Dr. Kremer has received consulting fees, speaking fees, and/or honoraria from Bristol-Myers Squibb, Centocor, Roche, and UCB (less than $10,000 each).
Interpreting registry-derived drug studies: Does societal context matter?
Article first published online: 29 OCT 2009
Copyright © 2009 by the American College of Rheumatology
Arthritis & Rheumatism
Volume 60, Issue 11, pages 3155–3157, November 2009
How to Cite
Kremer, J. M. and Greenberg, J. (2009), Interpreting registry-derived drug studies: Does societal context matter?. Arthritis & Rheumatism, 60: 3155–3157. doi: 10.1002/art.24880
- Issue published online: 29 OCT 2009
- Article first published online: 29 OCT 2009
- Manuscript Accepted: 11 JUL 2009
- Manuscript Received: 8 MAY 2009
If one were to set out to design a registry that would come as close to perfection as humanly possible, it would be ideal if disparate data sources from across the health care system could communicate with one another. That is, an inpatient registry of hospitalizations would connect with an outpatient registry of the particular disease and drugs being studied, which would also connect with a cancer registry and a death registry. And, for good measure, we could throw in an early disease registry of the particular malady being studied. But what if individuals moved or left the country? These details as well would be captured in a national population and emigrations registry. All of the data collected within each component of each of these registries would represent virtually complete electronic data within their respective domains for a society, and each would be able to communicate with the other so that the data on a single patient could be cross-referenced for events represented within each of the other registries. These overlapping systems would thus assure investigators working with the various data sets that important events associated with a particular diagnosis would not be missed, since a patient would inevitably show up in a matching data set if he or she were hospitalized for heart disease or malignancy, moved, emigrated, died, or simply shifted his or her care to a different site within the national network.
This system of electronic records actually exists in the Swedish national health care system, and investigators can mine the above-described national network of integrated registry domains, each with the ability to be cross-referenced with the other. The Swedish epidemiologic investigators in rheumatology have used their national system to produce a variety of significant contributions to our understanding of the epidemiology of rheumatic disease.
In this issue of Arthritis & Rheumatism, Askling and colleagues (1) report on the incidence of malignancies in their Swedish rheumatoid arthritis (RA) cohort starting therapy with a TNF inhibitor as compared with a group of RA patients starting therapy with methotrexate (MTX) and traditional disease-modifying antirheumatic drugs (DMARDs). The investigators report that the incidence of malignancies in the TNF inhibitor cohort is not increased as compared with the incidence in these other groups. This is important work because clinicians and patients are understandably very concerned about the long-term safety profile of these agents. As the authors emphasize in their article, it is not possible to adequately assess these long-term safety outcomes from the findings of randomized controlled trials (RCTs) of short duration. RCTs are designed with many exclusions for significant comorbidities, and patients undergo a washout period for previous interventions, although they typically have greater disease activity than is generally found in the population at large. This makes the external validity of the findings of RCTs less than ideal for “real-world” practice.
There is thus an obvious need for clinical information from large, prospectively followed registry data sets that can provide reliable data to the rheumatology community on a variety of outcomes of interest, including malignancy, cardiovascular disease, and serious infections. If the use of TNF inhibitors is associated with changes in the risk of developing such conditions, it is only by mining these large data sets that the many cogent questions regarding long-term toxicities can be legitimately addressed. These data sets should be derived from “real-world” experience, reflecting the actual prescribing patterns, disease activity, comorbid conditions, and treatment preferences endemic to the society of their origin.
What is remarkable and perhaps not widely appreciated is the degree of difference in patterns of utilization of biologic agents within the countries from which these data are derived—and the extent to which these differences are driven by each country's political, social, and economic realities. In Sweden as well as in the UK, with a nationalized single-payor health care system, the government is the single source of payment for drugs and can limit the use of biologic agents as well as apply minimum clinical disease activity criteria for their use. As a byproduct of the single-payor system, the government can define minimum levels of disease activity for access to biologic agents in the case of RA patients while mandating participation in a variety of registries.
This situation in the US is, of course, very different, with its scores of health care insurers, each with their own rules. But these payors need to remain sensitive to pressure from consumers within their payor network to accept and allow the prescribing of a biologic agent if it has been approved by the Food and Drug Administration. Since in the US, there is no payor with a large enough constituency to serve as the equivalent of the European government “stick” to promote the uniform collection of clinical information in a registry, the collection of data is also more of a challenge.
But, at the same time, RA patient consumers of biologic agents in the US have been spared the practice of defining a uniform, minimum level of disease activity in order to receive a biologic agent. Indeed, if a US rheumatologist says that a patient has failed to respond to MTX (and maybe another DMARD, such as hydroxychloroquine), that person can receive this class of drugs, even if their Disease Activity Score 28-joint assessment (DAS28) score is a scant 4.2 or less or if their tender and swollen joint counts are only 4 and 5, respectively, as has been documented as being typical in a large US registry of patients starting therapy with biologic agents (2).
Thus, just as the data from RCTs are very different from “real-world” registry data, it is apparent that not all registry data are the same. The registry data from each country will inevitably reflect the exigencies of the prescribing mandates within that society. Askling et al (1) describe the clinical characteristics of Swedish patients starting their first TNF inhibitor as ∼9 swollen and ∼9 tender joints with a DAS28 of 5.5. Approximately 10% of this RA cohort of Swedish patients have received a biologic agent. As noted, both the levels of disease activity and the frequency of treatment with biologic agents are more consistent with those reported in the UK (3), but are quite different from those reported in the US, where up to 40% of patients with RA are treated with these agents (4).
Can we accept the premise that levels of disease activity are driven by differing levels of inflammation and that this inflammation can be independently associated with a variety of outcomes of interest, including comorbid conditions such as malignancy, infection, and cardiovascular disease? If so, then it follows that different levels of disease activity in different populations may likely be associated with different outcomes, as reflected in registry data from these different societies. Indeed, particularly with regard to the risk of hematologic malignancies, the activity and severity of disease have been demonstrated to be a strong predictor of this comorbidity (5).
Another potentially important difference that is not widely appreciated is the different dosing protocols of biologic agents in the US versus those in some European countries. This may be particularly relevant for TNF inhibitors, given that a dose-dependent risk of malignancy has been reported from a meta-analysis of infliximab and adalimumab trials (6). European investigators raised criticisms of this RCT meta-analysis relating to the generalizability of these findings to countries such as the UK, where there is lower dosing (usually 3 mg/kg) of infliximab (7) than in some of the RCT arms. Of note, the UK dosing protocol for infliximab, for example, differs markedly from that in the US, where it has been noted to be ∼5 mg/kg in 1 US cohort (8). If there is in fact a dose-dependent risk of toxicity, as suggested by the meta-analysis, then it would follow that the European experience of biologic safety may not be readily generalizable to US patients being prescribed higher doses of these particular biologic agents.
The superb data in the report from Askling et al (1) are inherently valid, as are the many fine pharmacoepidemiologic studies from other European countries (4, 9, 10). It is perhaps more likely that the data from within Europe can be expected to be more consistent with each other than with those from the very different system in the US because of the potentially relevant and significant differences described. In other words, whereas the internal validity of the study may be strong, drawing conclusions about the generalizability (external validity) of the findings to RA patients in the US may be more challenging.
However, if the prescribing patterns differ among countries, such that biologics are prescribed for patients with more severe disease and higher inflammatory burdens in Sweden (and Europe) than in the US, then the finding of the same incidence of cancer in patients starting treatment with biologic agents versus traditional DMARDs renders the findings of this study even more compelling—as being indicative of an absence of an increased risk of malignancy associated with these agents.
But, in fairness to the process, what if the opposite finding had emerged? That is, if the investigators had hypothetically found that there was indeed an increased risk of malignancy in the population treated with biologic agents? This outcome would lend credence to a suggestion, or at least a consideration, of the possibility that a channeling bias for more severe disease in the population studied could diminish the possible relevance for a US population of patients with less severe disease receiving these same drugs. In addition, it might be pointed out that the drug-related skin cancer risk in Sweden (latitude 55–60+ degrees North) may not be as relevant to the US population (latitude 35–43 degrees North). Geographic differences in skin cancer are an obvious, but minor, point, and the issue is only raised in the context of the real differences, societal and geographic, between registries.
In summary, the report by Askling et al in this issue of Arthritis & Rheumatism is a superb example of what can be accomplished with a close-to-ideal government-supported infrastructure and a very talented group of epidemiologic investigators. It is, however, possible that the rheumatology community that reads and processes the published reports from these rich and varied data sources can begin to become a bit more aware of the many real societal and contextual differences that could potentially impact outcomes of interest, depending on the countries from which the data are derived.
- 3Predictors of response to etanercept and infliximab among patients with rheumatoid arthritis. Ann Rheum Dis 2005; 64 Suppl 3: 1398., , , .
- 9British Society for Rheumatology Biologics Register Control Centre Consortium, Silman AJ, et al, on behalf of the British Society for Rheumatology Biologics Register. Rates of serious infection, including site-specific and bacterial intracellular infection, in rheumatoid arthritis patients receiving anti–tumor necrosis factor therapy: results from the British Society for Rheumatology Biologics Register. Arthritis Rheum 2006; 54: 2368–76., , , ,