Care needs to be taken when reviewing malignancy data reported from observational studies: Comment on the article by Onel and Onel


  • Xavier Mariette MD, PhD,

    1. Assistance Publique-Hôpitaux de Paris, Hôpital Bicêtre, INSERM U 1012, and Université Paris-Sud 11 Le Kremlin Bicêtre, France
    Search for more papers by this author
    • Dr. Mariette has received consultant fees (less than $10,000 each) from Pfizer and UCB.

  • Alan Reynolds B Tech,

    1. Reynolds Clinical Sciences Eastleigh, Hampshire, UK
    Search for more papers by this author
    • Dr. Reynolds was employed by Pfizer until June 30, 2010 and has received consultant fees, speaking fees, and/or honoraria (less than $10,000) from Napp and (more than $10,000) from Pfizer.

  • Paul Emery MA, MD, FRCP

    1. University of Leeds, NIHR Leeds Musculoskeletal Biomedical Research Unit, and Leeds Teaching Hospitals NHS Trust Leeds, UK
    Search for more papers by this author
    • Dr. Emery has received consultant fees (less than $10,000 each) from Pfizer, Merck, Abbott, and UCB.

Care Needs to be Taken When Reviewing Malignancy data Reported from Observational Studies: Comment on the Article by Onel and Onel

To the Editors:

Onel and Onel address an important issue in their review published recently in Arthritis Care & Research, but we are concerned that some of the points they make are a result of misinterpreting the published data (1). Tumor necrosis factor (TNF) has a paradoxical role since it plays a part in host defense against cancer but also in the spread of some malignancies (2). Shortly after the introduction of TNF inhibitors for the treatment of rheumatoid arthritis (RA), concern was expressed about the long-term safety of these agents, including the possibility of a risk of malignancy (3). Some, but not all, meta-analyses of randomized clinical trials suggest that there may indeed be an increased risk of infection and malignancy (4–8). These analyses are limited, however, due to the relatively short duration of randomized trials and to the restrictions on eligibility that limit enrollment of patients with comorbidities. The availability of data from large-scale long-term observational studies of patients treated in ordinary care overcomes those limitations, but there are still issues resulting from the nonrandomised nature of the study design, which leads to differences between exposed and nonexposed cohorts. These differences then have to be dealt with by complex statistical techniques.

In the context of malignancy in particular, it is important to understand the exact nature of the populations with which the exposed cohort have been compared. In the randomized controlled trials this is usually the placebo control arm, i.e., patients with the same disease not exposed to the intervention. In the observational studies, comparisons may be made with a nonexposed patient control cohort or with the general population, adjusting for age and sex. In the comparison with a nonexposed cohort, the results are reported as relative risk (RR), odds ratio, or hazard ratio. When the comparison is with the general population, the standardized incidence ratio (SIR) is reported. Since patients with inflammatory diseases have increased (or reduced) risks of certain cancers compared with the general population (9), the results of these analyses can be very different. Unfortunately, the authors appear not to have appreciated these differences in some parts of the discussion in their review. Moreover, they compared data concerning lymphomas in one registry (10) with those concerning cancers in other registries (11, 12).

In the discussion on lymphoma, the authors correctly report the SIR from the French registry, Research Axed on Tolerance of Biotherapies (RATIO), as 2.4 (10), but they then compare this result with data from a Danish Crohn's disease (CD) registry (12) and the Swedish RA registry, Antirheumatic Therapies In Sweden (ARTIS) (11), both of which concern general malignancy. This is especially inappropriate when a specific lymphoma publication from ARTIS was available (13). The review reports that the SIR from these two studies was close to 1.0, which is very different from the 2.4 found in the RATIO. The authors state, “Nonetheless, there is a tremendous difference between a SIR of 2.4 and a SIR of less than 1.” Therefore, the authors have committed a double error, first by quoting the RR values (compared with nonexposed patients) yet calling it SIR, and second by comparing rates of two different pathologies, i.e., lymphomas versus general malignancy. Furthermore, in the discussion following, they mention again the specific SIR (from the French RATIO registry) for each anti-TNF agent and state that it concerns cancer development, although in fact it concerns only lymphoma.

The ARTIS lymphoma data are very clear: the SIR for lymphoma was 2.72 (95% confidence interval [95% CI] 1.82–4.08), whereas the RR was 1.35 (95% CI 0.82–2.11) (13). The results from the US National Data Bank for Rheumatic Diseases are similar: a SIR of 2.9 (95% CI 1.7–4.9) and an RR of 1.0 (95% CI 0.6–1.8) (14, 15). The differences between the RR and SIR values reflect the increased risk of lymphoma in patients with RA, which is probably a consequence of the inflammatory burden over time (16).

As a result of this error, the authors have unnecessarily raised concerns about “dramatically” different results between studies, whereas, in fact, the results were remarkably consistent. However, we do agree with their point that additional research is needed to provide greater precision to any risk since currently the CIs are wide.

We are also concerned about some of the discussion regarding risk of malignancy in patients with juvenile idiopathic arthritis (JIA). The authors comment, “Additionally, although cancer was most commonly associated with the use of anti-TNF agents in patients with CD, an increased risk for cancer was also found in children treated with TNF inhibition for other primary autoimmune disorders, including JIA. This observation suggested that it was the exposure to anti-TNF drugs that was driving the association, rather than the underlying disease pathology itself.” However, as very well stated elsewhere in the review, the risk of malignancy in juveniles with these diseases is not known; therefore, it is not possible to state that an increased risk has been found. Numerically the rates are higher than those found in the general population (25.7/100,000 versus 16.6/100,000), but this could occur for a number of reasons in addition to drug exposure (anti-TNF and/or associated immunosuppressive drugs). Indeed, later in the article the authors do consider factors that could confound the findings.

In conclusion, we believe that the authors have missed the opportunity to clearly summarize current evidence on the association between treatment with TNF inhibitors and the risk of malignancy and, by misunderstanding the data, have unnecessarily raised concerns about the lack of consistency between publications. To summarize, the SIRs clearly show an increased risk of lymphoma in TNF inhibitor–treated patients compared with the general population, but the RRs from the registries show no significant increased risk compared with disease controls. This should be the message for patients.

Xavier Mariette MD, PhD*, Alan Reynolds B Tech†, Paul Emery MA, MD, FRCP‡, * Assistance Publique-Hôpitaux de Paris, Hôpital Bicêtre, INSERM U 1012, and Université Paris-Sud 11 Le Kremlin Bicêtre, France, † Reynolds Clinical Sciences Eastleigh, Hampshire, UK, ‡ University of Leeds, NIHR Leeds Musculoskeletal Biomedical Research Unit, and Leeds Teaching Hospitals NHS Trust Leeds, UK.