Biopharmaceutical business development and licensing (BD&L) is a high-stakes arena, with manufacturers, private equity and other stakeholders simultaneously collaborating and competing to optimize their portfolios to achieve their strategic aims. Management teams must synthesize large quantities of information to make timely decisions about where to place large bets, and small changes in their underlying assumptions can determine whether a prospective investment is a success or failure.
One of the most critical assumptions BD&L teams use in evaluating development-stage novel biopharmaceuticals is how likely an asset or portfolio is to proceed through all stages of development and eventually win regulatory approval. This probability of technical and regulatory success (PTRS) informs how targets are screened and valued, and how deals are timed and structured.
Historically, most companies and investors have used this information primarily to assess potential downside risk. However, recent improvements in the quality of available data, analytical techniques, and artificial intelligence (AI) and machine learning (ML) technology now make it possible to uncover situations where the market may have “mispriced” risk. These developments allow both operating companies and investors to capture value by identifying development programs that may be more (or less) likely to succeed than the broader market believes.
We will demonstrate how applying a novel approach to quantifying risk can create a winning business development strategy based on three underlying beliefs:
- PTRS methodologies are not created equal. Novel methods can more accurately estimate drug development risk than traditional methods.
- Information asymmetry can create arbitrage opportunities. Those with superior insights about an asset’s likelihood of success can capture value by identifying under- or overvalued assets.
- Harnessing predictive power can create competitive advantages. Teams that embrace a consistent investment strategy across their portfolio based on improved risk quantification will have a competitive advantage over time.
A development program’s probability of success is defined as its likelihood of reaching its targeted endpoint and receiving regulatory approval. Historically, analysts have used a variety of methods to estimate this, including heuristics, historical benchmarks and statistical analyses of success rates for similar compounds. Traditional methods such as these suffer from three important limitations:
- Lack of validation. Validating probability of success methods requires (a) tracking a large sample of development programs in advance, (b) assessing success or failure in each case following completion and (c) comparing the actual aggregated results with the probability estimates. In practice, there has been limited validation of historical probability of success benchmarks.
- Input data limitations. Traditional methods of estimating probability of success are limited by the lack of harmonized input data on which to base predictions, especially in some newer therapeutic areas.
- Inference limitations. Existing probability of success methods must draw inferences from adjacent and often imperfect comparisons, possibly missing nuances of the indication at the trial level, such as varying clinical endpoints.
Advanced AI and ML techniques offer advantages over traditional methods for calculating probability of success. They can ingest a broader set of input data and identify inferences that analysts cannot spot by traditional means. Developers of these algorithms can integrate data from hundreds of sources into structured ontologies that yield not only a raw PTRS score but also insights into the relative contributions of various asset characteristics, such as drug biology, clinical trial design, trial outcomes, sponsor characteristics and regulatory pathway.
Novel AI and ML approaches have shown predictive power. Intelligencia, a health technology company that has developed an AI-based algorithm for PTRS, looked at pharma-sponsored oncology programs at the beginning of phase II trials to validate the accuracy of its algorithm. The algorithm correctly predicted 75% of programs that would go on to receive FDA approval and 78% of those that would not. Intelligencia, a ZS partner, is the source for all AI/ML-derived PTRS scores cited in this article.
To understand how companies and investors can leverage improved risk assessment as a competitive advantage in screening and valuing life sciences deals, we analyzed disparities in sources and methodologies for assessing development risk and their impact on asset valuation. This exercise identified strategies that can allow BD&L teams to identify winners early, improve confidence in valuation and inform transaction timing.
Identify winners early
During target screening, prospective buyers can assess drugs’ probability of success to identify likely winners that may be undervalued.
To illustrate the potential upside of this approach, ZS identified one case study in the lymphoma pipeline. In phase I, the ML-driven method pegged the asset’s probability of success at 28%—compared with a benchmark of 6% for similar drugs at the same development stage. The asset rose from a valuation of $60 million in phase I to a $2 billion buyout just three years later. An investor attuned to early, more accurate signals like this could construct a portfolio of comparatively lower-risk, early-stage assets for substantially less money than it would take to acquire a single late-stage winner. At scale, this strategy could yield a portfolio of assets with higher-than-average probability of approval, mitigating the idiosyncratic risk of each individual asset while improving long-term returns.
Improve confidence in valuation
When valuing assets, informed BD&L teams can use accurate risk measurements to improve their confidence in assets’ risk-adjusted net present value (rNPV). Having reliable information on which to base valuations can lead to stronger business cases for investment, better investment outcomes, smarter negotiation strategies and improved insight into the bid-ask spread in the context of a precise asset value.
Analyzing the leukemia pipeline, ZS identified 64 drugs in phase II and phase III development with at least two PTRS scores—one from an AI and ML-based method and one from a legacy one. For phase II assets, 40% had more than a 30-point spread between PTRS scores. For phase III drugs, more than half did.
For these assets, this variance equated to an average impact of more than $500 million in valuation.
Inform transaction timing
By accurately assessing risk, companies can drive value by understanding when to invest. For example, a biopharma company made headlines when it released positive results from a phase III immunology drug. Overnight, the company’s value nearly doubled to about $900 million. Based on AI and ML-informed probability of success estimates, however, ZS calculated that this company was potentially overvalued by more than $300 million post readout. Months later, the company’s valuation declined, falling back toward a figure that the AI and ML-driven estimate implied. A potential investor or partner using a risk-informed investment approach might have held out for this type of correction based on the risk data.
Early adopters who embrace this shift in approach to quantifying risk, and do it at scale, can create excess returns. In our experience, companies and investors typically begin by exploring “proof of concept” targets and transactions to vet the principle first. To scale, leading companies invest both in the data and underlying infrastructure to automate screening—things such as technology and dashboards to monitor target lists. Next, they systematize it as part of the business development process—for instance, when to evaluate as part of search and evaluation or as part of diligence. While relatively new as a concept, we expect companies will look to gain an advantage by investing early and using this capability as a differentiating investment tool.