Show me the value! Increasingly, that's what everyone—from Medicare and private insurers to doctors and patients—is demanding
before showing pharma the money for its drugs. What with the jump in generics, the fight for formulary placement, and ever-savvier
consumers, companies are coming to a painful realization: The evidence that satisfies FDA at drug-approval time may not satisfy
the marketplace. Being better than a placebo is necessary but increasingly not sufficient. More and more, drugs both new and
old are having to justify their price tags by proving that they're stronger, more effective, safer, or easier to take than
Bryan R. Luce
The congressional bill calling on the government to negotiate drug pricing for Medicare Part D—passed by House Democrats as
Pharmaceutical Executive went to press—will only accelerate the demand for this information.
Prove It? Not So Fast
The industry's resistance to head-to-head research is easy to understand: The studies are very expensive, time-consuming,
and, above all, risky. As Bristol-Myers Squibb famously learned from its PROVE-IT trial in 2003, real-world comparative data,
even when designed to maximize odds for success, can turn around and bite the sponsor.
BMS pitted Pfizer's potent statin, Lipitor, against a lower dose of its own Pravachol in 4,000-plus patients previously hospitalized
for angina or heart attack. That matchup may appear a no-brainer, but BMS had designed the study to minimize risk—or so it
hoped. First, PROVE-IT was a "noninferiority trial," so Pravachol had only to hold its own against Lipitor. Second, follow-up
ended at two years—the point at which differences among anti-cholesterol drugs were believed to be just approaching measurability.
But BMS got an unpleasant surprise when data after only 30 days showed that Lipitor patients had a 16-percent-lower risk of
heart attack than the Pravachol takers.
PROVE-IT, of course, reinforced industry's wariness of post-marketing comparative evidence. But a mere three years later,
the questions surrounding drug-versus-drug studies no longer turn on if but how—and how soon. The challenge for forward-looking drug manufacturers is to find a risk-limiting framework that renders such research faster,
easier, and cheaper—and then to be the first to market the information.
Adaptive Trials, Bayesian Model
An adaptive trial design, particularly if it uses Bayesian statistical methods, can go a long way toward overcoming the key
hurdles of comparative late-stage research. Donald Berry, a leading Bayesian advocate, argues that such Phase IIIb or IV trials
are 30 to 50 percent more efficient than the traditional blinded, placebo-controlled version.
Bayesian studies are generally smaller, swifter, and more focused than the large trials employing classical frequentist statistics.
For instance, they can be modified in midstream to capitalize on new data as they come in, allowing companies to optimize
time and sample size while limiting costs and risks. If an early analysis indicates that the drug appears to be particularly
effective in, say, elderly women, the randomization scheme can be rebalanced to recruit a higher percentage of these participants,
producing more meaningful data while improving patient outcome.
Bayesian trials also build off of all the relevant evidence known about a drug. Instead of the classical approach of starting
a new study from square zero—as if many millions of dollars haven't already been spent to develop data about the compound's
effects—a Bayesian method answers questions more efficiently. It says, in effect, "We know the drug is safe and effective
in a controlled setting, but this is not quite enough evidence for an informed decision about formulary placement. Now we
need to find out how well it performs in community practice."
Finally, Bayesian research yields information in a form ideally suited to healthcare decision-makers, who must adapt findings
to real-world situations. Rather than the statistical significance of traditional trials' output, Bayesian results are expressed
in terms of probability, such as:"Drug A is 70 percent more likely to improve health status than drug B."