More on the Avandia Debate: Common Sense vs. the Statisticians
I previously addressed the issue of “relative risk†in this blog, as it applies to the perceived hazard of using Avandia (rosiglitazone) to treat diabetics. To gain a better understanding of the true Avandia risk, I went back to the actual data submitted by Dr. Nissan et. al. in the meta-analysis which ignited this controversy. What I found supports my notion that the real risk is allowing statisticians to bludgeon common sense into immediate submission with a few technical terms.
In Nissan’s meta-analysis of 42 studies which compared Avandia to other diabetes treatments (“otherâ€), results from a total of 27,843 diabetics were analyzed (15,560 received Avandia and 12,283 “other†treatments). During the study period there were a total of 158 heart attacks (M.I.’s) and 58 deaths from cardiovascular causes. Compared to “other†treatments there were 14 extra M.I’s in the Avandia group then the “other†group. If your first reaction is “gee…14 extra deaths seem unacceptableâ€, remember there were 2300 more people in the Avandia group for bad things to happen to….see the blog on “relative risk†for more on that issue).
The over-all incidence of cardiovascular death for diabetics in the U.S. is generally accepted as 65% or more and the incidence of heart attack (M.I.) substantially higher. In Nissan’s study of 27,843 total diabetics 65% is equivalent to 17,730 total M.I.’s. The 14 extra M.I.’s in the Avandia group would make the M.I. rate 65.09%. Not a very alarming increase if it were true. In the “other†group if we equalize for the smaller number of participants in that group, we find the M.I. incidence would be higher at 65.12%. ( 72 M.I.’s for Avandia, 91 M.I.’s for “other†treatment).
Another way to look at the magnitude of the supposed increase in Avandia related events,
we find the “excess†number of M.I.’s is equivalent to 1 per 1250. For practitioners who treat diabetes and understand the enormous degree of variation between diabetic patients, trying to pin-point the factors accounting for one M.I. per 1250 in this group would be like trying to isolate one snowflake in a blizzard.
I simply do not believe that there is a way to validate the results of Nissan’s study. Believing the use of statistics can correctly pin-point the cause of 1 in 1250 M.I.’s within the chaos that is diabetes care, in my opinion, is being naïve to the true complexity of this disease and its treatment.
Gary Pepper, M.D.