descriptive epidemiology of RA is suggestive of a genetic effect. The occurrence of RA is relatively constant with a prevalence of between 0.5 and 1.0%, a frequency that has been reported from several European 12345678 and North-American populations 910. However, there are some interesting exceptions (Fig. 1).

Specifically, native American-Indian populations have the highest recorded occurrence of RA, with a prevalence of 5.3% noted for the Pima Indians 11 and of 6.8% for the Chippewa Indians 12. By contrast, there are a number of groups with a very low occurrence. Studies in rural African populations, both in South Africa 13 and in Nigeria 14, failed to find any RA cases in studies of 500 and 2000 adults, respectively. Studies in populations from Southeast Asia 15, including China and Japan 1617, have similarly shown very low occurrences (0.2–0.3%).

It is clearly difficult from a review of the descriptive epidemiological data to know whether environmental or genetic effects explain the differences between countries. One handle on this is to consider the occurrence in populations presumed to be of the same genetic origin but living in different environments. Such a situation arises by studying populations that have moved from one environment to another.

There are a few studies addressing this with respect to RA. A low occurrence of disease was found in one study of a Caribbean population of African origin living in Manchester, UK, suggesting that the protection to this group was indeed genetically determined 18. Similarly, the investigation of a Chinese population living in an urban environment in Hong Kong showed the same consistent low frequency 19. Recent data have shown, however, that Pakistanis living in England had a higher prevalence than those in Pakistan, but it is not as high as the prevalence in ethnic English populations 20. In general, the data on the geographical occurrence of RA would support the existence of genetic factors being important and explaining differences in disease risk.

The next stage in accruing evidence for genetic risk is to document an increased occurrence of disease in relatives of probands compared with the background population prevalence, so-called familial recurrence risk. Studies of hospital attendees are subject to bias as there may be a selection process whereby individuals would more probably be referred to hospital if they have an affected family member. Furthermore, several studies rely on family history as elicited by the proband, which again is subject to bias.

Few studies have been performed comparing familial recurrence risk in relatives of cases derived from population samples with those of controls. Indeed, such studies have only shown a modest increased risk 2122. For example, a study from the Norfolk Arthritis Register in England showed only a twofold increased risk 23. Such an observation does not negate the role of genetic factors, but underscores that their contribution to explain disease susceptibility may be modest. This is important as the familial recurrence risk is a key factor in determining the power of genetic linkage studies within affected family pairs. Indeed, in contrast to other autoimmune diseases such as insulin-dependent diabetes and multiple sclerosis, the familial recurrence risk in RA is certainly smaller, thereby making it harder for studies to identify new genetic factors.

A variant of studies of familial recurrence risk is the comparison of disease risk in the initially unaffected co-twin of monozygotic probands compared with dizygotic probands. The assumption is that the environmental sharing between these different twin pair types is the same and thus any increased disease concordance in the monozygotic twins confirms the genetic effect. It is important in such studies to ensure that it is only like-sexed dizygotic twin pairs that are compared with the monozygotic twin pairs. However, there may be a greater environmental concordance in monozygotic twins due, for example, to psychological and other factors.

Twin studies have consistently showed a fourfold increased concordance in monozygotic twins compared with dizygotic twins 24. This increased risk, however, is of little value in attempting to quantify the genetic contribution to disease risk. The concordance between twins is dependent on the prevalence of disease. As the population prevalence approaches 100%, the concordance will increase accordingly, independent of the true genetic effect. The appropriate way of quantifying genetic risk is to assess the heritability based on a series of assumptions of environmental sharing and genetic sharing between twin types. Such a study has recently been attempted using data from both Finnish and English twins 24. The results suggest that approximately 50–60% of the occurrence of disease in the twins is explained by shared genetic effects.

The role of HLA DRB1 alleles as a risk factor of RA has been known for 25 years. Associations between different HLA DRB1 alleles have been demonstrated in several populations across the world 25262728293031. Indeed, there have been few populations where associations have not been demonstrated.

Interestingly, there do appear to be differences in the strength of association between different alleles. For example, HLA DRB1*0404 is a much stronger susceptibility factor than HLA DRB1*0101 32 (Table 1). Although it has been suggested that the susceptibility alleles all share a single epitope 33, it is difficult to explain the variable risk under this model. Furthermore, the risk of disease is related not only to the presence of one single allele, but also to the full HLA DRB1 genotype 34. Individuals who carry the so-called 'compound heterozygote' genotype HLA DRB*0401/*0404 thus have a substantially greater risk than, for example, individuals who carry single HLA DRB*0101 alleles.

There is some suggestion that the relationship between human leukocyte antigen (HLA) and RA may be more related to the severity of disease, and that the development of arthritis per se is only weakly related. Support for this comes from studies from the Norfolk Arthritis Register population-based study of inflammatory joint disease in England 32. Data from this study show only a weak relationship between susceptibility to the disease and the HLA DRB1 genotype (Table 1). The association is fairly strong in those individuals who satisfy the criteria for RA. The data show there is an influence of genotype, with some genotypes having a stronger association as shown.

The HLA region on the short arm of chromosome 6 is a gene-rich area including several candidate genes that have an influence on the immune process. One of the most highly investigated is tumour necrosis factor (TNF). Studies have shown associations between TNF alleles and RA 3536, although one explanation may be linkage disequilibrium with HLA DRB1. Studies have also suggested, however, that the associations between HLA and TNF-c1 and TNF-b3 are independent of associations between HLA and the shared epitope 37. Other studies have shown an extended haplotype stretching from HLA through to TNF that has been implicated in disease 38.

Data from twin studies in the HLA association and sharing studies have been used to estimate that only 50% of the genetic contribution to RA can be explained by HLA 24. This has sparked a search for non-MHC genes.

The largest effort has been expanded in some whole genome screens on affected sibling pair families. Four such screens have now been undertaken in Europe 39, the United States 33, Japan 40 and the United Kingdom 41. A number of markers emerge from these studies suggestive of a linkage with RA, although the linkage with HLA is by far the strongest. One problem is that such studies often have only a weak power to detect defects. By contrast, because such studies may be simultaneously testing the possibility that any one of 200 regions may be linked with disease, the likelihood of a false-positive result is also very high. It is for this reason that it is not surprising studies often fail to replicate results both between themselves and on further samples within the population. It is therefore necessary to undertake further validation studies and more in-depth investigations, using more closely spaced markers.

An alternative approach is to use a candidate gene screen where there is no prior reason for looking at a particular region. Such an approach has been productive, and evidence has shown that corticotrophin releasing hormone 42, CYP19 (oestrogen synthase) 43, IFN-γ 444546 and other cytokines 474849 are linked to RA. Other approaches have addressed the possibility that genetic regions linked to other autoimmune diseases, such as insulin-dependent diabetes 50, may also be linked to RA. Indeed, linkage to a locus on chromosome X was shown in one study 51. A further strategy has been to use results and genome screens on animal models of arthritis to see whether syntenic regions are also linked to RA in humans. Such studies have suggested linkage to 17q22 52.

Whether any of these positive findings discussed will result in the identification of a true disease susceptibility mutation remains to be seen. However, one clear problem is that RA itself is probably heterogeneous and studies that fail to take notice of this heterogeneity may make it possible to find a positive result.

The increased risk of RA in females has lead to considerable effort in examining the role of hormonal and pregnancy factors in disease occurrence. In general, male sex hormones, particularly testosterone, are lower in men who have RA 25. By contrast, levels of female sex hormones are not different between RA cases and controls 53.

Interestingly, exogenous hormonal influences are implicated in disease risk. The most widely studied of these is exposure to the oral contraceptive pill, based on an observation made over 20 years ago 54 (Fig. 2). There have been several studies 55 confirming that women who take the oral contraceptive pill are at reduced risk of developing RA 25. There is no clear explanation for this and the association exists despite the formulation of the oral contraceptive pill varying enormously both between populations and over time. A follow-up of the original study was undertaken that suggested the oral contraceptive pill was protective. This showed that the initial protection was lost on follow-up 56. One conclusion might therefore be that oral contraceptive use may postpone, rather than totally protect against, the development of RA.

Pregnancy itself has been investigated as a risk factor in RA development. Studies on the influence of pregnancy on RA have produced conflicting results. A number of studies 25 have suggested that women who are nulliparous are at increased risk of developing the disease, although there is no increased risk in women who are single 57. It would thus appear that subfertility highlights a group at higher risk.

Recent studies have suggested that pregnancy might also be important with the interesting observation that the postpartum period, particularly after the first pregnancy, represents a strong risk period of disease development 5859 (Fig. 3). Subsequent investigations showed that much of this increased risk could be explained by exposure to breastfeeding and it is women who breastfeed after their first pregnancy who are at the greatest risk 60. The suggestion then arose linking breastfeeding with sub-fertility in so far as RA may be related to either increased prolactin or abnormal response to prolactin, this latter hormone being proinflammatory 61.

There have been a number of studies looking at other comorbidities that have an increased frequency in both subjects with RA and in their families. The most widely investigated has been the occurrence of other autoimmune diseases, particularly type 1 or insulin-dependent diabetes and autoimmune thyroid disease 62. Other diseases, for example schizophrenia, have been shown to be negatively associated with RA development 636465. The significance of these findings is unclear.

There have been relatively few studies on anthropometric factors associated with RA, although one recent case–control study suggested that people who were obese were at higher risk 66. The reason for this is unclear, and it is not certain whether this may represent a confounding factor of another exposure or whether people who are obese have, for example, increased production of oestrogens, which might pose a risk. A more recent case–control study found, however, after adjusting for age, smoking and marital status, that a link with obesity was nonsignificant 67.