To estimate the prevalence of and the annual number of ambulatory health care visits for pediatric arthritis and other rheumatologic conditions.
To estimate the prevalence of and the annual number of ambulatory health care visits for pediatric arthritis and other rheumatologic conditions.
We used physician office visit, outpatient department visit, and emergency department visit data from the 2001–2004 National Ambulatory Medical Care Survey and 2001–2004 National Hospital Ambulatory Medical Care Survey to estimate annual visits for the International Classification of Diseases, Ninth Revision, Clinical Modification codes thought to represent significant pediatric arthritis and other rheumatologic conditions (SPARC). We converted visit estimates into prevalence estimates using data on the number of prior annual visits per patient. Synthetic estimates for states were produced using national rates.
The average annualized estimate of the number of children with SPARC was 294,000 (95% confidence interval [95% CI] 188,000–400,000). The annualized number of ambulatory health care visits for SPARC was 827,000 (95% CI 609,000–1,044,000).
Pediatric arthritis estimates have varied widely because it is an umbrella term for which there are many definitions and because it is a relatively uncommon condition from a population surveillance perspective. Our estimates suggest that arthritis-related health care visits impose a substantial burden on the pediatric health care system. One advantage of this surveillance paradigm is that it has established a starting point for tracking the national prevalence of arthritis and rheumatologic conditions in children on an ongoing basis using existing infrastructure rather than expensive new surveys. This surveillance system will help us monitor and predict the health care needs of patients with these conditions.
The prevalence of childhood arthritis in the US is not currently known. Data reported from 34 worldwide studies conducted between 1966 and 1998 showed a juvenile arthritis prevalence range of 7–401 per 100,000 children (1, 2). Applying these ranges to the almost 73 million children younger than age 18 years in the US during 2002–2003 (3, 4) yields estimates of 5,100–292,600 children with arthritis. Using only studies conducted in the mainland US (1, 2), the prevalence rate range is 9.2–220 per 100,000. Applying these ranges yields estimates of 6,700–160,500 US children with arthritis during 2002–2003. These wide ranges result from the following: 1) differing study definitions and criteria for what childhood arthritis is, including “ever affected” or “currently affected”; 2) differing ways of ascertaining cases (e.g., hospital based versus community based); 3) differing definitions of what a child is (age cutoffs ranged from 12 to 18 years old); 4) small sample sizes or durations in studies leading to chance variation in rates; and 5) temporal trends and geographic differences (e.g., rates may vary over time and place with socioeconomic, environmental, and genetic factors) (1, 2).
The introduction of the Arthritis Prevention, Control and Cure Act of 2004 (S. 2338), although not passed, provided impetus for the Centers for Disease Control and Prevention (CDC) to conduct studies on the prevalence of arthritis and other rheumatic diseases and report data on juvenile arthritis. In response, CDC and the American College of Rheumatology co-hosted in December 2004 a 1-day summit of surveillance experts, pediatric rheumatologists, and key stakeholders to review available data, consider options for estimating prevalence, and draft a list of conditions for ongoing surveillance. In June 2006, after a year-and-a-half process of gathering and considering additional input, seeking comments on varying surveillance case definitions, testing possibilities, and consulting with key constituents and partners (e.g., the Arthritis Foundation; the American Academy of Pediatrics; and the National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health), the CDC Arthritis Program finalized a paradigm for ongoing surveillance of pediatric arthritis. Mirroring that of adult surveillance, the method uses selected International Classification of Diseases (ICD) diagnostic codes from health care and other diagnostically based data systems. The present study defined this new approach and used it to estimate the national (and synthetic state) prevalence and number of ambulatory health care encounters for pediatric arthritis and other rheumatologic conditions.
The International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code set for surveillance of pediatric arthritis was developed over the period December 2004 to June 2006. Various approaches were considered (e.g., using the codes from an existing pediatric arthritis registry) (5) and comments were sought from the American College of Rheumatology Section on Pediatric Rheumatology, the American Academy of Pediatrics Section on Rheumatology, and individual consultants in rheumatology and surveillance. Our intention was to focus on capturing conditions that met the following criteria: 1) expected duration of ≥3 months; 2) likely to significantly impact a child's life; and 3) likely to involve or already involving joint, cartilage, or muscle. As options were considered, it was noted that for well over 10 years, a set of ICD-9-CM codes developed by the National Arthritis Data Workgroup (NADW) had been used to analyze and report arthritis burden for adults (6). Ultimately, we decided that maintaining parallelism in adult and pediatric arthritis surveillance had far more benefits and fewer downsides than other available options. Therefore, the final codes chosen were: 1) the subset of adult NADW ICD-9-CM codes thought to capture conditions, such as juvenile rheumatoid arthritis (JRA), that are most relevant and likely to represent significant pediatric arthritis or other rheumatologic conditions in children younger than 18 years old and 2) additional ICD-9-CM codes for significant diseases (not symptoms or signs) frequently and consistently requested for surveillance by the American College of Rheumatology Section on Pediatric Rheumatology, the American Academy of Pediatrics Section on Rheumatology, and pediatric rheumatologist and surveillance consultants to the CDC (Table 1). These selected codes were termed significant pediatric arthritis and other rheumatologic conditions (SPARC) codes.
|277.3||Amyloidosis (includes familial Mediterranean fever)*|
|390||Rheumatic fever without mention of heart involvement|
|391||Rheumatic fever with heart involvement|
|446||Polyarteritis nodosa and allied conditions|
|710||Diffuse diseases of connective tissue|
|711||Arthropathy associated with infections|
|713||Arthropathy associated with other disorders classified elsewhere|
|714||Rheumatoid arthritis and other inflammatory polyarthropathies|
|715||Osteoarthrosis and allied disorders|
|716||Other and unspecified arthropathies|
|720||Ankylosing spondylitis and other inflammatory spondylopathies|
|727.0||Synovitis and tenosynovitis|
|729.0||Rheumatism, unspecified and fibrositis|
|729.1||Myalgia and myositis, unspecified|
We used physician office visit, hospital outpatient department visit, and emergency department (ED) visit data from the 2001–2004 National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) (7) to estimate annual visits for the SPARC ICD-9-CM codes (Table 1). These surveys sample visits to physician offices and hospital clinics (not individuals), include up to 3 diagnostic codes assigned by health care professionals, and allow for disease-specific estimations. Because the physician office and outpatient clinic (but not the ED) visit data also include information on the number of past visits within the last 12 months for the patient represented by the record (categorized as 0, 1–2, 3–5 or ≥6 and excluding the current visit), we used those survey data to estimate the prevalence of children with SPARC in the manner suggested by Burt and Hing (8). Conceptually, the method assumes that the weighted number of visits for 1 record divided by the number of annual visits for the patient reflected in the record (visits/patient) equals the number of patients with that condition (i.e., visits / [visits/patient] = patients). Thus, if a sampled visit was for a new patient with JRA who had a current visit and the weighted number of visits estimated for that record was 1,000, then 1,000/1 = an estimated 1,000 patients with JRA, where 1 is the total number of annual visits for that patient (1 current visit + 0 prior visits). Similarly, for a sampled visit of a patient with JRA who had been seen 3–5 times previously during the past 12 months, if the weighted number of visits estimated for that record was 1,000, then 1,000/5 = 200 estimated patients with JRA (5 is the total number of annual visits for that patient: 1 current visit + 4 prior visits [the midpoint of 3–5]). Using the midpoint for the number of prior visits, the total number of visits was calculated as follows: 0 prior visits = 1 total visit; 1–2 prior visits = 2.5 total visits; 3–5 prior visits = 5 total visits; and ≥6 prior visits = 8 total visits.
We used the SAS software, version 9.1 (SAS Institute, Cary, NC) for data extraction and analysis. We estimated the weighted number of visits and patients for each year for each diagnostic code and averaged the estimates to produce an annualized 4-year estimate. We made these estimates when a SPARC code was a listed diagnosis anywhere on the record. Programming steps were taken to avoid double counting of any records. In the case where more than one SPARC code was present on a record, the first instance was used to assign the condition. Standard errors and 95% confidence intervals (95% CIs) for both the average annual visit and prevalence estimates were calculated based on the complex survey design variables (CSTRATM and CPSUM) using SUDAAN software (Research Triangle Institute, Research Triangle Park, NC). Because the 2001 data set did not contain these variables, they were created from other variables present in the data set (Schappert SM: personal communication).
National estimates were rounded to thousands. Because NAMCS and NHAMCS estimates based on <30 records or with relative standard errors >30% are considered unreliable (9), we clarified how estimates for specific conditions relate to these criteria.
State health departments have requested prevalence data on juvenile arthritis from the CDC because they lack such data in their own states. To address this data gap, we converted the above national ICD-based estimate into a rate by dividing the national estimate by the average number of US children younger than age 18 years during 2002–2003 (3, 4). We multiplied that rate by the average in the year 2002 and year 2003 state-specific population estimates (3, 4) of the number of children younger than age 18 years to produce state-specific estimates. We then applied the upper and lower 95% CI estimates for the national prevalence estimates to produce the range in rates to use for the state estimates. All state estimates were rounded to the nearest hundred.
The estimated number of children with SPARC was 236,000 in 2001; 364,000 in 2002; 292,000 in 2003; and 284,000 in 2004, for an average annualized estimate of 294,000 children with SPARC (95% CI 188,000–400,000). The corresponding prevalence rate was 403 per 100,000 (95% CI 257–548 per 100,000). The top 5 most common conditions were synovitis and tenosynovitis, myalgia and myositis (includes codes for fibromyalgia), osteoarthrosis and allied disorders, diffuse diseases of connective tissue, and rheumatoid arthritis and other inflammatory polyarthropathies (includes all codes for JRA and related conditions) (Table 2).
|ICD-9-CM code and condition†||Percent of prevalence||Percent of visits‡|
|727.0||Synovitis and tenosynovitis||31.3||23.2|
|729.1||Myalgia and myositis, unspecified||22.9||26.9|
|715||Osteoarthrosis and allied disorders||10.9||6.7|
|710||Diffuse diseases of connective tissue||7.2||6.4|
|714||Rheumatoid arthritis and other inflammatory polyarthropathies||5.4||7.4|
|711||Arthropathy associated with infections||5.2||6.9|
|716||Other and unspecified arthropathies||3.0||3.0|
|446||Polyarteritis nodosa and allied conditions||2.4||3.4|
|Remaining SPARC codes§||6.4||9.2|
The number of ambulatory care visits for SPARC increased from 665,000 in 2001 to 813,000 in 2002 to 828,000 in 2003 to 1,000,000 in 2004, for an average annualized estimate of 827,000 visits (95% CI 609,000–1,044,000), including an average annualized estimate of 83,000 ED visits. State estimates based on the ICD-9-CM definition ranged from 500 children with SPARC in Wyoming to 38,000 children in California (Table 3).
|Children <18 years old, no.†||ICD-9-CM–based estimates, no. (95% CI)|
|United States||72,969,000||294,000 (188,000–400,000)|
|Dist. of Columbia||110,300||400 (300–600)|
|New Hampshire||308,300||1,200 (800–1,700)|
|New Jersey||2,129,500||8,600 (5,500–11,700)|
|New Mexico||501,300||2,000 (1,300–2,700)|
|New York||4,573,000||18,400 (11,800–25,100)|
|North Carolina||2,078,100||8,400 (5,300–11,400)|
|North Dakota||146,800||600 (400–800)|
|Rhode Island||241,600||1,000 (600–1,300)|
|South Carolina||1,001,300||4,000 (2,600–5,500)|
|South Dakota||195,500||800 (500–1,100)|
|West Virginia||390,000||1,600 (1,000–2,100)|
The national annualized prevalence estimate for 2001–2004 was 294,000 children with SPARC in the US. Childhood arthritis-related ambulatory health care visits were estimated to be 827,000, imposing a substantial burden on pediatric health care systems.
These ICD-based estimates have a number of limitations. First and foremost, much disagreement exists among experts about clinical case definitions of childhood arthritis. Indeed, our prolonged efforts at developing a surveillance case definition revealed much variation about what consultants thought should be counted as pediatric arthritis. In response, we also provide estimates for conditions that some constituents unsuccessfully nominated for SPARC (see Appendix A). Thus, a constituent who thought that enthesopathy should have been included can see how much that estimate would add to the national estimate for what they consider to be childhood arthritis.
Second, our estimates are based on small numbers of records meeting the SPARC definition and these estimates, especially for components of the definition, may be unreliable (9). Nevertheless, one can observe (Table 2) that the majority of SPARC do not come from the conditions most think of as juvenile arthritis, e.g., JRA or connective tissue diseases. One can also observe much year-to-year variation in the estimates, again attesting to the difficulty in stably estimating a relatively rare condition from a small number of records in these surveys. To compensate for this variation, we included 4 years of survey data to smooth estimates. However, there was a shift in NAMCS weighting procedures beginning in 2003 that makes trend analysis crossing this year potentially erroneous (10). The small numbers also precluded our using regional rates to synthetically estimate the prevalence in states. Moreover, the state-specific estimates do not account for the variability in size of more and less susceptible subpopulations in states.
Third, because these surveys are based on contact with the health care system, children with SPARC but no health care may have been undercounted. Thus, children in remission may also have not been counted. In addition, because there are pockets of specialists, e.g., pediatric rheumatologists, whose practices may not have been included in the sample, there may be undercounting of cases. In contrast, children with SPARC who saw more than one physician may have been overcounted if the survey samples included both of these practitioners.
Fourth, our ICD-based estimate is based on diagnoses listed in any of 3 diagnostic fields of the sampled visit, as opposed to only the primary diagnosis for the visit. We chose to use any listed diagnosis as the basis for estimation, because although a SPARC code may not have been the primary reason for a visit, by virtue of it being listed one can assume it impacted on the visit in some way. For example, if a child with JRA went to a physician for an acute infection, the infection might be coded as the primary reason for the visit and JRA as a secondary reason. We did not want to miss counting such children. This decision may have caused some overestimation. If we were to use only the first-listed (i.e., primary) diagnosis, the annualized estimate would be 23% less (i.e., 226,000 children with SPARC [185,000 in 2001; 227,000 in 2002; 262,000 in 2003; 230,000 in 2004]). The annualized number of pediatric outpatient visits for SPARC using only the first-listed (primary) diagnosis would be 24% less (i.e., 627,000 [65,000 of which were ED visits] with 474,000 in 2001; 647,000 in 2002; 668,000 in 2003; and 719,000 in 2004). In contrast, using any diagnosis meant that in the case where a child had ≥2 SPARC codes, only the first encountered was used to assign the condition. Thus, a child with tenosynovitis and JRA listed in that order might be assigned to tenosynovitis, with JRA underestimated.
Fifth, we were unable to convert ED visits into prevalence estimates, which no doubt led to an undercount. Using the most conservative assumption that if the data on prior visits had been present and that each ED visit had the upper range (n = 8) of prior visits (for conversion of visits to children), then an additional 10,000 cases were not counted because we were unable to convert the ED visit data into prevalence. In contrast, if each such ED visit represented the only annual visit for that child, then an additional 83,000 children with SPARC were not counted in the estimate.
Sixth, the conversion of visits to prevalence assumes that the prior annual visit data are accurate. To the extent this parameter is mistaken, the estimates are imprecise. Another limitation is that misclassification and incorrect ICD coding can lead to estimate errors. We note that osteoarthrosis and allied disorders was one of the more frequently coded diagnoses (usually not as the first-listed diagnosis); however, osteoarthritis is most unusual in childhood and suggests misclassification or misdiagnosis. These problems of misclassification and misdiagnosis continue to be an issue, especially considering that the vast majority of these children were not diagnosed by a physician with specialty training in pediatric rheumatology. Previous studies have shown that adult rheumatologists see many children with arthritis and joint symptoms in general because of the lack of adequate numbers of pediatric rheumatologists. The obvious miscoding of the diagnoses found in this survey points out once again the need for more pediatric rheumatologists who can make the correct diagnosis of joint pain and swelling in the pediatric population. This fact again speaks to the other issue addressed by the Arthritis Prevention, Control and Cure Act of 2004: the critical undersupply and maldistribution of pediatric rheumatologists.
Beyond how SPARC is defined, at least 3 clinical classification schemes for childhood arthritis exist: JRA, juvenile chronic arthritis, and juvenile idiopathic arthritis. All 3 schemes do not include many of the conditions considered as arthritis and other rheumatic conditions in adults, and while a case counted in 1 classification system may not be a case in another system, all schemes define childhood arthritis as occurring in children younger than 16 years of age. SPARC, however, includes 16- and 17-year-olds in the estimates because we elected to maintain national arthritis surveillance across the complete age spectrum (adult surveillance starts at age 18 years). Regardless of the classification schemes, the purpose of SPARC is for ongoing surveillance and trend analysis with a broad population focus and not clinical diagnosis of individual cases of illness.
To gauge the reasonableness of the SPARC estimates, we looked at 2 other data sources. First, we used self-reported data from the 2001–2004 National Health Interview Survey (NHIS). In 2001, adults were asked, “Have you ever been told by a doctor or other health professional that you have arthritis?” In 2002–2004, adults were asked, “Have you ever been told by a doctor or other health professional that you have arthritis, rheumatoid arthritis, gout, lupus or fibromyalgia?” Assuming the questions were, for practical purposes, identical, we summed the weighted annual 18-year-old prevalences for each year and averaged them. From a total of 3,428 sampled 18-year-olds, the average annualized estimate of self-reported doctor-diagnosed arthritis for 18-year-olds was 119,600 (95% CI 79,400–159,900). Second, the 2001–2004 NHIS also contained a sampling of children in households. Adult respondents were asked, “Has a doctor or health professional ever told you that (sampled child's name) had any of these conditions?” The respondent was given a list of 10 conditions, one of which was arthritis. From child samples ranging from 12,249 to 13,579 over the 4 years, the average annualized estimate of arthritis prevalence among children younger than 18 years old was 80,100 (95% CI 51,500–108,600) (11).
That both of these self-report–based estimates are lower than the ICD-based estimates is not surprising. The SPARC definition includes many other conditions that the respondent may not think of as the conditions mentioned in the questions. For example, a respondent with tenosynovitis or polyarteritis nodosa might have answered “no” to the self-report question. For the 18-year-old–based estimate, poor recall may underestimate prevalence; the estimate also reflects a cohort experience and, therefore, is a lagging indicator of the current prevalence. For the proxy report on the child having arthritis, studies have shown that parental reporting of child arthritis is imprecise (12). Nevertheless, both forms of self-report provide a potential tool to monitor national trends over time. Although the adult self-report question is also used in the state-based Behavioral Risk Factor Surveillance System (BRFSS), state-specific 18-year-old BRFSS sample sizes averaged only 42 and, therefore, are not likely to be useful for state surveillance.
One aspect of the Arthritis Prevention, Control and Cure Act of 2004 was to address the need for pediatric rheumatologists. Although the estimates here give a data-based estimate of the burden of rheumatologic diseases in childhood, they cannot be used to estimate the need for pediatric rheumatologists. These practitioners diagnose and treat a broad range of conditions in addition to those considered to be classic rheumatic diseases. Whereas internal medicine-trained rheumatologists often confine their practices to the diagnoses listed in the NADW list, pediatric rheumatologists often serve as diagnosticians in their respective departments, seeing patients with a much wider spectrum of diagnoses. Indeed, a number of pediatric rheumatologists asked for the inclusion of conditions listed in the Appendix A, because they believed that these diagnoses comprised a large number of the patients seen in their practices, even if they were not defined as SPARC. However defined, pediatric arthritis-related health care visits clearly impose a substantial burden on pediatric health care systems, with more than three-quarters of a million visits annually. Adding in the visits for just a few of the conditions that pediatric practitioners see in their practice (see Appendix A) suggests that annual ambulatory care visits for conditions needing evaluation easily top 1 million.
Pediatric arthritis estimates vary widely, and therefore much difficulty is encountered in describing its epidemiology. The reasons for this are 3-fold: pediatric arthritis is an umbrella term covering a number of types of arthritis and related rheumatic conditions, there are a number of different clinical case definitions for pediatric arthritis, and pediatric arthritis is a relatively uncommon condition from a population surveillance perspective. These difficulties have inhibited progress in the field. The strength/merit of this surveillance paradigm is that in this contentious field, it has established a starting point for tracking the prevalence of arthritis and rheumatologic conditions in children on an ongoing basis using existing infrastructure rather than expensive new surveys. Its advantage over prior estimates is that these estimates are data based and national in scope. This surveillance system will help us monitor and predict the health care needs of patients with these conditions.
Dr. Sacks 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 design. Sacks, Helmick, Ilowite, Bowyer.
Acquisition of data. Sacks, Luo.
Analysis and interpretation of data. Sacks, Helmick, Luo, Ilowite, Bowyer.
Manuscript preparation. Sacks, Helmick, Ilowite, Bowyer.
Statistical analysis. Sacks, Luo.
We thank Gary Langmaid for computer programming, and Brian Feldman, MD, MSc, FRCPC (Hospital for Sick Children, Toronto); Rosemarie Hirsch, MD, MPH (National Center for Health Statistics); Reva Lawrence, MPH (National Institute of Arthritis and Musculoskeletal Diseases); and Patience White, MD, MA, FAAP (Arthritis Foundation) for work at the December 2004 meeting on considering surveillance options and the case definition.
|ICD-9-CM code and condition†||Percent addition to prevalence||Percent addition to visits‡|
|719.4||Pain in joint||97.7||94.8|
|726||Peripheral enthesopathies and allied syndromes||53.1||37.6|
|729.5||Pain in limb§||34.7||44.9|
|727.4||Ganglion and cyst of synovium, tendon, and bursa||28.0||17.3|
|719.0||Effusion of joint||19.6||13.0|
|728.8||Other disorders of muscle, ligament, and fascia||19.1||18.9|
|729.8||Other musculoskeletal symptoms referable to limbs§||13.6||15.4|
|719.9||Unspecified disorder of joint||12.8||13.9|
|719.6||Other symptoms referable to joints||10.6||7.5|
|717.7||Chondromalacia of patella§||9.1||6.0|
|727.8||Other disorders of synovium, tendon, and bursa||7.8||4.1|
|719.8||Other specified disorders of joint||6.7||2.9|
|719.7||Difficulty in walking||6.6||3.6|
|279.4||Autoimmune disease, not elsewhere classified§||6.3||2.9|
|756.83||Ehlers Danlos syndrome§||3.8||1.5|
|337.2||Reflex sympathetic dystrophy§||2.9||4.0|
|728.9||Unspecified disorder of muscle, ligament, and fascia||2.8||5.4|