Re:Confounding by indication invalidates most observational studies on side effects of glucocorticoids

William G. Dixon, MRC Clinician Scientist/ Senior Clinical Lecturer,

Other Contributors:

February 13, 2012

Dear Editor,

We thank the authors for their interest in our paper and extending the discussion about how we can best assess the safety of glucocorticoid therapy. Ongoing uncertainty about the safety of steroids - particularly at low doses - despite over sixty years of clinical experience illustrates this is not an easy challenge. Nonetheless, it is important to tackle given the continuing widespread use. We must move towards clearer numerical risks for multiple safety outcomes (1). The two study designs available to us to address steroid safety are randomized controlled trials (RCTs) and observational studies. Each has advantages and disadvantages. The great benefit of RCTs is randomised treatment, meaning each group has a balance of known and unknown variables at baseline. The authors suggest a literature review on the topic: a good idea that we have already undertaken (2). Unfortunately, on reviewing all published RCTs, there was great variability in the quality of reporting in the methods and heterogeneity in the results for safety outcomes, making it difficult to have confidence in the final result of the meta-analysis. Furthermore, the total number of patients eligible for inclusion was less than 2000 after more than 60 years' of glucocorticoid trials in rheumatoid arthritis (RA).

Observational studies offer the chance to look at larger numbers of patients, which is crucial to study infrequent serious adverse events. This approach also allows us to examine 'real-life experience', particularly over the long-term. There are, however, many challenges. First, glucocorticoid therapy is used dynamically, meaning that conventional models of risk-attribution make assumptions that we know not to be true. This was the motivation for the second publication where we showed that novel methods for modeling exposure that considered changes to dose with time provided a better fit than traditional models (3). Second, adherence to therapy is a problem in steroid safety, as we know patients can take either more or less prednisolone than is prescribed. Third, observational studies provide associations that may or not be causal. Our finding of an association between GC therapy taken several years ago and current infection risk does not necessarily mean this is a direct effect with long latency. Whilst it may be having a direct effect by influencing adaptive immune pathways, it may represent an indirect effect, perhaps by impairing barrier functions such as skin integrity, or reflect the delay in hypothalamic-pituitary-adrenal axis resetting after glucocorticoid suppression (4, 5). Past GC use may also be a marker of current active disease, which brings us onto the fourth and largest challenge: confounding by disease severity. GC prescriptions are tightly coupled to disease severity, making this a difficult problem. We agree administrative databases do not contain good measures of disease severity in RA, meaning residual confounding is likely. We acknowledged this in both papers, but in the first went on to show that the residual confounding could not explain all of the observed association (6). Interestingly, RCTs do not resolve entirely the problem of confounding by disease severity. Randomisation ensures the groups are matched for disease severity at baseline, but as patients in the treatment arm respond to GC therapy, they will have lower disease severity throughout much of the study. Any measured infection risk will thus represent a mixing of effects of the direct effect of GCs on infection (perhaps increasing risk) plus the effect of lower cumulative disease severity (reducing risk).

We agree with the authors that we must strive towards a clearer understanding of steroid safety through carefully designed studies. Unfortunately, we cannot resolve all challenges at once. The weighted cumulative dose model has now shown benefit in modelling time-varying doses, and will enable us to move towards individualised safety estimates for given patterns of GC use. We have studies funded that will examine the extent of adherence to glucocorticoid therapy within a national database, allowing quantification of the effect of self-adjusted therapy on safety estimates. In order to address confounding by disease severity (both in observational studies and RCTs), we need to have regular measurements of both disease severity and treatment. We can then use techniques such as marginal structural modelling (7). It is rare that large databases contain sufficient information for such analyses, but by linking national datasets that each contains components of the total required information, we can move beyond our current knowledge (8). We agree a large and pragmatic RCT would be valuable, assuming disease severity is measured carefully throughout and the design allows for variation in GC exposure, both between and within subjects, to reflect real-life patterns of GC use.

The associations reported in our papers must indeed be interpreted with caution, as we do not yet have solutions to all of the challenges of observational drug safety research. This leaves us with uncertainty about the safety of these widely used drugs. We believe it is important to continue to address this issue whilst, in parallel, attempting to address the methodological limitations.


1. Hoes JN, Jacobs JW, Buttgereit F, et al. Current view of glucocorticoid co-therapy with DMARDs in rheumatoid arthritis. Nature reviews Rheumatology 2010;6:693-702.

2. Dixon WG, Suissa S, Hudson M. The association between systemic glucocorticoid therapy and the risk of infection in patients with rheumatoid arthritis: systematic review and meta-analyses. Arthritis Research & Therapy 2011;13:R139.

3. Dixon WG, Abrahamowicz M, Beauchamp ME, et al. Immediate and delayed impact of oral glucocorticoid therapy on risk of serious infection in older patients with rheumatoid arthritis: a nested case-control analysis. Ann Rheum Dis 2012.

4. Rhen T, Cidlowski JA. Antiinflammatory action of glucocorticoids - new mechanisms for old drugs. N Engl J Med 2005;353:1711-23.

5. McMaster A, Ray DW. Drug insight: selective agonists and antagonists of the glucocorticoid receptor. Nat Clin Pract Endocrinol Metab 2008;4:91-101.

6. Dixon WG, Kezouh A, Bernatsky S, et al. The influence of systemic glucocorticoid therapy upon the risk of non-serious infection in older patients with rheumatoid arthritis: a nested case-control study. Ann Rheum Dis 2011.

7. Thamer M, Hernan MA, Zhang Y, et al. Prednisone, lupus activity, and permanent organ damage. J Rheumatol 2009;36:560-4.

8. Suissa S. Assessing the safety of new arthritis drugs: are we there yet? J Rheumatol 2008;35:2295-7.

Conflict of Interest

None declared