Drug developers navigating ‘biology’s century’ need a model. We made one

Brian Finrow
10 min readMar 20


By Brian Finrow, Aleks Engel and Srinivas Akkaraju This piece was originally published at STAT NEWS on March 15th, 2023

STAT’s Matthew Herper struck an industry nerve when he wrote that the drug development industry is “not prepared for the next wave of biotech innovation.”

His report ends on a somewhat dismal note: the industry is long on diagnosis (clinical development is too expensive) and short on cures.

For an industry based on science and evidence, a good way to understand a complex situation like this is to test different scenarios in a simple model. Building on prior work by others in the field, we created such a model to help inform the much-needed debate that Herper’s article kickstarted.

To be sure, “All models are wrong,” as George Box once mischievously said, “but some are useful.” We believe this is one of the useful ones, and we invite readers to explore its implications with us.

Recent events and biopharma productivity

The productivity of new-drug research and development is poor and getting worse. Many others have noticed this. It’s an important, possibly existential, issue for the biopharma industry. A recent report from Deloitte, which is included with related analyses in the chart below, updates this long-term trend. The return on investment for new drug development fell to a shockingly low 1.2% in 2022.

Three independent estimates of the secular decline of biopharma R&D productivity (measured in dollar returns on investment). Adapted from a LinkedIn post by Kelvin Stott in 2017. ADAPTED FROM A LINKEDIN POST BY KELVIN STOTT IN 2017.

Recent events have inflamed this problem in two ways. First, because it takes more than a decade to develop a new drug, investment decisions are highly sensitive to the cost of capital. Rising interest rates put immense pressure on biopharma business models — especially those built in the recent era of cheap and abundant capital. Second, the market downturn has drained financial liquidity due to what Atlas Ventures’ Bruce Booth called a “risk-off” environment. Many biotechs are unable to raise funds at any price.

In the near term, the few companies that are able to transition to more efficient drug development cost structures will be in higher demand. Those that cannot will struggle through more layoffs and shutdowns.

In the past, most biotech investors and big biopharma companies have simply accepted rising costs and declining R&D productivity as immutable. Instead of fixing the underlying productivity problem, most have tried to solve it by investing only in the companies with the best odds of success or by focusing investment on multi-billion-dollar blockbuster categories like oncology.

But it’s hard to feel confident in a picking-the-winners strategy given drug development’s stubbornly high failure rates. And while the blockbuster approach worked for Merck with Keytruda, that happened famously more by accident than by design.

According to the Deloitte report, the average projected peak sales of a new drug launched in 2022 was actually just $389 million — far below blockbuster status. Worse, a majority of new drug launches fail to hit even these low targets. Clearly, the old strategies are no longer working.

Industry critics argue that cynical biopharma companies simply use price increases to fill the financial gap. If so, this “strategy” seems to have hit its political limits, as evidenced by the price negotiation provisions of the Inflation Reduction Act. It also abandons people living with diseases that lack blockbuster potential, especially those that are more common in low-income countries.

The best available tool short of a crystal ball: enhancing the Minikel model

It’s time to confront the R&D productivity problem, and model-based analysis can help. The drug development process has too many complex and interrelated components to visualize in one’s head. What’s needed is a set of actionable insights that will allow the industry to focus efforts on the key points in this complex system that will yield the biggest productivity improvements. Short of a crystal ball, a well-designed model is the best available tool for generating such insights.

A simple example of how this works has been published by Eric Minikel, a scientist at the Broad Institute in Cambridge, Mass. In his blog post about the model, he noted, with obvious puzzlement:

I have sometimes heard the claim that preclinical costs pale in comparison to clinical-phase costs, but I have yet to see data to back this up. … Rather than preclinical costs being negligible, I think the truth is just that they are very hard to estimate.”

We believe Minikel’s observation is true and under-appreciated (it does not appear in Herper’s article). It is a great example of how model-based analysis can generate useful and actionable insights.

A simple model of drug development investment decision-making, which builds on an original by Eric Minikel (Broad Institute).

Being embedded in the industry, we have good visibility into typical industry costs for preclinical activities, so we enhanced Minikel’s original model to add more detailed information about costs for cell-line and process development, toxicology studies, and commercialization. These are readily verifiable in the biologic drug development ecosystem, and we’ve linked to the key sources we used for developing the model. (The model is hosted here on DropBox under a Creative Commons license.)

A four-part prescription for biopharma

Exploring the model yields four concrete suggestions for anyone interested in fixing the productivity malaise in biopharma R&D.

1. For biologics, focus more on cutting cell-line and process development costs

Conventional wisdom holds that late-stage clinical costs are the main problem. Improving success rates for Phase 2 trials is certainly one way to improve things, but as noted above, Minikel’s model also shows that — viewed another way, and properly accounting for capital costs and development attrition — preclinical costs in fact dominate the financial analysis. This is especially true for preclinical biotech companies, which have higher capital costs.

In our model, the largest single expenditure in developing a new biologic drug is cell-line/process development — the late preclinical stage of development when the first batch of human-grade drug product is made. As shown in the figure below, in nominal terms this cost is almost invisible but, after properly adjusting for the time value of money and weighting for success probability, the effect can be seen clearly in year four.

Costs of a typical biopharmaceutical development program from preclinical development through commercial sales, with a 30-year time-horizon. Panel A: Nominal out-of-pocket costs and revenues. Panel B: The same costs and revenues adjusted for the time value of money (11% discount rate) and factoring in the number of replicates required at each stage to yield a single drug launch.

This might seem counterintuitive, since the direct cost is less than a typical Phase 2 study, but it aligns with findings published in 2010 by a team at Eli Lilly. This analysis sets the gold standard for historical cost analysis because the researchers had access to hard-to-obtain, non-public data on preclinical program costs and attrition rates. It also matches lived industry experience: writing a big check to a contract development and manufacturing company is the most sobering milestone in preclinical development.

2. Improve product convenience to cut development costs and expand markets

Most biologic drugs are given intravenously or by injection, generating a long list of product attributes that patients and health care systems dislike: most such products lack shelf stability; most cannot be self-administered; and trypanophobia (needle fear) affects about 25% of the population. These cumbersome attributes make it difficult for new drugs to be used by everyone who would benefit from them, and size-limited markets are less valuable, both financially and in terms of public benefit.

Removing these real-world access constraints can boost sales. Just look at Pfizer: Last year, it earned more than $50 billion from a pair of products — Comirnaty, the Covid-19 vaccine developed with BioNTech, and Paxlovid, the oral therapy for early Covid-19 infection — that are cheap, scalably manufactured, and easily distributed. And most of those revenues came from outside the U.S. The company has worked similar magic with its mass-market pneumococcal vaccine, Prevnar13, which has a list price of $220 and 2022 sales of $5.8 billion.

Pfizer’s experience counters the current industry consensus that profitability is best sought in rarer diseases with high unit prices. Cumbersome product attributes also drive up clinical development costs. Extra site visits for things like intravenous infusions drive up professional fees, recruitment costs, and attrition rates for clinical trial volunteers.

Treating trial participants and patients more like customers to be delighted on all dimensions of the product experience — not just the technical and regulatory aspects of product development — could yield big R&D productivity dividends.

3. Focus more on intrinsic safety

Improving drug safety is another big opportunity. Safety problems are a far bigger killer of clinical programs than lack of efficacy. This is intuitive: efficacy trials (Phase 2/3) are only run with drugs that first survive safety studies (Phase 1). Unexpected toxicity also kills drugs after launch, as happened with Vioxx and Rezulin.

The implications of product safety for R&D productivity can be easily overlooked because they are mostly indirect: the cost of the failed trial itself is emotionally salient, but in fact the direct cost is small compared to the indirect cost of throwing away all the prior investment in that program. In a sense, clinical trial failures amplify the cost of every earlier development expense by making it necessary for the rational investor to launch many early-stage programs to have a probability of one success — to be specific, 27 of them in the baseline configuration of our model.

Viewed another way, clinical attrition creates a malign effect that propagates backward in time through the model, multiplying the cost of every prior stage of development. To put hard numbers on it, in the baseline model it takes just a 2% improvement in Phase 2 success probabilities — from BIO’s 29% estimate to 31% — to completely cancel out the entire direct cost of Phase 2 trials. Investing more in safety could therefore yield huge productivity dividends.

Even better: If a drug can be made so intrinsically safe that regulators and ethicists concur that fewer and shorter studies are needed to ensure the safety of study volunteers, a developer would also save the direct cost — both money and time — of running those trials. This may sound fanciful to those in the world of systemically delivered cancer drugs, but many oral probiotic drug programs fall into this category, and the first of these was just approved by the FDA.

Recalibrating priorities and leveraging new tools like next-generation toxicology screens could yield a big jump in biopharma R&D productivity. And safer drugs have larger markets, which helps the ROI calculation from the other side of the equation.

4. Update statistical approaches to generate larger clinical effect sizes

The FDA requires evidence that new drugs are safe and effective, but approvable efficacy is a minimal threshold. Some approved drugs are simply better than others in the clinic, meaning they have larger effect sizes.

A larger effect size helps new-drug R&D productivity in both direct and indirect ways. It helps directly because, if everything else is equal, drugs with larger effects require smaller trials — a direct savings of time and money. Better drugs also sell better, a key consideration for the ultimate market size, another key variable in the ROI calculation.

Larger effect sizes help indirectly because drugs with big effect sizes are less likely to fail on statistical grounds due to the random background fluctuations in the natural occurrence rate of the disease symptom being treated, a factor critically important to the statistical validity of the clinical trial. Natural fluctuations in this background event rate — also sometimes called the placebo rate — is unavoidable, as Adam Feuerstein joked about on Twitter.

This last point may be preaching to the choir: Everyone in the biopharma industry is already hunting for big effect sizes. In a sense, our model just confirms what people in the industry are already doing. But here are two fresh ideas that we believe are under-rated and under-discussed.

First, drug cocktails. They have already revolutionized many disease areas via enhanced effect sizes. Common examples include oncology (chemotherapy combos), HIV/AIDS (antiretroviral cocktails), and hepatitis C (Harvoni). In a recent preprint, one of our team (B.F.) published a roadmap for exploiting this idea systematically and aggressively without breaking preclinical development budgets or putting study volunteers at risk.

Second, even greater productivity leaps may come from similarly clever statistical approaches to efficiently test novel drug combinations in the clinic but do it more quickly, cheaply, and systematically. Using 21st century statistical tools to better match dose size and cocktail composition to patient populations may yield effect size estimates that are a more reliable way to size smaller, less expensive Phase 3 studies.

Restoring new-drug R&D profitability

Biopharma R&D productivity has been in steady decline for at least three decades. Reversing this trend will require more than any one intervention. Players in the industry — big pharma, small biotech, and venture capital alike — share a moral obligation to fix the R&D productivity problem so it once again becomes affordable to develop new drugs for diseases other than those afflicting the most affluent sector of society. The end of cheap abundant investment capital also means that the market is standing by to discipline those who cannot meet the challenge, and reward those that do.

Brian Finrow is a co-founder and CEO of Lumen Bioscience, a clinical-stage biotechnology company in Seattle. Aleks Engel is a partner at Novo Holdings, the holding and investment company of the Novo Nordisk Foundation. Srinivas Akkaraju is the founder and managing general partner at Samsara BioCapital, a biotech-focused investment firm in Palo Alto, Calif. The views expressed here are the authors and do not necessarily reflect those of their respective organizations.



Brian Finrow

Jim and I started Lumen Bioscience in 2017 to develop and commercialize ultra-low-cost biologics.