Journal Club: Revisiting Eroom's Law

Episode Summary

Executive Summary: This Journal Club episode examines a Nature Reviews Drug Discovery opinion piece arguing that Eroom’s Law—the rise in drug development costs—may be flattening. Lauren Richardson, Vijay Pande, and Jorge Conde discuss three drivers of the trend shift: better biological data, better interpretation via AI/computation, and evolving regulatory/clinical trial practices that may lower development friction and risk.

Main Topics: Eroom’s Law and the recent cost plateau (Priority: 5/5): The episode centers on whether the long-term rise in drug R&D costs is finally stabilizing, based on post-2010 data analyzed in the referenced Nature Reviews Drug Discovery article. Better information from multi-omic, high-resolution biology (Priority: 5/5): Vijay argues that drug productivity is improving because researchers can now generate broader and deeper biological datasets, including genomics, transcriptomics, proteomics, and single-cell data. Better use of information through AI and machine learning (Priority: 5/5): The speakers emphasize that computational tools can synthesize complex datasets into more interpretable, decision-relevant insights, potentially improving both speed and quality of drug development. Automation, reproducibility, and engineering biology (Priority: 4/5): They frame a shift toward engineered, reproducible biology and automation as essential for making experimentation more scalable and less labor-intensive, which could reduce costs. Regulatory thresholds, biomarkers, and targeted trials (Priority: 4/5): They discuss how more targeted therapies and biomarker-based patient selection may reduce regulatory risk and improve efficacy signals, potentially making approval easier and faster. COVID’s effect and decentralized clinical trials (Priority: 3/5): The conversation suggests the pandemic accelerated modern trial designs such as virtual, distributed, remote, and continuously monitored studies, which may lower clinical development costs. Limits of the data and cost methodology (Priority: 4/5): They caution that estimates of R&D cost are methodologically contested and that different ways of slicing the data can support different narratives about whether costs truly fell.

Key Arguments: Biology is becoming more measurable at scale: broad multi-omic data, single-cell resolution, and computational analysis are improving insight generation and decision-making. Automation and reproducibility are prerequisites for engineering biology; once experiments are more standardized and machine-assisted, productivity can rise. Modern AI/ML is increasingly interpretable, which may make it more useful for drug development than earlier black-box approaches. Automating experiments and data analysis may reduce dependence on high-cost skilled labor whose costs inflate faster than ordinary consumer goods. Regulators may be more willing to approve therapies when trials are more targeted and patient selection is guided by mutations or biomarkers. Targeted development improves safety by excluding unlikely beneficiaries and improves efficacy by strengthening signal against noise. COVID accelerated acceptance of modern trial infrastructure, including virtual and distributed trials, decentralized monitoring, and at-home data collection. The apparent flattening in R&D cost may reflect a lag effect; innovations made 5-10 years ago may only show up in current data. The underlying cost of drug development remains high, risky, and time-consuming even if the trend has flattened. Methodology matters: different approaches to calculating R&D cost can materially change conclusions and are often used strategically in industry debates.

Data Points: Time window of cost analysis: Since 2010 - The cited paper analyzes drug development cost trends over the last decade and finds stabilization. Observed trend duration: Last 10 years - The speakers repeatedly reference a 10-year period in which costs appear to have flattened. Lag between innovation and market impact: 5 to 10 years - They note that changes in drug development today may not appear in cost data until years later because of development timelines. Clinical trial decentralization: Throughout the country - Described as a model where doctors in many locations can administer trials rather than only a single academic center. Data resolution: Single-cell level - Cited as an important increase in biological resolution for understanding disease.

Pivotal Quotes: "we can generate increasingly broad array of data across the various ways in which biology transmits information" — Vijay Pande: Explaining the first factor behind improving drug development productivity: better information. "The beauty of current machine learning and current AI is that they're being built to be interpretable." — Vijay Pande: Describing how AI can improve not just prediction but understanding in drug development. "the pendulum is definitely swinging towards trying to get these drugs in the hands of patients." — Jorge Conde: Discussing whether regulatory approval thresholds are changing in favor of faster access.

Implications: For listeners and industry, the takeaway is cautious optimism: new data, AI, and trial models may be easing Eroom’s Law, but drug development remains slow, costly, risky, and shaped by contested cost accounting.