Results

In the adult population covered by NOAR, the overall incidence of RA was 32 per 100 000 per year and 51 per 100 000 per year for IP. Five areas, on visual inspection, appeared to include a possible cluster. Three areas (table 1) showed significantly higher incidence rates. The total adult population of area C was only 85, who lived in two adjacent roads. During the period of observation, five of these 85 people developed IP, including three that satisfied criteria for RA. The Poisson probabilities of these number of cases occurring by chance in each of these three areas are, as shown, very small. The proportion of men and women in each of these three areas was identical to that observed in the original NOAR catchment population.6Furthermore, the proportion aged over 65 in the three areas were 18%, 20% and 21% respectively, the proportion of the NOAR area as a whole being 23%.6

We then compared the incidence of IP and RA in the areas immediately surrounding the clusters with that of the cluster areas. In the postcode areas surrounding areas A, B and C, the overall incidence of IP per 100 000 person years was 94.6, 61 and 55 respectively, in each case higher than the overall incidence of 51/per 100 000 person years.

Limited analysis of the clinical data available on the subjects in each of these three clusters (table 2) did not suggest a high degree of disease homogeneity, though the number of cases was clearly too small to make any definitive statement. In addition to the information presented, we examined the pattern of joint involvement, presence of nodules and the timing of onset, and no obvious similarity was noted within clusters. Perhaps the main point of interest is that, within area C, which had the highest probability of being a cluster, four of the five cases were rheumatoid factor positive. We examined data collected on all people at registration by NOAR for common triggers of disease. However, we could find no similarity between the cases in terms of symptoms of infection before disease onset, prior immunisation, occupation, pregnancy or smoking. Finally, none of the cases were related to one another.

Discussion

The interpretation of reports of clustering is fraught with difficulty, although their presence provides some interesting insights into the role of environmental factors in disease onset. Care has to be taken in the statistical evaluation of potential clustering. In a disease as infrequent in the population as RA, in any small area, the existence of a single case, by definition, would create a greater than expected frequency. Across the NOAR populationas a whole, there was an incidence density of 51/100 000 person years. Consequently, in a population as small a size as area C, only 0.3 cases would have been expected during the six year period of observation. A single case is therefore in excess of this.

The Poisson probabilities calculated allow for this random occurrence and estimate how likely it is that the observed number of cases would have occurred in a denominator population of that size by chance. The difficulty is then determining whether such clusters are “real”. It is perhaps useful to distinguish investigations that investigate a large population for the existence of clusters, from those, such as the present investigation, which report on what are essentially anecdotal collections of cases that appear to occur unusually closer together. The statistical approach used in this study confirms the anecdotal impression that these clusters are unlikely to have occurred by chance. The results obtained should be treated with caution in so far as only a very small proportion of the potential geographical areas that could have been investigated, were actually analysed. As an illustration, in the total NOAR catchment population of 413 000 adults, there are 4860 theoretical sub-populations, the same size as area C and 780 the same as area A and, theoretically, the p values obtained should be adjusted for this multiple significance testing. We have already shown elsewhere that in the entire NHA area there is modest evidence of spatial clustering.7 The current investigation adds to that observation and suggests that more formal techniques such as the use of a geographical analysis machine9 and other methods to detect what may be referred to as “smaller area database anomalies”10 would be appropriate.

One important concern is that any observed clustering represents localised selective over-reporting in specific neighbourhoods. Thus a person developing mild IP might be more likely to be ascertained if there is an awareness of other cases in the immediate vicinity. Similarly, reporting completeness is likely to vary by general practitioner, which might contribute to the likelihood of an apparent cluster. Notification to NOAR is based on attendance at primary care and we cannot exclude the possibility that an important proportion of mild cases is not reported, except in the areas of the apparent clusters observed in this report. Against this theory, however, is the fact that the cases observed in these clusters had similar spectrum of disease severity to NOAR cases as a whole.

The advantage of using a large population register to search prospectively for clusters, over anecdotal reporting of clusters, is that the possibility of clustering occurring by chance can be examined within the target population. As we have previously demonstrated, most cases of IP, including the specific subgroup with RA, are probably sporadic.7 This more detailed evaluation now suggests that there may be occasional georgraphical clusters that are unlikely to be attributable to chance. We also postulated that, if the same environmental trigger was responsible, then this would lead to a similar disease phenotype. This was not the case in the clusters identified.

The individual cases were ‘mapped’ to their address at the time of reported disease onset. This particular population is comparatively less mobile than the rest of the United Kingdom. However, given the unknown latency of IP, the relevant address at the time of any putative environmental exposure is unknown.

There are few clues as to the nature of any common environmental agent. Infection with parvovirus does not explain the clusters described in this report.11 Serum is, however, available for investigation of any future microbial hypothesis. Other putative environmental sources include the domestic water supply and, in this mixed rural-urban area, contamination from agricultural spraying.