Rheumatoid arthritis (RA) is the major disorder which dominates clinical rheumatologic practice, textbooks and journals of rheumatology, as well as presentations at conferences, including those of the American College of Rheumatology. Thus, rheumatologists should examine with great interest reports such as the one by Doran and colleagues, which appears elsewhere in this issue of Arthritis & Rheumatism (1). The authors suggest that the incidence of RA has fallen by half over the last 4 decades. Such observations are important, however, for those planning the provision of health care (including assessing the costs for new and expensive therapies) and for those seeking etiologic clues to the cause of this condition.
It is important to consider first how trends in the incidence of a disorder such as RA can be ascertained. In defining the burden of a disease such as RA, which has a variable activity and course (2), the incidence, a measure of cases of new onset, is far superior to the prevalence, a measure of the proportion “in state.” The ascertainment of RA incidence, however, requires the identification of all new cases that arise within a target population. Further, given the rarity of RA, the population needs to be large enough to obtain robust estimates, particularly when stratification by age and sex is required.
In addition, the determination of trends requires observation for several years. Unlike cancer, for instance, the diagnosis of RA is not easily attainable from routine sources, but requires the careful documentation of clinical and laboratory variables. Such a process is expensive, and there are very few examples worldwide where such an endeavor has been undertaken. The largest is the Norfolk Arthritis Register in the UK, which aims to identify (or “capture”) all new patients with inflammatory polyarthritis who come to primary care providers and to visit these new patients within 2 weeks of their first office visit (3). To achieve this for an adult population of around half a million requires 7 trained research nurses who are employed specifically to characterize the nature of the arthritis. This activity is clearly costly, and an appropriate alternative is to use existing diagnostic data sets from a fixed population base. In this way, it is possible to assess RA incidence retrospectively from routinely gathered clinical records.
Such a system can be developed nationally, as in Finland (4). That system was developed for health insurance purposes: registered individuals with RA (among other diagnoses) are eligible to receive free medications. The data gathered have been used over several decades to provide estimates of RA incidence (5, 6), although clearly, diagnostic misclassification is likely in an important proportion of patients, given the goals of the registry. In the US, there have been remarkably few attempts to estimate RA incidence. A study in Puget Sound, Washington, which was centered around a health maintenance organization, was designed primarily to recruit women for a study of RA and oral contraceptive use (7). That study provided an estimate of RA incidence in women. Data from another health maintenance organization in Massachusetts (8) also provided estimates, but such material, gathered over a short time span, cannot easily be used to assess trends in incidence.
It is against this background that the unique possibilities afforded by the record system covering the population of Olmsted County, Minnesota, based on attendance at the Mayo Clinic and associated health facilities, has been used to chart the occurrence of RA in that population over several decades, with interesting findings. The first report, covering 1950–1974 (9), suggested a substantial decline in RA incidence in women, but not in men. Such data were used to support the hypothesis that the increasing use of oral contraceptives was protective against RA (10). Interestingly, a case–control study of the Mayo Clinic cases observed during that period did not confirm a negative association with oral contraceptive use (11). Further reports of trends in RA incidence in Olmsted County have been published, incorporating, but extending, these earlier data. Thus, the second report covered the period 1955–1985 (12), with the current report (1) extending the observations to 1995. This current analysis has charted a continuing decline in RA incidence over the 4 decades, from 61.2/100,000 person-years to 32.7/100,000 person-years. The decline was substantially greater in women, but the smaller number of incident cases in men makes a true comparison of trends difficult.
As pointed out elsewhere (13), as well as in the report by Doran et al (1), there may be a number of artifactual explanations for such a decline. These include secular changes in diagnostic labeling, laboratory testing, and medical record quality. However, such explanations seem unlikely to explain the declines observed. Indeed, the decline in incidence in Olmsted County, Minnesota, is consistent with recently observed declines in several other populations, including Pima Indians (14) and the populations of Wakayama, Japan (15), and Finland (5). Thus, the decline seems to be a worldwide phenomenon.
In both Finland and Japan, the decline in RA incidence has been associated with a shift toward a more elderly age at onset (16, 17), an observation that merits further investigation. Such a shift was not, however, found in the study by Doran et al (1), although this could have been missed by their reliance on between-decade comparisons of the mean age at RA diagnosis, rather than a comparison of the age-specific incidence rates. The number of RA cases in Olmsted County over each of the 4 decades and across all age groups studied was constant. However, the investigators also reported that the RA incidence increased with age and decreased with time. These apparently incongruent findings can be reconciled only if the average age of the population itself has declined for demographic reasons. This being the case, then their data also support an underlying increasing mean age at onset of RA.
For the epidemiologist, the one obvious explanation for both a decline in incidence and shift toward an older age at onset is that of a birth cohort effect. The hypothesis is that successive generations, conditioned possibly by an early life influence, are successively less likely to develop RA. Hence, as time passes, it is the older individuals from more historic cohorts that are at greater risk. Indeed, such a phenomenon has been demonstrated for RA in both Japan and Finland (16, 17). Separating out the influences of age, year of onset (“period”), and year of birth (“cohort”) is not easy, since any 2 of these variables provide the value for the third. Formal age–period–cohort analysis can be used to compare the effects of “period” and “cohort,” although this was not undertaken by Doran et al. The graphic representations of their data are, however, consistent with such a decline (1).
Is there other evidence consistent with a birth cohort effect? A birth cohort decline in the production of rheumatoid factor (RF), independent of RA status, was recorded several years ago in an industrial population of the UK studied by Lawrence (18), who postulated that this was related to an improvement in air quality. Recent data from the Pima Indian population have also suggested a birth cohort decline in RF production (19). It was also shown 20 years ago that within RA patient populations, there was a birth cohort decline in RF production; i.e., at the same age, increasingly recent generations of patients with RA were less likely to be RF positive (20).
The fascinating data reported by Doran et al have broad implications that range from basic science to public health to health care finance and delivery worldwide. Each disease has its own ebb and flow in time. Perhaps RA will be like rheumatic fever and diminish in the West. While the basis for the changing incidence of RA will require future research, the findings by Doran and colleagues are important and give added impetus to the theory that the occurrence of RA is determined, at least in part, by an environmental agent, whose greatest effect might be in early life and whose occurrence or infectivity has declined.