Predictive performance of front-loaded experimentation strategies in pharmaceutical discovery: a Bayesian perspective

Abstract

Experimentation is a significant innovation process activity and its design is fundamental to the learning and knowledge build-up process. Front-loaded experimentation is known as a strategy seeking to improve innovation process performance; by exploiting early information to spot and solve problems as upstream as possible, costly overruns in subsequent product development are avoided. Although the value of search through front-loaded experimentation in complex and novel environments is recognized, the phenomenon has not been studied in the highly relevant pharmaceutical R&D context, where typically lots of drug candidates get killed very late in the innovation process when potential problems are insufficiently anticipated upfront. In pharmaceutical research the initial problem is to discover a “drug-like” complex biological or chemical system that has the potential to affect a biological target on a disease pathway. My case study evidence found that the discovery process is managed through a front-loaded experimentation strategy. The research team gradually builds a mental model of the drug’s action in which the solution of critical design problems can be initiated at various moments in the innovation process. The purpose of this research was to evaluate the predictive performance of frontloaded experimentation strategies in the discovery process. Because predictive performance necessitates conditional probability thinking, a Bayesian methodology is proposed and a rationale is given to develop research propositions using Monte Carlo simulation. An adaptive system paradigm, then, is the basis for designing the simulation model used for top-down theory development. My simulation results indicate that front-loaded strategies in a pharmaceutical discovery context outperform other strategies on positive predictive performance. Frontloaded strategies therefore increase the odds for compounds succeeding subsequent development testing, provided they were found positive in discovery. Also, increasing the number of parallel concept explorations in discovery influences significantly the negative predictive performance of experimentation strategies, reducing the probability of missed opportunities in development. These results are shown to be robust for varying degrees of predictability of the discovery process. The counterintuitive business implication of my research findings is that the key to further reduce spend and overruns in pharmaceutical development is to be found in discovery, where efforts to better understand drug candidates lead to higher success rates later in the innovation process.