Drug development is associated with a significant degree of attrition, which is generally thought to be in the 90 percent range, indicating that only about 10 percent of compounds entering development make it to regulatory submission, approval and market. During the past two decades numerous papers have been published on pharmaceutical attrition, attempting to arrive at an overall estimate of attrition, and the key reasons for attrition (e.g., Kola and Landis, 2004; Hay et al., 2014). Attrition rates have been reported to vary significantly depending on different attributes, such as therapeutic areas, phase of development and pharmacologic targets, and commonly cited causes of compound termination have included clinical safety, lack of efficacy, formulation, PK/bioavailability, commercial, toxicology, and cost of goods.
A general comment concerning the published literature on attrition and its causes, as well as assessments within individual pharmaceutical companies, is that definitions and attributions have not been standardized, and common methodology has been lacking. It is also noted that even within individual companies, when reviewing their compound termination databases, it’s not always clear what was considered the primary reason for the termination of a specific compound by the project team. Furthermore, the literature focuses primarily on small molecules, but success rates of biopharmaceuticals and vaccines are thought to be considerably better than those for small molecules.
Considering the significant negative implications of candidate compound attrition for the pharmaceutical industry, principally in terms of R&D expenditure, opportunity cost and productivity, the question arises how the industry might be able to reduce attrition, and thus be able to bring more new therapies to market. As an example, an R&D organization might base how many candidate compounds to bring into development each year on the desired number of drug candidates to be submitted for regulatory submission at a future date following clinical development. Thus, assuming a goal of one regulatory submission per year, at least ten candidate compounds would need to be brought into development each year. Such an inefficient drug development process obviously makes it much harder to address the many unmet medical needs areas.
The purpose of this commentary is to present a framework to address attrition of small molecules from a different perspective. It is not intended in any way to provide an extensive literature review of this complex topic; references are representative.
From Attrition to Predictivity
While different high-level categories have been proposed as reasons for attrition, such as scientific reasons, technical reasons, commercial reasons and regulatory reasons, each with their specific underlying set of causes, fundamentally the key causes of terminations are scientific, i.e., based on lack of efficacy, safety issues, and undesirable compound properties (compound properties are defined as determinants and descriptors of drug exposure). These are also the causes of terminations that together are by far the most common, and that lend themselves to addressing attrition by better understanding of the underlying scientific causes. This involves focusing on predictivity, i.e., how predictive are the preclinical methods and models of clinical performance, as expressed below:
PT = pe x ps x pc
Where PT stands for total predictivity with respect to scientific causes of terminations, and pe, ps and pc stand for the individual predictivities of efficacy, safety and compound properties, respectively. This is shown graphically below, and in an enlarged format here.
Note that the predictivities of these three scientific components are considered to be multiplicative, which assumes these are independent of each other. Also note that this leaves aside terminations by attrition categories such as cost of goods, formulation issues, budget/resource constraints, portfolio rationalization, potential values, patent issues and regulatory hurdles, as these are fundamentally unrelated to the three scientific causes mentioned above, and are typically business-related rather than scientific.
Predictivity Rates of Discovery and Preclinical Data
As was mentioned above, the data and methodologies that have been used to address attrition and attrition categories, as reflected in the published literature, have generally not been uniform or standardized. Yet in order to identify opportunities to reduce attrition, what’s needed are well-defined studies with robust statistical analyses of how predictive current preclinical studies of efficacy, safety and compound properties are with respect to clinical performance. With an overall success rate of only about 10%, for small molecules at least, one has to conclude that our understanding of how to confidently progress candidate compounds from preclinical to clinical is limited. Indeed, to quote Sir Peter Medawar, Nobel Laureate in Physiology and Medicine 1960: “No branch of science can be called truly mature until it has developed some form of predictive capability.”
A table of examples of preclinical methods and models of efficacy, safety and compound properties and examples of clinical failures due to lack of efficacy failures, safety/toxicology issues or undesirable compound properties is shown below, and in an enlarged format.
At the present time, specific values for predictivities of efficacy, safety and compound properties are not readily available in the publishedliterature, but it is possible to arrive at a general range for these, at least for compound properties and safety. While there are obviously a variety of complicating factors, such as the chemical properties of the candidate compounds, which therapeutic areas, and which specific parameters are being addressed for these scientific causes of termination, such guesstimates can be useful for identifying where the weaknesses in the current preclinical-to-clinical paradigm lie.
Compound Properties. A number of papers have been published on the predictivity of compound properties based on preclinical data and using a variety of computational models to predict PK parameters in humans and compared these with the observed parameters in man, typically from early Phase I studies (e.g., PISC/PhRMA group, 2011; Jones et al., 2013). The parameters addressed have included PK parameters, such as total clearance, AUC, maximum concentration, elimination half-life, bioavailability and food effect. The computational methods have included allometry, in vitro-in vivo extrapolation (IVIVE) and physiologically-based pharmacokinetics (PBPK). The published reports have included data from individual companies and from multi-company precompetitive collaborative efforts. In general, the predictivity for the individual compound properties parameters have broadly ranged from about 0.4 to 0.7 (predictions within a factor of two), although there has been considerable variability from report to report; PBPK appears to be the favored method.
Safety. There have also been numerous published studies addressing predictivity of safety, typically based on the concordance between adverse effects in man (after Phase I or after clinical development) and the findings from preclinical safety studies (e.g., Olson et al., 2000; Monticello et al., 2017). In general, these studies have found good concordance between the findings in preclinical and clinical studies, although there has typically been considerable variability with respect to different organ systems. While methodologies based on concordance are not specifically addressing predictivities, these are generally thought to be of a similar magnitude as those for compound properties, in the broad range of 0.4 to 0.7. It is noteworthy that regulatory requirements for preclinical studies for safety and compound properties are well standardized and prescriptive, including ICH guidelines.
Efficacy. In stark contrast to predictivity of safety and compound properties, the situation with respect to how predictive preclinical efficacy studies and methods are with respect to efficacy in humans is significantly more limited. Numerous challenges in this area have been suggested, such as non-uniform preclinical studies and methodologies associated with preclinical animal models, small sample sizes and translational challenges. However, for illustrative purposes in this commentary, one can grossly estimate these to be around 0.3, by solving for total predictivity of 0.10 and using 0.6 values for both predictivities of safety and compound properties; using values of 0.5 and 0.7 for safety and compound properties, efficacy predictivity values would be in the 0.4 and 0.2 range, respectively. Of the three scientific causes of attrition, clearly predictivity of efficacy represents the weakest link in successful progression of candidate compounds from the preclinical to the clinical stage of development.
The present commentary focuses on scientific causes of attrition, and the predictivities of those: efficacy, safety and compound properties. Non-scientific causes of attrition, as reported in the published literature, have varied somewhat, in percentage of total attrition and being generally business-related. Since these are not included, it follows that the above predictivity estimates are likely to be somewhat undervalued.
These considerations suggest that in order to meaningfully reduce attrition rate, much more work is needed to better define the predictivities of preclinical methods and models, particularly those with respect to efficacy. Work in this general area to date has clearly shown the value of precompetitive collaboration across the industry, as exemplified by recent work on compound properties (e.g., PISC/PhRMA group) and on safety (e.g., DruSafe group). It is important to note however that such collaborative work is both resource and time intensive and requires long-term commitment by all the concerned participants. Yet, it is hard to imagine how pharmaceutical productivity can be improved without it.
Predictivity => Productivity
Hay M, Thomas DW, Craighead JL, et al. Clinical development success rates for investigational drugs. Nat. Biotechnol., 32: 40-51, 2014.
Jones HM, Mayawala K, Poulin P. Dose selection based on physiologically based pharmacokinetic (PBPK) approaches. The AAPS J. 15(2): 377-387, 2013
Kola I, Landis J. Can the pharmaceutical industry reduce attrition rates? Nat. Rev. Drug Discov., 3:711-716, 2004.
Monticello TM, Jones TW, Dambach DM, et al. Current nonclinical testing paradigm enables safe entry to First-in-Human clinical trials: The IQ consortium nonclinical to clinical translational database. Toxicol. Appl. Pharmacol., 334: 100-109, 2017.
Olsen H, Betton G, Robinson D, et al. Concordance of the toxicity of pharmaceuticals in humans and in animals. Regul. Toxicol. Pharmacol., 32(1): 56-67, 2000.
PISC/PhRMA group. J. Pharmaceut. Sci, 100(10): 4045-4157, 2011. This issue has five papers from a PhRMA working group on predicting human pharmacokinetics, plus an editorial and a commentary by its two consultants.