1. Top of page
  2. Abstract


The HLA–DRB1 “shared epitope” (SE) genotypes are associated with rheumatoid arthritis (RA), but it remains controversial whether the association is with incidence, severity, or both, whether there are associations in seronegative patients, and whether different DRB1 alleles that contain the SE have similar effects on RA susceptibility and/or severity. The present study was undertaken to study these issues in a large cohort of patients with RA.


White patients with RA of <6 months' duration (n = 793) were enrolled in an inception cohort. HLA–DRB1 typing was performed, and patients were categorized into 21 DRB1 genotype groups. The disability index of the Health Assessment Questionnaire was the primary outcome measure.


DRB1 associations in seronegative RA patients closely resembled those in controls. Of seropositive patients, 21% had 2 copies of the epitope, 52% had 1 copy, and 27% had none. However, not all genotypes with 1 copy were associated with increased susceptibility; for example, frequencies of DRB1*0404/X and *01/X did not differ from those in controls. Absolute differences between seropositive RA patients and controls were greatest for DRB1*0401 homozygosity (3.8% versus 0.8%, respectively) and *0401/0404 heterozygosity (4.7% versus 1.0%). DRB1*0404 was increased in frequency in seropositive RA but, unlike *0401, an increased frequency was seen only with 2 epitope copies. The relatively rare DRB1*10 had an unexpected association with seropositive RA, being present in 1.7% of seropositive RA patients and 0.7% of controls, and also showed a trend toward association with greater disease severity. The presence of 2 epitope copies was associated with increased frequency of seropositivity and younger age at disease onset, not with disease severity. Treatment indication bias was substantial and may have accounted for some of these effects. HLA–DRB1*0401/0404 was found much more frequently in men and in patients with a lower age at disease onset, and there was a trend toward a higher frequency of *0404/0401 in women.


This large inception cohort study confirms previously identified major associations and provides additional insights. Only one dominant association was found: *0401, which differs from other SE alleles in a single Lys-for-Arg substitution. The association of the rare DRB1*10 allele has not previously been postulated. Sex associations were confirmed. Associations with seronegative RA were not seen. Not all genotypes containing an SE copy showed increased susceptibility to RA. The association of SE genotypes found in this study related to disease susceptibility rather than severity.

The association of the DRB1 shared epitope (SE) genotypes with rheumatoid arthritis (RA) has been extensively reported (1–16). However, it is less certain whether they predict severity or susceptibility, whether they are associated with seronegative RA, whether the associations differ between sexes, and whether there are additional unrecognized associations between genotype and RA. The SE is a sequence in the third hypervariable region (amino acids 67–71; Leu-x-x-Glu-Arg/Lys) of the DRB1 locus common to several different DRB1 alleles. A continuing controversy is whether different DRB1 alleles that contain the SE do, in fact, have similar effects on RA susceptibility and/or severity and whether the SE “dose” (1 versus 2 copies) affects susceptibility or severity. A number of investigators have found an association between the SE and disease severity (1–7), while others have not (8–16), even though the populations investigated were all or mostly white, in both the studies showing and those not showing an association. These issues are important to our understanding of genetic and environmental interactions in RA, since susceptibility might imply an abnormal response to a particular pathogen or pathogens, while severity might imply a more or less intensive immunologic response to a perturbation. If different genotypes with the SE epitope are associated with different effects, then it is clear that alleles, rather than any single sequence motif, determine the extent of risk.

Unfortunately, the interaction between treatment and disease severity has not been studied (1–16), yet differences in aggressiveness of treatment between genetic groups could easily confound severity/genetic associations, and the effects of this bias could vary greatly depending on the treatment philosophy implemented for the cohort. Currently used disease-modifying antirheumatic drugs (DMARDs) have major effects on the usual dependent variables, including disease activity, function, and radiographic score (17–22). Thus, the potential bias is that more severe disease is associated with one genetic group versus another. The treating physician, of course unaware of the patient's genetic group, sees the aggressive disease and treats vigorously. After the treatment response, the genetically more aggressive disease now appears to be less severe than the other.

Study of these issues has proven difficult for a number of reasons. A study group obtained by selecting consecutive patients in a clinic can be biased toward patients with more severe disease since sicker patients attend clinic more frequently. Most studies have been cross-sectional; only a few have been longitudinal. As described above, more severely ill patients may be treated with stronger agents earlier in the course of their disease. Distinctions between seropositive and seronegative patients also may be confounded, since some patients with early RA are seronegative although they later will become seropositive. It is likely, particularly in inception cohorts, that some seronegative patients do not have RA at all and that their diagnoses may change over time. The numbers of patients in most earlier studies have been small (often not more than 100), control groups needed for susceptibility testing may have had even fewer subjects, and controls may not have been ethnically matched. Finally, a large number of possible associations have been examined simultaneously.

We sought to clarify these issues by assembling a large inception cohort of white RA patients initially seen during the first 6 months of disease. This cohort was first studied cross-sectionally, as described in this report, and is being followed up longitudinally. We have extensive clinical and treatment data collected prospectively, with latex fixation (rheumatoid factor [RF]) and C-reactive protein (CRP) testing performed at a central laboratory. We were able to examine 21 separate genotypic groups both alone and aggregated, to use a large control group, to match ethnically by including only white patients and controls, to investigate for associations with rare alleles, and to examine confounding by extraneous variables such as treatment indication bias.

We sought to address the following questions: 1) Are all genotypes containing the SE associated with greater susceptibility, or do different DRB1 alleles containing the SE motif have different effects on disease risk? 2) Are there additional previously unrecognized associations between RA and DRB1 alleles? 3) Are there differences in disease severity (disability, pain, patient global assessment, RF status, erythrocyte sedimentation rate [ESR], CRP) associated with different alleles containing the SE or with combinations of these alleles? 4) Are there differences in use of DMARDs and prednisone associated with specific genotype groups and if so, could this bias account for reduction in disease severity in certain genotype groups (treatment indication bias)? 5) Are there differences in associations between seropositive and seronegative patients and/or between men and women? 6) Are there increased risks with 2 epitope copies versus 1, and if so, are these similar for all alleles or unique to certain homozygous or heterozygous combinations?


  1. Top of page
  2. Abstract


The Arthritis, Rheumatism, and Aging Medical Information System, National Inception Cohort of Rheumatoid Arthritis Patients was assembled in 1996–1997 by recruiting 352 rheumatologists from the American College of Rheumatology (ACR) membership to enroll consecutive patients with a clinical diagnosis of RA of <6 months' duration. Initially, 1,085 patients were enrolled by 161 rheumatologists and constituted a national sample. We excluded those who refused to participate, who could not be contacted, who had a change in diagnosis, or who had symptoms for >1 year. Nine hundred fifty-two patients provided informed consent and were enrolled, completed the Health Assessment Questionnaire (HAQ) (23), had clinical data provided by their physicians, and had cells and serum obtained and frozen at −70°C. Cells were subsequently typed as described below, and CRP and latex fixation tests performed on the serum at a central laboratory. This initial study was cross-sectional, using data from the first HAQ and initial clinical observations within the first 6 months of study. Results of longitudinal studies will be reported in the future; the cohort is being followed up with 6-month HAQ data, yearly clinical data, and 5-year radiographic data.

For the study reported herein, patients with a change in diagnosis in the first 3 years of observation have been excluded, and the patient group has been restricted to whites to facilitate comparisons with control populations and other studies. The final cohort consisted of 793 patients.

Clinical and laboratory variables.

The primary dependent variable for disease severity was the disability index of the HAQ, with the HAQ pain scale and patient global assessment as secondary measures. The disability and pain measures are scored from 0 to 3 and the global measure from 0 to 100; higher values represent greater severity. Laboratory indexes of disease activity were the CRP (primary) and the ESR. RF (by latex fixation) served as a stratifying laboratory covariate. RF and CRP tests were performed at a central laboratory in Wichita; in addition, RF and ESR values were submitted by referring physicians using results obtained from their own laboratories. We defined seropositivity as a positive result on one or both of these latex fixation tests, in order to minimize false-negative results.

Additional prespecified covariates included age, sex, DMARD use, prednisone use, and use of specific DMARDs including methotrexate (MTX), hydroxychloroquine (HCQ), sulfasalazine, intramuscular gold, and D-penicillamine. The most frequently used DMARDs were MTX and HCQ and their combination. At the time of study initiation, leflunomide, etanercept, infliximab, and anakinra had not been approved.

HLA typing.

Polymerase chain reaction (PCR)–based HLA class II typing was performed. DNA was extracted from frozen blood using QIAamp kits, according to the instructions of the manufacturer (Qiagen, Valencia, CA). All samples were PCR amplified for the DRB, with loci detection and interpretation performed using immobilized probe arrays (Dynal Biotech, Lake Success, NY). Group-specific amplifications and probe hybridizations for subtyping of DR4 alleles at the DRB1 locus were performed as previously reported (24). DRB1 alleles observed in this cohort that contain copies of the SE (25) were DRB1*01, DRB1*04, and DRB1*10. However, only some DRB1*04 alleles contain the SE (DRB1*0401, *0404, *0405, *0408). Furthermore, the DRB1*0401 allele contains the amino acid 67–71 sequence of Leu-Leu-Glu-Glu-Lys, whereas the DRB1*0404, *0405, and *0408 alleles contain 71 Arg. Therefore, the DR4 alleles were divided into the following groups: DRB1*0401, DRB1*0404 (including *0405 and *0408), and DRB1*04X (not including the SE). All other DRB1 alleles (excluding DRB1*01 and DRB1*10) were grouped into the DRB1*X category. Frequencies of genotype pair groups were determined by direct counting in the RA inception cohort. Control group genotype pair frequencies were calculated from control DRB1 allelic frequencies, assuming Hardy-Weinberg equilibrium. Our high-resolution HLA analysis was limited to the DR1 and DR4 serogroups. Since we have not conducted high-resolution analysis of the DR11 and DR13 serogroups at this time, we have not been able to examine the DERAA motif in this cohort.


The population data for 2 white control groups (from Paris and New Mexico) were generated from prior work at Roche Molecular Systems, using HLA genotyping techniques identical to those used in this study. The Centre d'Etude du Polymorphisme Humaine (CEPH) control group is restricted to white families, for whom HLA class I and II data have been previously reported (24); allele and gene frequencies are from parents. The New Mexico population was regional and was recruited from subjects in a study of HLA effects on the natural history of cervical human papillomavirus infection. This control group was exclusively female, recruited from obstetrics/gynecology clinics in New Mexico. Subjects were not included if they currently had, or had a history of, cervical disease. RA was not an exclusion criterion. There were no substantial differences between the 2 control populations with regard to the frequencies of any DRB1 allele groups or subgroups, and they had no remarkable features compared with other white control groups; hence, they were combined in these studies to provide larger numbers and greater data stability for fine analyses.

Statistical analysis.

Analyses were performed using the first value measured during the first 6 months of followup for each variable. Odds ratios (ORs) were constructed to compare genotype frequencies between cases and controls. Point estimates and their standard errors were obtained by the method of Christensen (26). An average, weighted by quantity of independent haplotypes, was calculated for each genotype across the CEPH and New Mexico control groups.

Nonlinear least squares was used to regress HAQ disability on 4 predictors in a generalized logistic model (27). A generalized logistic was selected because it accounts for the bounded nature of HAQ disability (0–3 scale). Predictors were number of SE copies, sex, baseline age, and number of DMARDs taken at baseline. Cost was regressed on quantity of epitopes, using ordinary least squares. A confidence interval (CI) on the coefficient of the slope was estimated with the bootstrap method (28), via resampling of observations.

Permutation testing (29) permitted examination of the relationship between mean HAQ disability and percentage of patients using DMARDs. Three hundred seven mixture distributions (the number of patients taking DMARDs) were calculated from the original data. Each mixture was created by pooling 2 random samples, 1 sample from the group of patients who did not take DMARDs (off) and the other from the group of patients who did (on), across a series of mixing ratios in sample sizes of 1:307, 2:306, … 306:2, 307:1. For each mixture the overall mean was calculated. This process was then repeated 500 times on randomly permuted data. For each permutation, assignments to DMARD usage (on or off) were randomly shuffled among all patients. Together the 500 permutations produced a pertinent reference set for the null hypothesis that observed HAQ disability is unrelated to DMARD usage in this study. The Pearson product-moment correlation between mean HAQ disability and percentage of patients taking DMARDs was calculated for the original data and each of the 500 random permutations. The 2.5% and 97.5% percentiles of the permutation distribution were calculated to estimate if the probability of obtaining a correlation as extreme or more extreme than the observed correlation for the original data was <5%.


  1. Top of page
  2. Abstract

Patient characteristics.

The 793 patients represent a geographically diverse, ethnically homogeneous cohort of RA patients who are likely to be broadly representative of patients with early RA seen by rheumatologists. Elimination of patients whose diagnosis had changed within 3 years increased confidence in the diagnoses. Because the short disease duration could affect the number of ACR RA criteria (30) met, we used instead the clinical diagnosis, sustained over time. The mean ± SEM age of the patients was 54 ± 0.5 years, and 75% were female. Sixty-seven patients (8%) were seropositive based on testing at the central laboratory, and 68 (9%) were seropositive based on testing at either the central or the home laboratory. The mean ± SEM HAQ disability index was 1.01 ± 0.02 on a scale of 0–3, the HAQ pain score was 1.03 ± 0.04 on the same scale, and the HAQ patient global score was 34 ± 1.1 on a scale of 0–100. The mean CRP value was 1.8 mg/dl, and the mean ± SEM ESR was 41 ± 1.0 mm/hour. Thirty-eight percent of the patients had already received 1 or more DMARDs in the first 6 months of disease (early aggressive treatment), and 26% were taking prednisone, at an average dosage of 4 mg/day.


Table 1 shows genotype frequencies in white RA patients as a total group, seropositive RA patients, seronegative RA patients, and the combined control groups, categorized by number of copies of the SE. Of all RA patients, 66% had at least 1 epitope copy, compared with 45% of the controls. Of seropositive RA patients, 73% had at least 1 copy. The distribution of genotypes in the RA group and in the seropositive RA subgroup was generally consistent with previously reported frequencies in white populations (1–16). Of interest, the genotype frequencies in seronegative RA patients did not differ appreciably from those in controls.

Table 1. Genotype frequencies in white rheumatoid arthritis (RA) patients and controls by number of copies of the shared epitope*
No. of copiesAll RA (n = 793)Seropositive RA (n = 528)Seronegative RA (n = 249)Controls (n = 791)
  • *

    Values are percents.

  • For 16 patients, information on rheumatoid factor status was not available.

1 or 266734945

Patients were stratified into 21 genotype groups to examine the role of individual DRB1 alleles carrying the SE as well as potential associations with homozygosity and heterozygosity. Table 2 presents genotype frequencies for the largest 13 of the 21 groups, stratified by the presence of 0, 1, or 2 copies of the shared epitope, in the entire RA cohort, seropositive RA patients, seronegative RA patients, and the combined control groups. “X” denotes a DRB1 allele other than those that contain an SE. The 2 control groups closely resembled each other, with the CEPH group and the New Mexico group having, respectively, 54% and 56% with no epitope, 39% and 38% with 1 copy, and 7% and 7% with 2 copies. In the tables these groups have been combined via sample size–weighted averages to increase the stability of the numbers and the clarity of presentation.

Table 2. Specific frequencies of genotypes occurring in at least 7 subjects among white rheumatoid arthritis (RA) patients and controls
DRB1Total nFrequency, %Odds ratio (95% CI), seropositive RA vs. controls*
All RASeropositive RASeronegative RAControls
  • *

    95% CI = 95% confidence interval.

No epitope copies      
 04X/X364. (1.1–2.3)
 XX23529.622.445.452.20.3 (0.0–0.5)
1 epitope copy      
 0401/X15719.822.713.312.62.1 (1.8–2.3)
 0404/X719. (0.7–1.5)
 01/X12615.915.217.314.31.1 (0.8–1.4)
 10/X121. (0.1–1.7)
2 epitope copies      
 0401/0401232. (4.2–6.1)
 0401/0404283. (4.0–5.6)
 0401/01273. (2.0–3.3)
 0404/040470. (2.2–5.0)
 0404/01182. (1.8–3.5)
 01/01182. (1.4–3.2)
 10s121. (1.3–3.4)

Several major findings were evident. First, the frequencies of genotypes containing 1 copy of the SE were not increased in any given group, with the notable exception of those containing *0401 (DRB1*0401/X). The frequency of DRB1*0401/X genotypes in seropositive RA patients was approximately double that in seronegative patients and controls; the number of patients was quite large, and the confidence limits of the OR indicated statistical significance. This observation suggests that *0401, but not other single-copy genotypes, acts as a “dominant” gene in terms of susceptibility or penetrance, and indicates that a single copy of a DRB1 allele, other than DRB1*0401, containing the epitope does not necessarily increase susceptibility to RA. It should be noted that the SE sequence in DRB1*0401 contains Lys at amino acid position 71, while all other SE alleles contain Arg at this position.

Second, for the genotypes with 2 SE alleles, in seropositive RA patients frequencies were generally increased, with ORs of 2–5, and the CIs always excluded 1. The DRB1*0401-containing genotypes were present in ∼13% of seropositive patients and ∼4% of controls; the *0404-containing genotypes were present in ∼9% of seropositive patients and ∼3% of controls. The highest OR in seropositive RA patients versus controls was for DRB1*0401/0401 (5.1), followed by *0401/0404 (4.8).

Third, the pooled DRB1*10-containing genotypes (*10/10, *01/10, *0404/10) also had a large relative frequency in seropositive RA patients versus controls (OR 2.3); this reached statistical significance, but the numbers were quite small. Numbers were also small, but the OR was also statistically significant, in the DRB1*0404/0404 group. These data suggest that DRB1*0404 and *10 genotypes act as “recessive” genes for susceptibility or penetrance and require 2 copies of the SE to increase susceptibility.

Fourth, when DRB1*0401 was present with either another copy of *0401 or with *0404, ORs were twice those with *0401 alone and also twice those with any other combination of 2 SE copies.

Table 3 contrasts selected genotype frequencies in seropositive patients by sex and shows ORs and CIs versus controls. For the great majority of genotypes, such as DRB1*0401/X, there were no differences between men and women. With DRB1*10/X there was an increase in men over women, but the numbers were small and the differences were not significant. However, there were statistically significant differences in 2 instances: HLA–DRB1*0401/0404 was more frequent in men, and DRB1*0404/01 was more frequent in women. This association of DRB1*0401/0404 with male sex (and with younger age at disease onset) has been previously reported by Weyand et al (1) and MacGregor et al (5).

Table 3. Selected genotype frequencies among seropositive rheumatoid arthritis patients, by sex*
DRB1 (n)Quantity of epitopesFrequency in male patients (n = 131), %Odds ratio (95% CI) vs. controlsFrequency in female patients (n = 397), %Odds ratio (95% CI) vs. controls
  • *

    95% CI = 95% confidence interval.

0401/X (120)120.61.8 (1.3–2.3)23.42.1 (1.8–2.4)
10/X (10)13.81.8 (0.8–2.8)1.30.6 (0.0–1.6)
0401/0404 (25)29.29.6 (8.7–10.5)3.33.2 (2.4–4.1)
0404/01 (15)20.80.7 (0.0–2.8)3.53.3 (2.4–4.1)

Clinical features.

Table 4 compares clinical characteristics of patients with 0, 1, or 2 copies of the SE, in all RA patients and in those with seropositive RA. The age at onset of disease was inversely correlated with the number of epitope copies, particularly in the seropositive patients, among whom the average onset ages were 56 years, 53 years, and 51 years, respectively, in those with 0, 1, and 2 copies. The percent males increased somewhat with the number of copies (21%, 26%, and 27% among seropositive patients with 0, 1, and 2 copies, respectively). The frequency of seropositivity increased strikingly with the number of epitope copies, although the mean RF titer among seropositive patients did not rise. These observations suggest increasing susceptibility (or penetrance) with epitope number and do not have implications with regard to severity.

Table 4. Characteristics of rheumatoid arthritis (RA) patients (all patients and seropositive patients) with 0, 1, and 2 shared epitope copies*
Characteristic0 copies1 copy2 copies
All RA (n = 273)Seropositive RA (n = 143)All RA (n = 387)Seropositive RA (n = 274)All RA (n = 133)Seropositive RA (n = 111)
  • *

    Values in parentheses are the standard error of the percentage. RF = rheumatoid factor (determined by latex fixation); ESR = erythrocyte sedimentation rate; CRP = C-reactive protein; DMARDs = disease-modifying antirheumatic drugs; MTX = methotrexate.

  • By Health Assessment Questionnaire (0–3 scale for disability and pain; 0–100 scale for global assessment).

 Age, mean ± SEM years55.8 ± 0.955.5 ± 1.354.2 ± 0.853.4 ± 0.952.0 ± 1.350.8 ± 1.4
 % female78.0 (2.5)79.0 (3.4)74.2 (2.2)74.1 (2.7)70.7 (4.0)73.0 (4.2)
 Disability, mean ± SEM1.03 ± 0.10.96 ± 0.11.03 ± 0.11.02 ± 0.10.88 ± 0.10.84 ± 0.1
 Pain, mean ± SEM1.04 ± 0.10.98 ± 0.11.07 ± 0.11.08 ± 0.10.90 ± 0.10.84 ± 0.1
 Global, mean ± SEM34.0 ± 2.032.6 ± 2.634.9 ± 1.634.9 ± 1.933.4 ± 2.931.9 ± 3.2
 RF titer, mean ± SEM148.8 ± 17.0262.4 ± 28.8190.4 ± 13.6253.7 ± 17.1222.3 ± 20.8255.4 ± 22.7
 % seropositive53.2 (3.1)100 (0.0)72.5 (2.3)100 (0.0)85.4 (3.1)100 (0.0)
 ESR, mean ± SEM mm/hour38.4 ± 1.641.7 ± 2.042.4 ± 1.542.2 ± 1.740.6 ± 2.540.8 ± 2.8
 CRP, mean ± SEM mg/dl1.77 ± 0.21.92 ± 0.21.85 ± 0.12.01 ± 0.21.93 ± 0.21.97 ± 0.3
 No. of DMARDs taken, mean ± SEM0.49 ± 0.10.58 ± 0.10.45 ± 0.10.47 ± 0.10.57 ± 0.10.58 ± 0.1
 % taking any DMARDs39.2 (3.0)44.8 (4.2)37.5 (2.5)38.3 (2.9)42.1 (4.3)43.2 (4.7)
 % taking MTX20.1 (2.4)25.2 (3.6)20.2 (2.0)21.2 (2.5)28.6 (3.9)27.0 (4.2)
 % taking prednisone25.3 (2.6)25.9 (3.7)25.3 (2.2)25.5 (2.6)27.1 (3.9)25.2 (4.1)

Clinical severity.

The patient-derived outcome variables of HAQ disability, pain, and global assessment showed unexpected trends, with clinical disease severity tending to actually decrease with increasing number of epitopes. Values for each of these variables were lowest in the groups with 2 epitope copies. Laboratory disease activity variables of mean RF titer, ESR, and CRP showed only inconsistent associations with the number of epitope copies.


There were few dramatic treatment differences when groups were stratified by number of epitope copies. MTX, possibly the strongest DMARD in general use at the time of patient enrollment (31), however, was used by ∼29% of patients with 2 copies compared with ∼20% in the other 2 groups. Other treatments, such as HCQ and prednisone, failed to show consistent correlations with the number of epitope copies. Medical costs might be considered an additional surrogate measure for disease severity, but we found no associations between direct medical costs and specific genotype or number of epitope copies (data not shown).

Table 5 shows the same clinical variables across specific genotypes of interest. Results are consistent with those seen in Table 4, but the numbers of patients are smaller. Of note, the group of patients with DRB1*0401/0404 was younger (mean 45 years) and had a higher percentage of males (nearly one-half) (similar to findings in studies by Weyand et al (1) and MacGregor et al (5), while the group with DRB1*0404/0404 was a little older (mean 49 years) and had a higher proportion of females (86%). If a susceptibility or increased penetrance genotype may be inferred by association with lower age at onset and greater proportion of men, then *0401/0404 appeared to have the strongest association with susceptibility. The most aggressive therapy was given to the relatively small *10s group, made up of *10/10, *01/10, and *0404/10, the 3 genotypes containing *10 and a second copy of the epitope.

Table 5. Clinical characteristics of seropositive RA patients with specific genotypes*
Characteristic1 epitope copy2 epitope copies
0401/X (n = 120)01/X (n = 80)0401/0401 (n = 20)0401/0404 (n = 25)0404/0404 (n = 7)0401/01 (n = 23)10s (n = 9)
  • *

    See Table 4 for explanations and definitions.

Age, mean years52575345495051
% female78737052868367
% seropositive (parent cohort)786591891008582
Disability score, mean1.
Pain score, mean1.061.331.200.650.300.891.25
CRP, mean mg/dl1.742.213.641.940.461.601.11
No. of DMARDs taken, mean0.510.400.400.640.290.570.89
% taking any DMARDs42343544294467

We addressed the question of biases that might result from treatment differences across groups. If early aggressive treatment is associated with improvement in outcome variables and if treatment is given earlier or more aggressively to certain genotype groups because of their greater disease severity, then a treatment indication bias could mask an actually greater biologic disease severity in that group. Table 6 compares patients who were and those who were not taking DMARDs, stratified by the number of epitope copies and by RF status. Disability was clearly reduced in those receiving DMARD treatment, and the magnitude of reduction (difference of 0.17, 0.24, and 0.36 on a 0–3 scale for those with 0, 1, or 2 epitope copies, respectively) was much larger than differences in disability among untreated subjects. Paradoxically, the percentage of patients taking any DMARD was 45%, 37%, and 12%, respectively, among those with mild, moderate, and severe disability (HAQ scores of 0–<1, 1–<2, and 2–3, respectively). In contrast, patients who had not yet received DMARD therapy had similar degrees of disability when stratified by number of copies. There did not appear to be major differences in treatment by number of epitope copies, although such differences might have been present in subgroups.

Table 6. Differences among seropositive RA patients receiving and those not receiving DMARD therapy*
Characteristic0 epitope copies1 epitope copy2 epitope copies
No DMARD (n = 79 [55%])DMARD (n = 64 [45%])No DMARD (n = 169 [62%])DMARD (n = 105 [38%])No DMARD (n = 63 [57%])DMARD (n = 48 [43%])
  • *

    See Table 4 for explanations and definitions.

Age, mean years555753545150
% female778172786879
% seropositive (parent cohort)486172738487
Disability score, mean1.040.871.110.871.000.64
Pain score, mean0.781.
CRP, mean mg/dl1.682.221.932.142.111.78
No. of DMARDs taken, mean0.

Figure 1 shows the relationship between HAQ disability scores and the percentage of patients taking DMARDS, ascertained using the permutation test described in Patients and Methods. There was a strong and consistent inverse association, which was statistically significant (P < 0.05). This relationship was sufficiently strong that even small differences in treatment between groups could mask small differences in disease severity. Nonlinear generalized logistic regression analysis indicated that the average number of DMARDs was the best predictor of greater disability levels (P < 0.001), followed by female sex (P < 0.01) and age (P = 0.10). Number of epitope copies was not a significant predictor of greater disability (P = 0.5). A similar regression analysis was performed using the same covariates and the number of physician visits as a surrogate measure of medical costs with this analysis as well; the number of epitope copies was not included in the model.

thumbnail image

Figure 1. Relationship between disability as determined by the Health Assessment Questionnaire (HAQ) and use of disease-modifying antirheumatic drugs (DMARDs).

Download figure to PowerPoint


  1. Top of page
  2. Abstract

This large data set from an inception cohort of white RA patients contains extensive HLA–DRB1 genotyping, demographic, clinical, and treatment data. In examining data obtained within the first 6 months of study and the first year of disease, we have been able to confirm the general findings of DRB1 associations described previously. In addition, with the large number of patients (n = 793) included in the present study, we have been able to postulate some new associations, demonstrate differences among DRB1 alleles containing the shared epitope, and develop insights into a treatment bias that may have contributed to confusion in the earlier literature.

One issue that has remained controversial has been the association, or lack thereof, between the SE genotypes and seronegative RA. Clinically, it is obvious that any group of RA patients labeled “seronegative” is very likely to include some who will later become seropositive. Some patients become seropositive later in the disease course, and some are inconsistently seropositive. As a consequence, any genotype associations that are found only in seropositive patients are likely to be present also in “seronegative” patients but to appear with reduced frequency, creating false-negative results. We found only minimal examples of this “contamination” of seronegative groups by seropositive patients in our cohort, perhaps because we used a high-quality central laboratory for RF testing, also used rheumatologist-confirmed diagnoses, and had 2 serum samples analyzed for seropositivity for most patients. Our data strongly suggest that the SE associations in whites are limited to RA patients who are seropositive, which raises again the issue as to whether seropositive and seronegative RA are the same disease.

Genotypes containing DRB1*0401, unlike other SE-containing DRB1 alleles, increase susceptibility or penetrance even if only 1 copy of the SE is present. We note that DRB1*0401 encodes Lys in amino acid position .71, while all other DRB1 alleles containing the SE encode Arg. In contrast, DRB1*0404- and *10-containing genotypes appear to require 2 copies of the SE to increase susceptibility. DRB1*0401 in combination with either itself or *0404 further increases the association with susceptibility. All combinations of 2 copies of the epitope are significantly associated with susceptibility. Clearly there is much heterogeneity of risk among DRB1 alleles containing the SE, and clinical associations appear to be dependent as much on the specific genotype as on the presence and number of the SE.

The association of DRB1*0401 with heightened susceptibility is also supported by its association with younger age at disease onset and male sex, if one assumes that younger onset implies greater patient susceptibility as does male sex; such attributes might imply lessened resistance to a perturbation in individuals who would otherwise be more resistant to RA. These sex difference findings are robust, and the younger age and greater proportion of males among patients with DRB1*0401/0404 have been reported previously by others (1, 5). Again, the message is that of heterogeneity in susceptibility among DRB1 alleles containing the SE. Additional genetic factors, such as tumor necrosis factor microsatellite markers (32), HLA–A, HLA–DQ, or others, may be involved, or environmental factors such as smoking or infection (33, 34) could interact differently with different genotypes.

We found an increase in frequency of genotypes containing the *10 allele in combination with another SE DRB1 allele among seropositive RA patients, an observation that has not been described previously. The numbers remain rather small, however, and this association should be considered hypothetical at present. Of potential clinical significance, this subgroup is the only group examined in which there was an association with disease severity and increased DMARD treatment.

Our cohort is large, with well-established diagnoses and careful protocol-based outcome observations, and with extensive treatment data. The patient sample is likely to be broadly representative of rheumatologist-treated RA patients in North America. Yet, we did not find severity effects as assessed with disability, pain, patient global score, or laboratory variables. We did not study radiologic outcomes, although recent work has suggested that there is little association between epitope status and radiologic outcomes after up to 5 years (8, 13). We did not include data on extraarticular features, since we were examining patients during the first year of disease, in which nodules are rare and other extraarticular features are generally not seen.

Previous studies have not investigated the potential biases introduced by treatment status in relation to the question of whether genotype can predict severity of illness. It is clear that these biases can be of major importance. Consider 2 patients first seen by a rheumatologist after 5 months of illness and enrolled in an inception cohort. One has severe disease and his primary physician prescribed MTX only 3 months after disease onset. The other has less severe disease and is still DMARD-naive when enrolled in the cohort. Since MTX treatment is known to result in decreases in HAQ disability scores of ∼0.30–0.40 units within a few months (17–19, 31), the biologically more severely affected patient may well have lesser apparent disease severity, since any apparent differences in severity between genotypes are much smaller than the effects of the MTX treatment. Thus, a genotype associated with disease severity might lead to more vigorous treatment and an ensuing response, which could more than account for the differences in severity at the time of study.

Radiographs were not available for many of the patients at this point in the study, but a treatment bias could also invalidate severity data involving radiographic erosions; the rate of erosion development is substantially reduced by a number of contemporary treatments. Longitudinal study may help elucidate some of these issues but will likely not conclusively answer the questions since treatment bias will be more and more complex with greater disease duration and multiple treatments. Only if clinical data are available for all patients in a cohort prior to institution of DMARD therapy can the bias be avoided; we hope that by observing our patients who meet this criterion we will be able to provide more definitive answers. In general, however, we believe that any associations of disease severity with specific genotypes or with number of epitope copies must be small and not of clinical significance, at least in early disease.

Nevertheless, we believe there may be hidden associations of disease severity with certain genotypes which cannot yet be identified with certainty. We found, for example, that the patients with DRB1*10 carrying an additional allele with the SE were more likely to have been treated with MTX. Since treating physicians were blinded to genotype and our analysis in this instance was limited to seropositive patients (some clinicians might use seropositivity itself as an indication for more aggressive treatment), it seems possible that the treating physician may have made the more aggressive treatment choice on the basis of greater perceived severity.

It is likely that the magnitude of treatment indication bias is not the same across different published studies. It will be largest when the philosophy of treatment for the cohort is most aggressive. Future studies must consider the effects of treatment upon the apparent disease severity associated with specific genotype groups.


  1. Top of page
  2. Abstract