The recent NEJM article by Nissen and Wolski purporting to show an increased risk of cardiovascular outcomes among patients taking rosiglitazone is of particular interest to nephrologists because of the large number of diabetic patients that we see and the fact that alternative drugs are problematic in patients with a reduced GFR.

I believe that the study has a number of very important limitations, some of which have been raised by the authors themselves, but many of which seem to have been forgotten by some inflammatory editorialists.

Study selection – of the 48 studies that met the inclusion/exclusion criteria, the data from six were not included in the published analysis because they “…did not report any myocardial infarctions or deaths from cardiovascular causes and therefore were not included in the analysis because the effect measure could not be calculated.”

I find this argument to be specious. The effect measure that they refer to is the odds ratio and while it is true that this statistic cannot be calculated for each of the six trials individually (since 0/0 is undefined), that does not preclude the inclusion of the data in a summary odds ratio pooled from all 48 trials. The effect (i.e. whether the summary OR would move closer to or farther away from unity) of including the data on the pooled odds ratio is variable and depends on the relative number of patients in the rosiglitazone and control groups. The magnitude of the effect depends on the number of patients in the six trials relative to the other 42. Sadly, this vital information is lacking.

Heterogeneity – ensuring that trials are homogeneous is a basic tenet of meta-analysis since the average of apples and oranges is meaningless. To support homogeneity, the authors use the Cochran’s Q statistic and find that the null hypothesis of no heterogeneity could not be rejected. However, one has to ask: what is the statistical power of such a test when heterogeneity seems to be clearly manifest in table 4 in the paper? Here we see a 3-4-fold higher risk of MI and a 2-3-fold lower risk of CVD in the ADOPT study versus the other two sources. Furthermore, the risk of CVD is higher in the control group of the ADOPT study and lower in the other two sources. This has been attributed to the different spectrum of patients involved in the ADOPT trial ( i.e. patients further along in the course of their diabetes) – but that’s just another way of saying that the studies are heterogeneous.

Effect direction – The author’s use the “Peto method” to derive their summary OR for MI of 1.43 with a 95% confidence interval that just misses unity. However, if one calculates an un-adjusted OR for MI using the data in table 4, the OR favors rosaglitazone (OR = 0.97). One has to wonder about an adjustment process that not only changes the magnitude of an effect but also its direction.

Effect size – the cardiovascular events described in the study were uncommon (~0.6% for MI and 0.3% for CVD). Calculating an odds ratio in this case is problematic because one has to divide a small number by another small number. This creates tremendous instability in the resulting ratio. A few cases either way and the OR could change significantly in terms of magnitude and direction. This is especially concerning since all but one of the trials (DREAM) had un-adjudicated end-points.

Overall, the evidence that this meta-analysis provides is weak. What should we, as nephrologists, tell our patients? For those diabetic ESRD patients or CKD patients with GFRs that preclude the use of other antiglycemic agents contemplating using rosaglitazone or who are already on Avadia, my plan is to explain that the study has major flaws, that, so far, pioglitazone appears to have a favourable profile (but that the latter conclusion is subject to change) and that insulin therapy is probably the safest option, and then let the patients decide.

Here are a couple editorials which address this article that some might find interesting.

NEJM editorial
http://content.nejm.org/cgi/content/full/NEJMe078099

Lancet editorial
http://multimedia.thelancet.com/pdf/rosiglitazone_editorial.pdf