A recent paper on drug development costs did not impress me. But if possible, it impressed Matthew Herper at Forbes even less. That’s the one where the authors looked at a number of companies that had been around long enough to develop one drug – they figured that this would give a cleaner read on what that one drug cost, as opposed to trying to work it out from a larger or more well-established company’s budget, and the figure that they came up with was $648 million.
Herper takes issue with this in the same way that I did:
Prasad and Mailankody assert this analysis takes into account the high attrition rates of drug development because each company was developing between 2 and 11 experimental medicines, only one of which reached the market. But this assumes that the companies were developing a large enough number of medicines to capture the high failure rate of drug development. Given that 9 in 10 medicines fail, it seems unlikely that looking at companies that had made 4.3 attempts at creating a drug, on average, would capture this. Conceptually, this is no different from simply looking at companies that had only tried to develop a single drug and happened to succeed. Researchers call this “survivorship bias” – it’s like estimating an average lifespan by asking people their ages, but not finding out if anyone already died.
He has the data to back this up, too. If it really costs about $648 million to develop a drug, then you would figure that as a company gets large enough to have several projects running, those costs should converge more and more on a figure in that range, as you average out the slightly-cheaper and slightly-more-expensive ones. Heck, it should probably converge on something even lower, because a larger company would certainly have some economies of scale in its development costs that a smaller one-drug company has had to pay full whack for.
But that’s absolutely not what happens. As Herper shows, the more drugs a company has developed, the higher the average cost per drug. That, my friends, is because the cost for the one-drug-only companies is not representative, for just the reasons mentioned above. When you look at what companies spend to keep on developing drugs, year after year, the true costs become apparent, and the number is not pretty. It comes out to a bit over $2 billion per, these days.
And “these days” is an important qualifier, because these costs have not been static. One of the things you come away with from studying this issue is that the cost-per-drug has truly been increasing, and that the (relatively steady) productivity of the drug industry as a whole is due to lots more cash being frantically bulldozed into the furnaces behind the scenes.
The real problem is that the amount spent to develop every new drug seems to be increasing. Prasad and Mailankody present the $2.7 billion Tufts estimate as being in contrast with an estimate of $320 million produced by the group Public Citizen. But the Public Citizen estimate, which is also based on R&D budgets divided by the number of drugs approved, is based on drugs approved in the 1990s. And the most obvious explanation for the difference is that the R&D dollars spent per drug have increased ten-fold in two decades. This is exactly the observation that Jack Scannell and his co-authors noted in Nature Reviews Drug Discovery in 2012, calling it “Eroom’s Law” – the reverse of the tech industry’s Moore’s Law, which predicts that transistors become exponentially cheaper over time.
Exactly. I would like to lay down a challenge for future authors who want to show that “come on, drug development’s really not all that expensive, right guys?” Unless they can show that they have read, understood, and refuted the best writing and evidence on this subject, with detailed reasons why the previous analyses are wrong, they are not to be taken seriously. Everyone who comes after this topic mentions the Tufts figures (which are damned close to what Herper arrives at, by the way), but generally to brush them off with a “ah, that can’t be right” and an insinuation that they must be a bunch of industry shills. But that won’t cut it. If you want to show that drugs are actually a lot cheaper to find, then you not only need to engage with the Tufts numbers, in detail, but you need to also deal with the writings of Jack Scannell, Bernard Munos, and others, with those others very much including Matt Herper. Show just why they’re wrong. Don’t just wave your hands at them in annoyance and say that these numbers are obviously inflated or something – show why they’re wrong. Because I don’t think that they are.