Avandia: Lies, Damned Lies…and Statistics

There are no whole truths; all truths are half-truths. It is trying to treat them as whole truths that plays the devil."b       "Truth is rarely pure, and never simple."c       The news is again a-twitter with stories of another fallen hero: Avandia. While I have no need for the drug, it seems to echo the Vioxx and Celebrex stories of the recent past. I have arthritis, so those stories DID affect me. Having had to switch medications in mid-stream, as it were, I decided to take a closer look at the numbers being used. [Any of my readers who know my background know I have some interest in statistics and Chemometrics from my work in near-infrared (NIR) spectroscopy.] As with any news story (e.g., "Man bites dog!"), seldom do the pundits read beyond the first paragraph. The headline on the Today Show this morning was "Avandia causes 30% increase in heart attacks." Let's look at that first by going to the FDA website where the information was posted "for medical professionals," as per the news. [Under the law, ALL citizens have access to the data, so don't be put off.] It seems that "8,604 patients on rosiglitazone (Avandia) [too bad "litazone" sounds so much like litigate] were compared with 5,633 patients randomized to a variety of alternative therapeutic regimens, including placebo." The population seemed well dispersed among various lengths of disease occurrence, history of other problems and surgeries, etc. All but four of the 42 studies were of six months duration. The overall incidence of myocardial ischemia in the Avendia versus comparators was 1.99% versus 1.51%. This was seen as a 30% excess risk factor. Now, I seemed to remember something similar in the 55 mile-per-hour national speed limit controversy, so I went back and looked at some notes. When Maine switched from 55 mph to 65 mph, the death rate doubled! That's right, it doubled! It went from one to two for the year! Hence the title of this piece! So, let's read farther"¦often newspeople and lawyers [there are already"¦same day as newscast"¦lawyers posting for patients taking Avandia] hope we have short attention spans and will not read further (it is, after all four large pages of print with no picture). A Balanced Cohort Study used 33,363 patients in a managed care database for its study. "Propensity matching" was used to match risk factors (how novel) for cardiovascular disease and other considerations. The "composite cardiovascular endpoint" was hospitalization for heart attacks (myocardial infarctions) and bypasses (revascularizations). Patients in the study began use of the (dreaded) drug in question between 2002 and 2004. The groups were only rosiglitazone, metformin, or sulfonylurea; oral dual therapy combinations, and insulin combinations. The result?  With a follow-up time of 1.2 years, the incidence of the composite cardiovascular endpoint was 1.75 events per 100 patient years for the rosiglitazone (Avandia)-containing regimen and 1.76 events per 100 patient years for the other treatments. Now, this is NOT a headline-grabber. Further studies also do not support the first stated study. This could be why the FDA voted 22-1 to continue the sales of the product and to allow doctors and patients to decide whether the risk is acceptable.  This is all very interesting you say, but what does it have to do with process analytical technologies and Quality by Design (PAT/QbD)? Well, it seems the study that took ALL the factors into consideration was approaching Chemometrics"¦or, as we say, "Design of Experiment." When you use a portion of your data to support a conclusion (e.g., there is a better than 0.9 correlation between the rise and fall of the Mississippi River and the birth rate in Paris, France), you take a chance of using PAT to make your product worse. When we attempt our first (or tenth for that matter) PAT project, it is critical to have enough inputs and use correct math to glean the causes of "good" v. "bad" product. If we merely look at, say, hardness versus dissolution rate, we totally ignore the distribution of lubricant or particle size of API or moisture distribution or etc., etc. etc. Vis á vis the topic of this column, I could point out that 100% of the people who go bald drank water prior to baldness and anyone who breathes air will die. These are true, but what conclusions are you drawing from these statements? The discipline needed for good science is not limited to the lab or the clinic or the production line. Good science and engineering are prerequisites at all stages of health care. But, since I have a limited charter, let it suffice to say, use all information when building a better product. May the force (a.k.a. Chemometrics) be with you. References and Quoted Dignitaries
  1. Benjamin Disraeli, 1804-81
  2. A. N. Whitehead, 1861-1947, Dialogues, Prologue
  3. Oscar Wilde, 1854-1900, The Importance of Being Earnest, I
By Emil Ciurczak, NIR/PAT expert and Contributing Editor, PharmaManufacturing.com.  Comments, as always, invited.