One of the most intractable problems in drug development has been productivity.
Some feel that productivity has decreased substantially while other believe that it has merely remained obstinately the same. What is clear is that investment and costs have been rising, and the output has at best remained constant.
I’ve had a lot of conversations about this with people and have read just about every piece of published literature out there on this topic. One relatively common theme is rising regulatory hurdles for approval. I think this might be one of the factors, but I’m not convinced it’s the main reason.
Now, it is very difficult to discern what factor plays what role in productivity, since there are few controls to compare against. So for the most part, the debates are based on opinions.
I think one way to try to tease out the effect of regulatory burden is to look at other industries and compare. Our closest sister industry is actually the pesticide industry. They work on similar chemicals, their regulatory hurdles have been rising very quickly (arguably more quickly than ours), and they turned to similar combichem and high throughput screening at around the same time that we did. If regulatory bar is the main driver for productivity, then we might expect their productivity to have dropped faster than ours. How did they fare?
Well, if you talk with people in that industry, you hear similar complaints: that costs are going up, fewer molecules are being developed, etc. There is not much objective data, but what there is tells a story. Kind of.
Yes, their costs have been going up. But at a far slower pace than drug development costs. These costs don’t include failures. (These charts are from Phillips McDougall report, The Cost of New Agrochemical Product Discovery, Development and Registration in 1995, 2000 and 2005-8).
But take a look at the table below. This is the average number of molecule at each stage of development per company. The data points are sparse, but it appears that the failure rate has not increased, at least as much as in drug development. There are fewer molecules in development, but the output is what counts.
One way to interpret data could be that, unlike us, they may be getting better at predicting success before they put a molecule into development. Put another way, if drug industry could get 1 drug approved for every 1.3 IND filed, well – that is the Holy Grail isn’t it? Some people think the goal of development is to have a funnel, something like 100 Phase I candidates, 50 Phase II, 20 Phase III, and 15 approvals. 100 into the pipeline, and 15 out. I would be think the goal should be 15 in and 15 out.
If this dataset is correct, it means that the actual cost per approved molecule cost may have increased even slower than suggested by the above graph. The pesticide industry may not be doing great with regard to productivity, but probably better than the drug industry.
I emphasize, the dataset is very small, so we need to temper our conclusions (I think it’s really just hypothesis-generating), but this is the best data I’ve found. I’ve only found one other set of data, also sparse, but it also shows similar weak trend.
If it’s not just the stricter regulation that is driving down productivity, then what? Some people believe that the easy targets have all been harvested, and that the remaining targets are just more difficult. Yet others blame the mergers and growing size of the companies. It’s probably true that innovation doesn’t typically come from large companies. I was talking to someone who worked in the soft drinks industry who told me that the innovation even in developing new soft drinks is difficult in a large company.
What is clear that combichem, rational drug design, genomics, high throughput screening– all those things that were supposed to have improved the rate has not had much of an impact. Or at least any impact that is demonstrable. Some might argue that in absence of those fancy techniques productivity would have been even lower than it has been–that these techniques have helped avoid even worse decline in productivity.
Others however, argue that the old methods were better than how drug discovery and development is being done now. When we jumped into HTS and combichem, and target based rational drug deign, there was no data showing that HTS or genomics would improve drug development. It just sounded so…compelling. So people jumped into it with both feet, and here we are today.
So has drug development improved? We don’t actually know. One factor that makes it difficult to distinguish good from bad drug development is that drug development is a low-yield endeavor. With the success rate hovering between 5 and 10%, even the best drug developers will fail most of the time. It is not easy to tell between those who are good vs. those who are lucky.
It’s tricky to identify skilled drug developers. The percentage of correct predictions on drug development decision is not the right way to judge competence. Since the failure rate is 90%, someone who predicts that every drug will fail will be right 90% of the time. However, that person is of much less value than someone who predicts success 30% of the time and is right 10% of the time.
My personal theory is that drug development hasn’t gotten any better or worse. Admittedly, that’s not a terribly bold insight. But here is the second part of the theory: I would posit that we have not reason to expect that the rate would have improved over time, because we don’t have an effective feedback loop in drug development.
Most people learn from making mistakes. You perform an action, see the results, and make adjustments from success or failure of those actions. In drug development, this is very difficult. The timelines are very long, on the scale of decades. It is very difficult for a chemist at the beginning of the drug development process to take into account what happens years later and incorporate that into what he or she does. In addition, there are multiple factors that can affect the success of a drug development candidate, and it is not easy to disentangle them.
More importantly, we don’t have a culture of learning in drug development. If a drug fails in Phase II or III, very little effort is spent trying to understand and learn from the experience. I don’t mean in instances where a drug fails and the team believes that the drug still has a chance to become a success if they tried again, perhaps with a different indication or dose. I mean in instances where the drug fails and it is clear that there is no more hope for it.
In terms of improving the drug development process and productivity, value of a failed drug is just as great as a successful one. Yet we spend much of our time studying successful drugs and trying to emulate them, and not much time trying to understand the differences between a drug that looked very promising going into Phase II or Phase III and failed vs. one that looked equally promising and failed.
You cannot learn by looking only at successes.
As an example, let’s try to figure out how to become rich by studying rich people. What are the similarities between Bill Gates, Warren Buffett, Nelson Rockefeller, and Jay Gould? Here are some of the things common to 100% of rich people:
Not very helpful, but as Kahnemann pointed out, we have a cognitive bias toward thinking like this.
If we want to improve drug development, I believe the most helpful drugs to study are ones that failed that we thought were going to succeed, and ones that were dropped at some point that turned out to be successes.
In addition, we have to disentangle force magnifiers from directional determinants. There are certain factors that magnify the effects of certain actions. Force multipliers. For example, if you look at the very successful people, you might find them to be very determined and not easily swayed by other people’s opinions. However, you might find the very same personality traits in the most unsuccessful people. Being strong-willed may in and of itself may not make you successful or not – it may just magnify the effects of intelligence, hard work, and other factors.
In some studies, as another example, it has been shown that some of the biggest disasters in corporate history comes from a single individual or a small group of individuals who manages to push through a project over objections of almost everyone else, and manages to get the company to commit enormous amounts of resources to something that turns out to be a very bad idea. When I was at one of my previous companies, an article to this effect was widely circulated.
Of course, some of the greatest successes have also come from a small group of committed individuals pushing an idea ridiculed and opposed by the vast majority of people.
In this fashion, we have to separate the most successful drugs in terms of sales from successful drugs that turned out tobe safe and effective. The size of the market doesn’t impact learning for productivity.
One of the issues if we want to study how to improve drug development of course is that the sample size is low. It is hard to draw lessons from only a few thousand drug candidates. However, the sample size can be enlarged to look at the business development decisions. Each drug that is developed undergoes evaluation aw potential BD deals dozens of times. There are confounding factors of course, but this might be one way of enlarge the sample size.
So, I think we need to really look at how we learn from our mistakes in drug development and try to integrate those learnings into our practices if we want to have any hope of improving the success rate.