Prevalence of rheumatoid arthritis in persons 60 years of age and older in the United States: Effect of different methods of case classification

Authors


Abstract

Objective

To determine prevalence estimates for rheumatoid arthritis (RA) in noninstitutionalized older adults in the US. Prevalence estimates were compared using 3 different classification methods based on current classification criteria for RA.

Methods

Data from the Third National Health and Nutrition Examination Survey (NHANES-III) were used to generate prevalence estimates by 3 classification methods in persons 60 years of age and older (n = 5,302). Method 1 applied the “n of k” rule, such that subjects who met 3 of 6 of the American College of Rheumatology (ACR) 1987 criteria were classified as having RA (data from hand radiographs were not available). In method 2, the ACR classification tree algorithm was applied. For method 3, medication data were used to augment case identification via method 2. Population prevalence estimates and 95% confidence intervals (95% CIs) were determined using the 3 methods on data stratified by sex, race/ethnicity, age, and education.

Results

Overall prevalence estimates using the 3 classification methods were 2.03% (95% CI 1.30–2.76), 2.15% (95% CI 1.43–2.87), and 2.34% (95% CI 1.66–3.02), respectively. The prevalence of RA was generally greater in the following groups: women, Mexican Americans, respondents with less education, and respondents who were 70 years of age and older.

Conclusion

The prevalence of RA in persons 60 years of age and older is ∼2%, representing the proportion of the US elderly population who will most likely require medical intervention because of disease activity. Different classification methods yielded similar prevalence estimates, although detection of RA was enhanced by incorporation of data on use of prescription medications, an important consideration in large population surveys.

Rheumatoid arthritis (RA), a multisystem disorder characterized by chronic destructive synovitis, can have substantial adverse health consequences for those who are affected. Although the clinical course and outcomes vary widely among individuals (1), it is well established that persons with persistent RA experience progressive disability (2–7) and early mortality (8–11). Periodic documentation of the national prevalence of RA is important in order to characterize the disease burden and evolution in the population (12). Furthermore, national prevalence estimates are critical to the process of establishing priorities for public health action and resource allocation (12). In the US, the last national prevalence estimates for RA (13) were derived from the First National Health and Nutrition Examination Survey (NHANES-I), which was conducted between 1971 and 1975. In NHANES-I, the case definition of RA was based on the clinical diagnosis of the examining physician and not on established, recognized classification criteria. Since that time, classification criteria for RA have been revised (14), which warrants updated prevalence estimates.

The accuracy of prevalence estimates is determined in part by the method of case classification. However, the classification of RA is complicated by the lack of a definitive “gold standard” (15). For the purposes of epidemiologic studies, the methods of classifying RA have evolved over the last 45 years (14, 16, 17), but the criteria have remained descriptive. Current understanding of RA lacks an etiologic basis for explaining pathologic disease processes and for differentiating between similar rheumatic conditions (15). Thus, criteria used in epidemiologic studies rely on the presence of clinical signs, laboratory and radiographic findings, and self reports of symptoms to classify persons with RA (14). In recent studies, application of these criteria has resulted in a sensitivity of 91–94% and a specificity of 89% (14). However, early-stage RA and mild or self-limited RA may not be ascertained using these criteria. Furthermore, other diseases with similar clinical features may be incorrectly classified when these criteria are applied (14, 18–20). Understanding the effect of different disease classification methods on prevalence estimates is paramount to the accuracy and interpretation of enumerations of persons with RA.

In the present study, data from NHANES-III (21) were used to compare prevalence estimates generated by 3 different RA classification methods in noninstitutionalized individuals ages 60 years and older. Our findings are presented herein.

SUBJECTS AND METHODS

Data source.

NHANES-III was a cross-sectional survey conducted between October 1988 and October 1994 by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (21). The NHANES-III used a multistage, stratified probability cluster design to select a sample representative of the civilian noninstitutionalized population who were ages 2 months and older and residing in the 50 states of the US. Children younger than 5 years, adults 60 years and older, black persons, and Mexican American persons were oversampled to improve the precision of estimates for these subgroups. Data were collected through an interview in the participants' homes and a standardized physical examination conducted in a mobile examination center (MEC). The NHANES-III plan of operation and sample design have been described in detail elsewhere (22).

Subjects.

Of those eligible to participate in the NHANES-III, 33,994 infants, children, adolescents, and adults received the household interview (86%). Of those interviewed, 91% were examined at the MEC and 1.5% had a limited examination at home. Examination variables necessary for RA classification were available for persons ages 60 years and older. Of the 6,596 persons interviewed in this age range, 5,302 completed the MEC examination (80%).

Variables of interest.

Demographic data, self report of RA, self report of morning stiffness, and information about prescription drug use were obtained from the household interview. Upper extremity physical examinations performed at the MEC documented swelling of the proximal interphalangeal (PIP), metacarpophalangeal (MCP), wrist, knee, and first metatarsophalangeal (MTP) joints, and the presence of rheumatoid nodules. Blood samples were drawn at the MEC to determine the presence of rheumatoid factor (RF). Hand radiographs were also obtained at the MEC.

Demographic data.

Demographic data included sex, age, race/ethnicity, education, marital status, poverty index, and health insurance coverage. Self-reported race and ethnicity were recoded to four race/ethnicity groups for analysis: non-Hispanic white, non-Hispanic black, Mexican American, and other. Respondent age (<70 years; ≥70 years) and education (less than high school completion; high school completion and above) were each divided into 2 categories for analysis. Four categories characterized marital status (married, widowed, divorced/separated, and never married). Responses of “married, spouse in household,” “married, spouse not in household,” and “living as married,” were grouped into 1 category (married). Persons reporting a status of divorced or separated were grouped into 1 category. The poverty index relates family income to the national poverty threshold, as determined annually by the Census Bureau. For descriptive purposes, the poverty index (≤1.0; >1.0) and report of health insurance coverage within the previous month (yes; no) were each divided into 2 categories.

Criteria for RA.

Questionnaire, examination, and laboratory data were used to identify cases of RA based on the American College of Rheumatology (ACR; formerly, the American Rheumatism Association) 1987 revised criteria for the classification of RA (14). Fulfillment of the ACR criteria was determined as follows. For criterion 1, morning stiffness lasting for more than an hour was determined by responses to the questions “Have you ever had stiffness in your hands when first getting out of bed in the morning on most days for at least 6 weeks?” and “How long after getting up and moving around does the morning stiffness last?”

For criterion 2, the presence of arthritis in 3 or more joint areas was determined from examination variables documenting the presence of soft tissue swelling in 3 or more of the following 10 joint areas: right or left PIP, MCP, wrist, knee, and first MTP joints. Data indicating the presence of joint swelling were available for 4 PIP joints and 5 MCP joints bilaterally. Swelling was classified as present in each joint area if it was found in 1 or more of the individual joints in these areas. Examination data on elbow and ankle swelling, which are components of the ACR 1987 criteria for determining joint swelling, were not collected in the NHANES-III.

For criterion 3, the presence of arthritis in the joints of the hand was determined from examination variables documenting the presence of soft tissue swelling in at least 1 right or left PIP, MCP, or wrist joint area.

For criterion 4, the presence of symmetric arthritis was determined from examination variables documenting the presence of soft tissue swelling of the same joint areas on both sides of the body simultaneously (right and left PIP, MCP, wrist, knee, and first MTP joints).

For criterion 5, the presence of rheumatoid nodules was determined from the examination variable documenting the presence of subcutaneous nodules observed on the shaft of the right forearm, the left forearm, or both forearms.

For criterion 6, the presence of RF was determined from blood samples using the Singer-Plotz latex agglutination test. Blood specimens were screened using latex-enhanced nephelometry prior to obtaining a titer (23). RF assays were unavailable for 337 persons. For the purposes of this study, all non-zero, non-missing values were classified as being RF positive. The value of the lowest RF titer was 1:20. All zero values were classified as being RF negative. Management of missing RF titer values is discussed under case definition methods.

For criterion 7, anteroposterior radiographs of the hands and wrists were collected in NHANES-III, but readings for the presence of RA were not available for analysis at the time of the present study.

Medication data.

Use of disease-modifying antirheumatic drugs (DMARDs) and glucocorticoids provided additional data with which to identify RA case status. During the household interview, respondents were asked: “Have you taken or used any medicines for which a doctor's or dentist's prescription is needed, in the past month?” For each medication reported, the interviewer asked to see the medication container to record the product name. If the container was unavailable, the interviewer queried the subject for this information. Each prescription drug reported in the survey was subsequently identified in the Physicians' GenRx, and assigned a standard generic name and 4-digit generic code for that product. For this study, classification of medications (Table 1) was developed using the Women's Health and Aging Study classification system as a model (24) to identify persons who reported the use of any DMARDs (A-list) or glucocorticoids (B-list) within the previous month.

Table 1. Use of DMARDs and glucocorticoids as reported by NHANES-III respondents*
A-list medications (DMARDs)B-list medications (glucocorticoids)
  • *

    DMARDs = disease-modifying antirheumatic drugs; NHANES-III = Third National Health and Nutrition Examination Survey.

Medications used by ≥1 NHANES-III respondent 
 AuranofinMethylprednisolone
 Gold sodium thiomalatePrednisone
 Methotrexate sodiumPrednisolone
 AzathioprineBetamethasone
 PenicillamineTriamcinolone
 Sulfasalazine 
 Chloroquine phosphate 
 Hydroxychloroquine sulfate 
Medications not used by any NHANES-III respondent 
 Gold compoundsPrednisone stelagate
 Gold sodium thiosulfate 
 Aurothioglycanide 
 Aurothiopropanol sulfonate 
 Aurothioglucose 

Self report of RA.

Self report of RA was identified by responses to the questions “Has a doctor ever told you that you had arthritis?” and “Which type of arthritis was it? Was it rheumatoid arthritis or osteoarthritis?”

RA case definition.

Constructed questionnaire, examination, and laboratory variables reflecting ACR criteria were classified as being present or absent for the purpose of identifying cases of RA. Three classification methods were used. For method 1, the “n of k” rule (14) was applied, such that subjects who met 3 of the 6 available ACR criteria were classified as having RA. If RF titers were missing, subjects who met 3 of the 5 available ACR criteria were classified as having RA.

For method 2, the classification tree algorithm described by Arnett and colleagues (14) was applied. This method allows the use of surrogate classification variables when a primary classification variable is unavailable. In the present study, MCP swelling was substituted for radiographic findings and wrist swelling was substituted for missing RF titer values.

For method 3, the ACR classification tree algorithm used in method 2 was modified to incorporate the use of DMARDs (A-list medications) to classify RA. Respondents who reported having RA and using A-list drugs were classified as having RA. Respondents who did not report having RA but reported using A-list drugs had to meet at least 1 of the ACR criteria to be classified as having RA. Respondents who did not report the use of A-list drugs were classified according to the ACR tree algorithm. B-list drugs were not used for case classification since their prescription is not specific to the treatment of RA. However, the effect that incorporation of B-list drug use would have on case classification is discussed below.

Statistical analysis.

Concurrence between self report of RA and the findings of the 3 classification methods was determined using the unweighted sample by identifying the proportion of persons classified as having RA by methods 1–3 relative to the total number of persons self reporting RA. Kappa coefficients (25) were calculated to determine pairwise agreement between methods 1–3 when classifying persons with or without RA. The unweighted sample was used to determine agreement, correcting for chance agreement.

Population prevalence estimates and 95% confidence intervals (95% CIs) were determined using each of the 3 classification methods. Stratified prevalence estimates (with 95% CIs) were generated for each method according to sex, race/ethnicity, age, and education. Due to limited sample size, estimates for persons from “other” race/ethnicity groups are not reported. To obtain population prevalence estimates, the sample was weighted using the total 6-year examined (MEC) sample final weight produced by the NCHS. Sample weights adjust for differential selection probability from sampling methods (i.e., oversampling), noncoverage, and nonresponse. In the NHANES-III, data are clustered (correlated) due to sampling methods, which affects variance estimation. Statistical software capable of accommodating the complex survey design (26) was used to generate standard error estimates via Taylor linearization methods (27). Analytical guidelines recommended by the NCHS for uncommon events (P ≤ 0.25) were used to determine the reliability of prevalence estimates (28).

RESULTS

Demographic characteristics.

Demographic characteristics of persons ages 60 years and older in the US are described in Table 2. The majority of persons were women (57.2%), were of non-Hispanic white race/ethnicity (84.4%), married (60.9%), had a high school education or more (58.1%), had health insurance (96.7%), and were living above the poverty threshold (80.0%). There was an even distribution of persons younger than 70 years and persons 70 years of age and older.

Table 2. Demographic characteristics of persons 60 years of age and older in the US (n = 5,302)*
 Sample sizeWeighted proportion, %
  • *

    Some groups of percentages do not total 100% because of missing data.

Sex  
 Male2,58742.8
 Female2,71557.2
 Total5,302100
Age  
 60–69 years2,30950.2
 ≥70 years2,99349.8
 Total5,302100
Race/ethnicity  
 Non-Hispanic white3,04384.4
 Non-Hispanic black1,0738.3
 Mexican American1,0272.3
 Other1595.0
 Total5,302100
Education  
 Less than high school2,99041.4
 High school or more2,27458.1
 Total5,26499.5
Marital status  
 Married3,07860.9
 Widowed1,57227.2
 Divorced/separated4187.5
 Never married2254.3
 Total5,29399.9
Poverty index  
 At or below the poverty threshold1,00010.4
 Above the poverty threshold3,67880.0
 Total4,67890.4
Health insurance coverage  
 Yes4,99796.7
 No2012.4
 Total5,19899.1

Definition of RA cases.

The 3 methods used to classify RA cases yielded similar results (Figures 1–3). Consistent with the findings of Arnett and colleagues (14), the ACR classification tree (Figure 2) identified more cases of RA than did the “n of k” decision rule (Figure 1). These methods classified 132 and 128 subjects as having RA, respectively. Addition of the medication data to the ACR tree algorithm increased the number of cases classified as RA. Using method 3, a total of 144 persons were classified as having RA. Overall, 12 additional cases of RA were classified using this method compared with the ACR tree algorithm alone.

Figure 1.

Results of classification method 1 (n of k) used to identify rheumatoid arthritis (RA) cases. Highlighted boxes depict the number of cases of RA that were identified using this method (n = 128). NHANES-III = Third National Health and Nutrition Examination Survey; MEC = mobile examination center; ACR = American College of Rheumatology.

Figure 2.

Results of classification method 2 (classification tree) used to identify rheumatoid arthritis (RA) cases. Highlighted boxes depict the number of cases of RA that were identified using this method, where metacarpophalangeal (MCP) joint swelling was substituted for radiographic findings, and wrist swelling was substituted for missing rheumatoid factor (RF) findings (n = 132). NHANES-III = Third National Health and Nutrition Examination Survey; MEC = mobile examination center.

Figure 3.

Results of classification method 3 (classification tree/disease-modifying antirheumatic drugs [DMARDs]) used to identify rheumatoid arthritis (RA) cases. Highlighted boxes depict the number of cases of RA that were identified using this method, where the American College of Rheumatology (ACR) classification tree and reported use of DMARDs were used to verify cases (n = 144). NHANES-III = Third National Health and Nutrition Examination Survey; MEC = mobile examination center; MCP = metacarpophalangeal; RF = rheumatoid factor.

When the 3 methods were compared, classification differences occurred in <1% of the study sample (Table 3). When corrected for chance, agreement (25, 29) was excellent between methods 1 and 2 (κ = 0.921), methods 2 and 3 (κ = 0.955), and methods 1 and 3 (κ = 0.879). When method 3 was expanded to incorporate B-list drugs into the algorithm, such that RA was identified in those who self-reported RA and took a B-list drug, an additional 11 persons were classified as having RA, yielding a total of 155 cases. Expansion of method 3 was done to identify potential underestimation of RA cases, but it was not used to determine actual prevalence estimates for RA.

Table 3. Agreement between the 3 RA classification methods*
Method 1 (n of k)Method 2 (classification tree)Method 3 (tree/DMARDs)Frequency, no.
  • *

    The overall percentage of agreement between methods is 99.4% (n = 5,270) for agreement and 0.6% (n = 32) for disagreement. Pairwise agreement between methods corrected for chance is as follows: for methods 1 and 2, κ = 0.921; for methods 2 and 3, κ = 0.955; and for methods 1 and 3, κ = 0.879. RA = rheumatoid arthritis; DMARDs = disease-modifying antirheumatic drugs.

RARARA120
No RANo RANo RA5,150
No RANo RARA12
No RARARA12
RANo RANo RA8
128 cases132 cases144 cases 

Among persons who self-reported RA (n = 389), methods 1–3 concurred with the self reports in 22 (5.7%), 29 (7.5%), and 39 (10%) cases, respectively. Of the 105 respondents who did not report RA but were classified as having RA by method 3 (the method that classified the most persons with RA), 19 (18%) reported osteoarthritis, 46 (44%) reported arthritis but did not know the type, 2 (2%) reported arthritis but did not respond to the question about the type of arthritis, and the remaining 38 respondents (36%) did not answer affirmatively to the question “Has a doctor ever told you that you had arthritis?”

Estimates of the prevalence of RA.

Population prevalence estimates for RA, stratified by age, sex, education, and race/ethnicity, for the 3 classification methods are shown in Table 4. Overall prevalence estimates using the 3 classification methods were 2.03% (95% CI 1.30–2.76), 2.15% (95% CI 1.43–2.87), and 2.34% (95% CI 1.66–3.02), respectively. The prevalence of RA was generally greater (by ∼1.5-fold) in women (2.35–2.71%) than in men (1.59–1.85%) and in persons ages ≥70 years (2.46–2.80%) than in persons ages 60–69 years (1.59–1.89%). The prevalence of RA in persons who did not complete high school ranged from 2.25% to 2.64%, compared with 1.82% to 2.09% in those who had a high school education or more. Point estimates for the prevalence of RA by race/ethnicity ranged from 1.94% to 2.55% in non-Hispanic blacks, 2.10% to 2.41% in non-Hispanic whites, and 2.43% to 2.80% in Mexican Americans.

Table 4. Stratified population prevalence estimates for RA using the 3 classification methods*
 Sample sizePrevalence of RA, % (95% CI)
Method 1 (n of k)Method 2 (classification tree)Method 3 (tree/DMARDs)
  • *

    Separate prevalence estimates for “other” race/ethnicity groups are not reported because of limited sample size. All other prevalence estimates include all race/ethnicity categories. RA = rheumatoid arthritis; 95% CI = 95% confidence interval; DMARDs = disease-modifying antirheumatic drugs.

All subjects5,3022.03 (1.30–2.76)2.15 (1.43–2.87)2.34 (1.66–3.02)
Sex    
 Male2,5871.60 (0.84–2.36)1.59 (0.82–2.36)1.85 (1.04–2.66)
 Female2,7152.35 (1.35–3.35)2.57 (1.58–3.56)2.71 (1.73–3.69)
Race/ethnicity    
 Non-Hispanic white3,0432.10 (1.30–2.90)2.21 (1.42–3.00)2.41 (1.65–3.17)
 Non-Hispanic black1,0731.94 (0.48–3.40)2.26 (0.84–3.68)2.55 (1.22–3.88)
 Mexican American1,0272.43 (0.67–4.19)2.47 (0.71–4.23)2.80 (1.02–4.58)
Age    
 60–69 years2,3091.59 (0.98–2.20)1.66 (1.04–2.28)1.89 (1.31–2.47)
 ≥70 years2,9932.46 (1.26–3.66)2.63 (1.42–3.84)2.80 (1.59–4.01)
Education    
 Less than high school2,9902.25 (1.44–3.06)2.49 (1.66–3.32)2.64 (1.81–3.47)
 High school or more2,2741.82 (0.90–2.74)1.86 (0.94–2.78)2.09 (1.24–2.94)

DISCUSSION

In this study, the prevalence of RA in the civilian noninstitutionalized population of the US ages 60 years and older was estimated using 3 classification methods. The NHANES-III provided a large, nationally representative sample for analysis. All of the ACR 1987 criteria for classifying RA, except radiographic data, were available from the NHANES-III. Data on the use of antirheumatic and antiinflammatory medications were also available. Product names were verified from container labels, ensuring accuracy of medication reporting. These strengths contribute to the validity and accuracy of the prevalence estimates provided in this study.

The 3 methods of RA classification used in this study, the “n of k” ACR criteria, the ACR classification tree, and the ACR classification tree augmented by medication data, yielded similar results. Classification differences occurred in <1% of the study sample when applying these methods. Consistent with the findings reported by Arnett and colleagues (14), we found that the classification tree was more sensitive than the “n of k” method for classifying RA. One benefit of the tree classification method is the comparability of classification when a surrogate variable is substituted for missing data (14). Since radiographic data were not available for the current study, MCP swelling could be substituted without negatively affecting the properties (i.e., sensitivity and specificity) of the classification method.

In method 1, the potential for overestimation of the prevalence of RA existed because 3 of 6 rather than 4 of 7 ACR criteria were used. Since method 2 allowed substitution of variables, the comparability of RA prevalence estimates between methods 1 and 2 suggests that the absence of radiographic data did not adversely affect case classification and therefore prevalence estimates. Furthermore, the classification agreement between these methods was excellent (κ = 0.921). This is an important finding, since collection and interpretation of radiographic data in large population studies is costly.

Data on the use of medications that are typically prescribed for the treatment of RA proved to be useful in identifying additional cases of RA. Compared with method 2, a total of 12 additional cases of RA (9%) were classified by incorporating the use of DMARDs (A-list medications) into the classification tree algorithm (method 3). A total of 16 additional cases of RA (12.5%) were classified with method 3 compared with method 1.

Overall and pairwise agreement between methods was excellent, although this can be attributed in part to the low prevalence of RA. Method 3 was more sensitive than the other methods, although this method may have incorrectly identified rheumatic conditions other than RA for which DMARDs may be prescribed, such as psoriatic arthritis. Early treatment of RA with antiinflammatory and antirheumatic drugs has been advocated to reduce inflammatory activity and minimize joint damage (30–32). Since the use of these drugs may reduce symptoms of RA, such as joint swelling, classification methods that rely predominantly on the presence of signs and symptoms of active disease may not detect cases of RA that are under pharmacologic control. Thus, classification of RA in large population surveys may be enhanced by the incorporation of data on prescription medication use. However, method 3 may overestimate the prevalence of RA by identifying persons with other rheumatic conditions who take DMARDs.

In addition to the RA cases identified by method 3, another 11 cases of RA were classified by incorporating the use of glucocorticoids (B-list medications) into the classification tree algorithm. Since prescription of glucocorticoids is not specific to RA and since the diagnosis of RA was potentially inaccurate because it was self reported, use of these medications was not incorporated into the final algorithm for method 3. Therefore, prevalence estimates based on method 3 may be underestimated.

Estimates of the prevalence of RA from this study are roughly comparable to those in other reports from the US. Overall, we found the prevalence of RA to be between 2.03% and 2.34%, depending on the classification method, for adults ages 60 years and older. In women, RA prevalence ranged from 2.35% to 2.71%, and in men, it ranged from 1.60% to 1.85%. In the Tecumseh Community Health Study (conducted from 1959 to 1960), the prevalence of definite RA in persons ages 60 years and older was ∼1% for men and ∼2% for women, as defined by the ACR 1958 criteria (33). In a recent review of RA prevalence in the US, Lawrence and colleagues (13) reported that in the National Health Examination Survey (for the years 1960–1962), definite RA was identified in 1.8% of men and 4.9% of women between the ages of 65 and 79 years, using the ACR 1958 criteria for definite and classic RA. In 1964, residents of Sudbury, MA, were examined for RA (34). Using the ACR 1958 criteria, 3.8% of women and 1.3% of men ages 15–75 years met the criteria for probable or definite RA. Separate prevalence estimates were not provided for definite RA. However, of 117 persons identified with probable and definite RA, only 39 had definite RA. Using the sample of persons 15–75 years of age as the denominator (n = 4,552), this yields an overall prevalence of 0.9% for definite RA. In the NHANES-I (for the years 1971–1975), the overall prevalence of clinically diagnosed RA was 1.5% in persons ages 65–74 years (13). In Olmstead County, MN, the prevalence of RA in persons ages 35 years and older was 0.74% in men and 1.4% in women, as defined by the ACR 1987 criteria (35).

There are a number of reasons for the variations in prevalence estimates from study to study and in comparison with the current study. Methods for classifying RA have evolved over time, which affects case identification (14, 16, 17). Cases of probable RA that were included in the prevalence estimates from older studies may not be classified as RA based on current criteria (13). The size of the study samples differed, and some of the studies did not achieve national representation. In some studies, sample sizes were small for the population ages 60 years and older. Comparability of estimates is best determined between studies that use similar criteria. Although conducted in Finland, rather than the US, one of the most recent studies that used the current ACR criteria (the 1987 criteria) showed that the prevalence of RA in a random sample of 1,317 people ages 65 years and older was 1.2% in men and 2.2% in women (36). These values are remarkably similar to those identified in the present study.

In this study, concurrence between self-reported RA and classification via methods 1–3 was low (6–10%). Using a variety of classification methods, other investigators have also reported low concurrence, on the order of 20%, between RA classification methods and self reports of RA (37–39). Data from the NHANES-III provided a partial explanation for this lack of agreement. The majority of those who did not report RA but were classified as having RA according to method 3 (64%), misreported or did not know their type of arthritis. This suggests that while self reports of “arthritis” may be accurate, reporting errors increase when questions about the specific type of arthritis are included.

Persistent RA is marked by periods of exacerbation and remission. Disease is considered active when patients have signs and symptoms of active or chronic synovitis, with joint swelling and tenderness, often accompanied by elevation of one or more acute-phase reactants, such as the erythrocyte sedimentation rate or C-reactive protein level. Disease is considered inactive or in remission in the absence of these signs and symptoms; this usually occurs with DMARD therapy. Eight scenarios can be used to characterize RA disease states captured in cross-sectional studies such as the NHANES-III (Table 5). Disease burden in the population is represented best by classification into 1 of the first 7 categories. These are individuals who require ongoing medical intervention and in whom disability may exist and can progress. Respondents in 5 of these 7 categories (categories 1, 2, 3, 5, and 6) were captured in the NHANES-III survey. Since residual joint deformity was not part of the NHANES-III examination, cases in the seventh category would not have been detected. Failure to identify persons with inactive disease who have residual joint deformity could lead to underestimation of RA prevalence. MacGregor and colleagues found that the use of criteria that rely on the presence of current signs and symptoms of RA, such as the ACR 1987 criteria, led to clinically important underascertainment of cases (40, 41). Thus, the RA prevalence estimates presented here may be conservative. Respondents in the fourth category represent those with mild, early, or self-limited disease, those with an atypical presentation, or those with a different inflammatory polyarthritic disease. These cases would not be identified in the NHANES-III. In studies that evaluated the ACR 1987 criteria, it has been noted that early-stage, mild, and self-limiting disease may not be ascertained (18, 42). Respondents in the eighth category do not require ongoing medical intervention since they are not taking medications, do not have permanent joint deformity, and are not functionally limited. Therefore, ascertainment of these cases may not be warranted in large, costly population studies.

Table 5. Potential scenarios characterizing RA disease states in cross-sectional studies*
ScenarioActive diseaseACR criteria metDMARD prescriptionResidual deformity
  • *

    Subjects cannot satisfy the American College of Rheumatology (ACR) criteria in the absence of active disease. Residual deformity is defined as permanent abnormal structural change, such as limitation of joint motion or joint subluxation. RA = rheumatoid arthritis; DMARD = disease-modifying antirheumatic drug; NA = not applicable.

1YesYesYesNA
2YesYesNoNA
3YesNoYesNA
4YesNoNoNA
5NoNoYesNo
6NoNoYesYes
7NoNoNoYes
8NoNoNoNo

In summary, using data available from the NHANES-III, ∼2% of persons ages 60 years and older in the civilian noninstitutionalized population of the US were identified as having RA. Different classification methods yielded similar prevalence estimates, although identification of RA was enhanced by incorporating data on the use of DMARDs, an important consideration in large population surveys. The prevalence estimates reported here represent the proportion of the population who will most likely require medical intervention because of disease activity. Since the proportion of people in the US who are over the age of 60 years is growing and since the prevalence of disability from all causes increases with age, interventions that minimize the sequelae of RA will have a great impact on reducing disability among older Americans.

Ancillary