Here’s an updated look at clinical success rates in the biotech industry, from the BIO industry organization. They’re looking at only company-sponsored programs aimed at FDA registration, nothing investigator-sponsored, and it’s a big data set:
This is the largest study of clinical drug development success rates to date. Over the last decade, 2006-2015, a total of 9,985 clinical and regulatory phase transitions were recorded and analyzed from 7,455 development programs, across 1,103 companies in the Biomedtracker database.
So how is everyone doing in the clinic? Sheesh, you had to ask. The overall Likelihood of Approval” (LOA) from Phase I is 9.6 per cent. So no, if you were hoping for an upturn in the more recent numbers, it doesn’t seem to be there. Somewhere around ten per cent success has been the estimate for some time now, and here it is again. The worst therapeutic area is oncology, with a 5.6% LOA. That reflects some of its traditional features: lower barrier to entry, which brings in a lot of marginal ideas, difficult diseases to target once you’re in the clinic, and a corresponding willingness to try some fairly out-there stuff.
These numbers shouldn’t be too surprising to folks in the industry itself. What I’ve found while writing this blog, though, is that people who don’t know drug discovery tend to be very surprised indeed. It bears repeating: ninety per cent of all drugs that go into the clinic fail. And ninety-five per cent of all cancer drugs. That’s the central problem of the whole industry, right there. If you can make those numbers look better, hundreds of millions of grateful patients (and untold billions of dollars in revenue) are waiting for you.
That means that each step of the process (Phase I, Phase II, Phase III and regulatory approval) is slightly better than a 50/50 shot on average, slightly worse in oncology. (A flat 50% attrition rate in each would give you an overall 6.25% LOA). But the four stages are not the same. Approval rates have been running at about 85% during this period (with some resubmissions, mind you), but Phase II is still the killer, as it has always been: there’s a 30% success rate across that stage of the process. If you can get through Phase II, the overall LOA for the industry during the past ten years is just about 50% – a coin flip, but better than what you’re facing before Phase II itself!) All this, of course, reflects that Phase II is generally the first time that your new idea actually contacts human beings with the disease you’re targeting, and as you can see, T. S. Eliot was right on target: we cannot stand very much reality. (As an aside, if anyone from BIO is reading this, you need to fix the title of the chart that illustrates this point – right now, it says “Probablity of Success”, which I assume isn’t quite what you were aiming at, and four more charts in the report are labeled the same way).
Broken down by disease, by far the best area was Hematology (anemia, hemophilia, etc.), with an overall LOA of 26.1%. (That was the only therapeutic area with a better than 50% chance of making it through Phase II, by the way). Infectious disease was second overall at 19.1%. As mentioned, oncology was worst, but (as many will have guessed), CNS/psychiatric indications is right down there too, at only 6.2% LOA. Those two areas also had the highest number of clinical programs, so you can see where the overall success rate is getting hammered down. (If you exclude oncology, the overall LOA is about 12%). Interestingly, the next worse was cardiovascular, at 6.6%. The last big survey of this sort of data had that field ranking a lot higher, and I wonder if we’re seeing the vanishing of the low-hanging fruit in these numbers. Oncology, by the way, is the only therapeutic area with a Phase III success rate of under 50%.
So you can argue, reasonably, that oncology (which accounted for 31% of all the clinical programs in the data set) has an outsize influence on the total numbers. But even without cancer programs, the view is unappealing: instead of an average of one out of every ten drugs making it through, it’s one out of eight, which is still a hard way to make a living. If you look inside the oncology data, you see a clear split between hematologic cancer and solid tumors – the latter are, of course, much harder to work with, and pancreatic cancer is (no surprise) at the very bottom of the list.
The report also breaks out the rare-disease programs by concentrating on all the FDA orphan disease designations (and taking out oncology) in order to look at inborn genetic indications. The numbers bear out the general impression of greater success in these areas: 25% overall LOA, which is more like it. Success rates were the same or higher in every phase, with an especially big difference in Phase II (50% chance of success in this area). Another interesting breakout was with the programs that indicate that they’re using some sort of biomarker as a selection criterion for the clinic. In these cases, success rates were also higher at every stage, with an overall LOA of 26%. (It’s worth noting that this report was <s>partly sponsored</s> by a biomarker company, Amplion, but the numbers do speak for themselves). Update: I should note that Amplion put up no money, just data and data analysis).
The bind that the industry is in is also well illustrated here. If you want to go after a high prevalence, heterogeneous, chronic disease – that is, one where a new therapy would have the biggest impact – then you’re looking at the absolute worst odds and the most expensive development path possible. Think. . .Alzheimer’s. On the other hand, if you want to go after a rare orphan indication with a genetically homogeneous patient population, your chances of success are definitely higher – not a slam dunk, because we don’t have those, but still cranked all the way up to only a 75% chance of failure as opposed to 90%. But the patient population is so small that the cost of the drug will have to be expensive for the effort to make financial sense at all. You spends your money and you takes your choice.