Pod Insights a16z Podcast Journal Club: Revisiting Eroom's Law About this episode With @jorgecondebio @vijaypande and @lr_bio
On this special holiday weekend edition of the a16z bio Journal Club we discuss a recent opinion article suggesting the end of Eroom’s Law—the decades long trend in biopharma R&D towards higher and higher costs per new drug—and our take on the big trends in this space.
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Speaker 00 Hello, I'm Lauren Richardson and this is the A16Z Journal Club, our podcast where we cover recent scientific advances, why they matter, and how to take them from proof of principle to practice. For this special holiday weekend edition, we've got something a little bit different. In this snack-sized episode, A16Z general partners Vijay Pandey and Jorge Conde join me to discuss an opinion piece published in Nature Review's Drug Discovery called "Breaking Eroom's Law." Eroom's Law is Moore's Law spelled backwards. Whereas Moore's Law describes technologies becoming exponentially faster and cheaper over time, Eroom's Law describes the trend of drug development becoming exponentially more and more expensive over time. In "Breaking Eroom's Law",
0:45 Speaker 00 the authors analyze the data since 2010 and find that costs appear to have stabilized over the last 10 years. But what has contributed to this critical and exciting trend shift? In our conversation, Jorge and Vijay discuss their views on the technologies and policies impacting this trend and their opinion on whether Eroom's Law is broken for good. We start by discussing the three possible factors for this shift cited in the "Breaking Eroom's Law" article, with the first being, quote, "better information."
1:15 Speaker 01 In my mind, there are sort of three vectors that are converging that I think become increasingly important when it comes to impacting productivity and therefore Eroom's Law. The first one is the fact that we can generate an increasingly broad array of data across the various ways in which biology transmits information. So we can do genomics, transcriptomics, we can do proteomics, all of these things at increasing scale. The second point is we can also do them at increasing resolution. We can go now increasingly down to the single-cell level, which, you know, in many cases is the sort of the functional unit of disease at a single-cell level. The fact that we can query biology at that level of resolution, I think, also makes a big difference.
1:57 Speaker 01 And thirdly, we can take all of these various streams of information and make better sense of them with the use of, among other things, advanced computational capabilities. It means that we not only have better information, we can generate better insights than we could before. And the hope, of course, is that with that, we'll be able to make better decisions.
2:20 Vijay Pande? If you want to escape E. Eroom's law and escape the flattening of E. Eroom's law to get to something more law-like, the natural way to do it is to do it by engendering a sense of engineering. And that probably will have so many different components. There'll be a huge biology and automation and reproducibility component. You can't engineer something if you can't have reproducible experiments. And so naturally, that's a huge part, and there's a lot of working on going on there. And then with all that automated biology, you've got to understand all that data that's coming in. And machine learning and AI is a very natural connector there.
2:56 Speaker 00 So what I'm hearing you say is that it's not just better information, it's not just more biological data, but it's also better use of this information. And, quote, "better use of information" is the second factor that the authors cite in breaking E. Eroom's law. How will AI and other methods of data interpretation aid in drug development costs?
3:17 Vijay Pande? The beauty of current machine learning and current AI is that they're being built to be interpretable, that not only can we get a sense of what the answer is, and hopefully in a way that has potentially be superior to human beings in many cases, but we can understand why that answer came about. And it's in that interpretability and in that understanding, I think, will be even greater advances to come. The other thing, too, is certain things inflate with higher inflation rates than the standard cost of living. Like the cost of buying socks at Target is inflating relatively slowly, but the cost of college, of lawyers, of drugs, is actually inflating much higher because it's a different labor pool. To the extent that more of this becomes automated, whether we're talking about automated
4:00 Vijay Pande? experiments or machine learning and better use of data, when you can automate more and more and more, maybe that also can help shift the curve because we're using sort of a high end skilled labor that inflates at a much higher rate, much less or in a different way. So in this process of putting all of it together, the end result will be really exciting. Hopefully it'll be a great decrease in the cost of drugs as well as an increase in our ability to understand biology, which itself has this sort of year over year advantage. That will hopefully be this great virtuous cycle that will snowball in all of our favors.
4:33 Speaker 00 Yeah, the combination of better information and better use of information is powerful. The third factor listed by the authors is changes to the approval threshold. So there had been this documented progressive lowering of risk tolerance of drug regulatory agencies like the FDA, which raised the bar in terms of demonstrating safety and efficacy for the introduction of new drugs. Do you two think that this threshold is changing now?
5:00 Vijay Pande? I think the pendulum is definitely swinging towards trying to get these drugs in the hands of patients.
5:05 Speaker 01 I think one of the things that helps the cautious regulator dynamic is the fact that increasingly novel drug programs are very targeted. They're targeted from a patient selection standpoint because you're going after a specific mutation or a specific genetic disease. But this whole rise of using biomarkers or other ways to select patients means that you can have a much more targeted development effort, which hopefully has two benefits to it. One is the safety side of things. If you have a better targeted therapy, you're picking the patients that are most likely to benefit from it, excluding the patients that are unlikely to benefit from it, which means from a cautious regulator problem, you're sort of minimizing risk to those
5:49 Speaker 01 that are least likely to benefit. And number two, on the efficacy side of the equation, to the extent that you're targeting the right thing, you're more likely to have an outsized impact. And so you're likely to see a stronger signal against noise, where a signal, of course, being efficacy and noise being the potential of things not working or worse, having some sort of toxic or detrimental effect on patients that wouldn't have benefited anyway. So you can hopefully have more thoughtful trial design as a result and move more quickly as well. The other factor I would say is I think that COVID will have the impact of accelerating the future in terms of modernizing how we think about running the appropriate studies for making sure that we can get therapies to patients safely and effectively, but also quickly.
6:35 Vijay Pande? Yeah, and I think we're already seeing advancements in the manner in which trials are done. So there's a huge push to virtual or distributed trials where instead of having to go to a single academic center, there could be doctors throughout the country administering the trial.
6:52 Speaker 01 The technology is emerging and increasingly being embraced that will help us rethink how we do clinical trials in the first place. So we can continuous monitoring of patients, you know, in a decentralized or at home setting that we can generate a lot more data points more quickly, more reliably for for patients. So I think technology outside of the lab itself, but going out into the community, so to speak, could have a real impact on what the all in cost to develop a drug looks like going forward.
7:25 Vijay Pande? This is kind of like a quiet revolution and it takes a while for those changes to be seen in something like a Evers law graph. But the the things that I think were started even like eight, 10 years ago are now very much coming to fruition.
7:39 Speaker 00 So my question is, do you really believe that Evers law is broken? Or is this leveling off that we've seen in the last 10 years, you know, a blip, temporary, or maybe the start of something resembling Moore's law?
7:52 Vijay Pande? It definitely looks like there is some major trend shift. Now, to say that we've shifted from sort of Evers law to something better is unclear. But even just flattening alone would be a major, major accomplishment for the industry. And so how long will this flattening last? I wouldn't be shocked if the flattening lasts for five years or 10 years, in large part because the things that we're talking about putting into drug design today won't lead to drugs to market for some time. And so there is a little bit of a lag here. These data points we're seeing is talking about the changes that were done five or 10 years ago. And so the changes that are happening now will only be seen later.
8:30 Speaker 01 The other thing I think it's worth mentioning is there's real debate in terms of whether or not the cost of R and D is calculated correctly. People debate that methodology. There have been alternative approaches that showed the number is actually much lower. I think people pick and choose the data depending on the argument that they're trying to make in many cases. And so if you're the pharmaceutical industry, you make the case that the cost of R and D is incredibly high to justify the pricing for those that are successful. If the costs are much lower, then obviously the prices that the system is forced to bear for therapies becomes less defensible. And so, you know, I think we should also keep in mind when we look at this paper that there are many, many ways to slice the data in terms of what costs really look like. And
9:14 Speaker 01 in turn, you know, I think it's, it's hard to definitively say in which direction the trends have moved and will move going forward. One thing that is clear is that developing drugs is time consuming, risky and expensive. That's probably not going to change anytime soon.
9:30 Speaker 00 Too true. Thank you both for joining me on Journal Club this week and for covering these big trends in the space.
Previous Preventing Pandemics with Genomic Epidemiology Next Read-Alouds, Continued Page updated October 26, 2026 · transcript and summary are machine-generated