Background The magnitude of the association of cancer with rheumatoid arthritis, especially after anti-TNF use, remains in dispute. We recently proposed (1) an important selection bias was potentially inherent in the registry data especially when the comparator for the sought cancer incidence in RA was the cancer incidence in the “mother population”, the population from which the registry is derived from. In a mother population within a given time there would be many patients with cancer who would not have the chance to develop RA since a. a sizeable fraction would die from their disease before having the chance to develop RA; b. The cancer treatment could potentially prevent the development of RA or finally, and particularly in the case of anti-TNF registries, c. If they remained alive and developed cancer the likelihood of them being included in such a registry would be small. All 3 factors could render the incidence ratio (incidence in the registry/incidence in the mother population) less than unity even if biologically there are no real differences in the cancer frequencies between the 2 populations. Lastly this ratio would decrease in time as was observed in the Taiwan registry (1).
Objectives We formally surveyed whether the same potential selection bias was present in other manuscripts reporting similar data.
Methods We conducted a PubMed search with the search terms “registry” “cancer” and “rheumatoid arthritis” among 7 high-impact rheumatology journals between 2001 and May-2011 (inclusive). First, articles that reported cancer incidence in patients with RA were retrieved in full-text. Among these, those manuscripts which reported an incidence ratio at 2 or more time-points were included in this survey. We specifically sought a. whether the comparisons were made between the registry and a “mother population” and b. the changes in the above described ratio between the first and last time points.
Results We retrieved 36 articles among 1274 search results. In 6/36 the incidence ratio comparisons were made between at least one RA registry and its mother population at more than 1 time point. The number of time-points ranged from 3 to 11. Sampling-period-length ranged from 8 to 40 years. In 5 of the 6 articles when a comparison was made with the mother population there was a reduction over time in the incidence ratio which ranged from 29 to 99%. In the remaining study there was no appreciable change in the incidence ratio.
Conclusions There are many important methodological issues for a fair assessment of cancer incidence in RA especially with the use of anti-TNF agents (2) We propose that the selection bias we try to highlight here should be included in this list of issues especially when we compare the cancer in a RA cohort with that in the mother population.
Arthritis Rheum. 2011;63:352-8
Pharmacoepidemiol Drug Saf. 2011;20:119-30
Disclosure of Interest None Declared
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.