 |
How many go to Las Vegas or Atlantic City and play the roulette wheel? While it may be fun to go once or twice, I doubt that any of us would resign our faculty positions at Johns Hopkins to make a career playing roulette. Why? Well for one thing, the payoff is too low to compensate for the risk involved. The expected value of playing roulette is negative, not positive—over time, it costs more than it returns.
Now consider for a moment that the success rate for drug development is about 1 in 200. Not great odds, unless the payoff is considerably greater than 200 to 1, right? In general, the payoff for drug development thus far has more than compensated pharmaceutical companies for their investments. Okay, but following the mergers of smaller pharmaceutical companies into bigger ones, consider that the emphasis has shifted to the development of so-called blockbuster drugs—ones that bring in more than $1 billion in annual sales. Such drugs include statins, antihypertensive medications, analgesics and antidepressants—drugs that not only have a large market, but also that are taken chronically. Now the rewards are much higher—but so are the stakes. The odds of success go down way below 1 in 200, while the payoff increases far beyond 200 to 1 as well. But by emphasizing blockbuster drugs, the big pharmaceuticals set themselves up to fail in two ways.
The first is related to the so-called St. Petersburg paradox first posed in 1713 by Swiss mathematician Nicholas Bernoulli. The paradox considers games of chance with a positive expected value in which the probability of winning is low but the payoff is huge. Each play produces a small loss, but in an extraordinarily rare event, the payoff is infinitesimally large. Even though the expected payoff for this game is ultimately positive, most gamblers will go bankrupt before they hit the jackpot because they don't have enough cash reserves to stay solvent. For example, consider a game of slots where the probability of a jackpot is 1 in a million, but if you hit the jackpot you win $2 million. For each pull of the slot machine, if you don't hit the jackpot you lose $1. If you have very deep pockets, your average expected winnings will be approximately $1 million ($2 million minus $999,999). But in order to make $1 million, you need to have a large sum of money to avoid going bankrupt before hitting the jackpot.
Similar forces are at work with blockbuster drugs. The payoff is huge, but the losses incurred to get there, though less than the payoff, are quite large—some companies may not have the financial stamina to stay in the game. Blockbuster drug discovery, just like blockbuster Hollywood movie production, is nearly impossible to sustain. Studies of Hollywood movie studios support this assertion. I don't know of any similar ones in the pharmaceutical industry, but I suspect it is true there as well.
Vioxx, on the other hand, was already a successful blockbuster drug, so Merck—the drug's manufacturer—can be said to have beat the St. Petersburg paradox. But Merck stumbled into another pitfall that blockbuster drugs engender—fooled by randomness. If you develop a drug that is going to be used chronically on large populations, you are going to experience measurable numbers of adverse outcomes (irrespective of the safety of the drug) as the size of the population on which the drug is used increases. The more patients using the drug, the more likely adverse events will appear. Moreover, while these adverse events may occur as random fluctuations, they will appear to many to have a causal correlation with the use of the drug. If you do enough studies on a large enough sample, you can expect to find serendipitous correlations that may, or may not, be causally related.
The difficulty is in proving the null hypothesis—namely, that the correlation between the drug and an adverse outcome, like myocardial infarction, is not causally linked. Just as we have difficulty establishing the safety of hormone replacement therapy in postmenopausal women, it is hard with a drug like Vioxx to determine causality within samples of relatively heterogeneous populations. Controlled randomized prospective trials with millions of patients are simply not feasible, so we are faced with having to make deductions from populations of (from a statistician's point of view) imperfect composition.
He who lives by the sword, dies by the sword. An unfortunate aphorism perhaps, yet probably more true for Merck than they would care to recognize. In order to promote Vioxx to blockbuster status, Merck used direct-to-consumer advertising to boost sales. Increasing usage, however, increased the likelihood that adverse effects would appear that might be correlated with use of the drug. Even if there is absolutely no causal relationship whatsoever, there can be the appearance of one, especially when presented by a sophisticated legal team to a jury of nonstatisticians. Fooled by randomness.
How many go to Las Vegas or Atlantic City and play the roulette wheel? While it may be fun to go once or twice, I doubt that any of us would resign our faculty positions at Johns Hopkins to make a career playing roulette. Why? Well for one thing, the payoff is too low to compensate for the risk involved. The expected value of playing roulette is negative, not positive—over time, it costs more than it returns.
Now consider for a moment that the success rate for drug development is about 1 in 200. Not great odds, unless the payoff is considerably greater than 200 to 1, right? In general, the payoff for drug development thus far has more than compensated pharmaceutical companies for their investments. Okay, but following the mergers of smaller pharmaceutical companies into bigger ones, consider that the emphasis has shifted to the development of so-called blockbuster drugs—ones that bring in more than $1 billion in annual sales. Such drugs include statins, antihypertensive medications, analgesics and antidepressants—drugs that not only have a large market, but also that are taken chronically. Now the rewards are much higher—but so are the stakes. The odds of success go down way below 1 in 200, while the payoff increases far beyond 200 to 1 as well. But by emphasizing blockbuster drugs, the big pharmaceuticals set themselves up to fail in two ways.
The first is related to the so-called St. Petersburg paradox first posed in 1713 by Swiss mathematician Nicholas Bernoulli. The paradox considers games of chance with a positive expected value in which the probability of winning is low but the payoff is huge. Each play produces a small loss, but in an extraordinarily rare event, the payoff is infinitesimally large. Even though the expected payoff for this game is ultimately positive, most gamblers will go bankrupt before they hit the jackpot because they don't have enough cash reserves to stay solvent. For example, consider a game of slots where the probability of a jackpot is 1 in a million, but if you hit the jackpot you win $2 million. For each pull of the slot machine, if you don't hit the jackpot you lose $1. If you have very deep pockets, your average expected winnings will be approximately $1 million ($2 million minus $999,999). But in order to make $1 million, you need to have a large sum of money to avoid going bankrupt before hitting the jackpot.
Similar forces are at work with blockbuster drugs. The payoff is huge, but the losses incurred to get there, though less than the payoff, are quite large—some companies may not have the financial stamina to stay in the game. Blockbuster drug discovery, just like blockbuster Hollywood movie production, is nearly impossible to sustain. Studies of Hollywood movie studios support this assertion. I don't know of any similar ones in the pharmaceutical industry, but I suspect it is true there as well.
Vioxx, on the other hand, was already a successful blockbuster drug, so Merck—the drug's manufacturer—can be said to have beat the St. Petersburg paradox. But Merck stumbled into another pitfall that blockbuster drugs engender—fooled by randomness. If you develop a drug that is going to be used chronically on large populations, you are going to experience measurable numbers of adverse outcomes (irrespective of the safety of the drug) as the size of the population on which the drug is used increases. The more patients using the drug, the more likely adverse events will appear. Moreover, while these adverse events may occur as random fluctuations, they will appear to many to have a causal correlation with the use of the drug. If you do enough studies on a large enough sample, you can expect to find serendipitous correlations that may, or may not, be causally related.
The difficulty is in proving the null hypothesis—namely, that the correlation between the drug and an adverse outcome, like myocardial infarction, is not causally linked. Just as we have difficulty establishing the safety of hormone replacement therapy in postmenopausal women, it is hard with a drug like Vioxx to determine causality within samples of relatively heterogeneous populations. Controlled randomized prospective trials with millions of patients are simply not feasible, so we are faced with having to make deductions from populations of (from a statistician's point of view) imperfect composition.
He who lives by the sword, dies by the sword. An unfortunate aphorism perhaps, yet probably more true for Merck than they would care to recognize. In order to promote Vioxx to blockbuster status, Merck used direct-to-consumer advertising to boost sales. Increasing usage, however, increased the likelihood that adverse effects would appear that might be correlated with use of the drug. Even if there is absolutely no causal relationship whatsoever, there can be the appearance of one, especially when presented by a sophisticated legal team to a jury of nonstatisticians. Fooled by randomness.
