How the FDA forgot the evidence: the case of donepezil 23 mg.

What is the difference between 20 and 23? If you said three, you are off by millions—of dollars in sales, that is—at least from the perspective of Eisai, the manufacturer of donepezil (marketed as Aricept by Pfizer). A little context helps make the maths clearer. Donepezil, the biggest player in the lucrative market for Alzheimer’s disease treatments, was a blockbuster, with over $2bn in annual sales in the United States alone. But the drug, first approved in 1996, had reached the end of the road: the patent expired in November 2010. Investors call this “going over the cliff,” an anxious reference to plummeting sales as market share is lost to generic competitors. Necessity, however, is the mother of invention. Just four months before the expiry of the patent, the US Food and Drug Administration (FDA) approved a new dose for moderate to severe Alzheimer’s disease: donepezil 23 mg. Is 23 an odd number? Not really, when you consider that you cannot get to 23 mg using the 5 mg and 10 mg doses that were going generic. The “new” 23 mg product would be patent protected for three more years. Now it was time for the marketing to begin. In addition to their sales force, the manufacturers deployed dedicated teams of “Aricept 23 mg clinical nurse educators” to reach prescribers. They focused particularly on “priority targets”—neurologists and high volume facilities for the long term care of people with Alzheimer’s disease—to promote the idea that “there are no ‘stable’ AD [Alzheimer’s disease] patients—therefore aggressive treatment is required.” But the marketers had to overcome some inconvenient facts. The drug had been approved only over the objections of the FDA’s medical and statistical reviewers. 3 In fact, approval breached the FDA’s own regulatory standard. In study planning meetings, the director of the FDA’s Division of Neurology Products (the division that oversaw drug approval) told Eisai that the regulatory standard for approval of a drug for Alzheimer’s disease required superiority on both a cognitive and a global measure. The rationale for the global measure is to ensure that any statistically significant cognitive difference (for example, performance of tasks such as the correct repetition of a series of three, four, or five digits) represents a real clinical benefit (that is, an improvement that patients or care givers notice). The FDA and Eisai explicitly agreed that the 23 mg dose would be approved only if the drug were shown to be superior to the 10 mg dose on both measures. But it didn’t. In the approval trial of more than 1400 patients, there was a small statistically significant improvement in cognition (a 2.2 improvement over the lower dose on a 100 point scale) but no statistically significant difference in global functioning (a 0.06 improvement on a seven point scale). Further, the higher dose was not superior on either of the pre-specified secondary outcome measures, which, as the FDA medical reviewer pointed out, argues that the cognitive difference was not meaningful. (See table.⇓) The other problemwas side effects of the 23 mg dose, including substantially more nausea and vomiting. Russell Katz, director of the FDA’s neurology products division (who ultimately approved the drug), said that “there is a clear increase in the incidence of adverse events on the 23 mg dose compared to the 10 mg dose . . . These are not trivial events; in these patients, these could lead to significant morbidities and even increased mortality.” And yet the drug was approved. Why? The same FDA division director (who signed the correspondence stating the regulatory standard for approval, of superiority on both a cognitive and a global measure) said, “In my view, this strongly argues for a conclusion that the 23 mg dose is very likely to also have an effect on overall functioning, despite this not having been