lessons learned from the fate of astrazeneca's drug pipeline: a...

Given the wealth of scientific and technological advances in recent decades — ranging from ultra-high-throughput screening to the many breakthroughs in the understanding of disease biology catalysed by the ‘omics’ revolution — it might be anticipated that pharmaceutical research and development (R&D) productivity would be at an all-time high. In fact, despite huge increases in R&D investment, costs (especially those linked to drug development) have skyrocketed while the output of new drugs has remained largely static, resulting in a widely discussed ‘productivity crisis’ in the industry1–5.

The possible reasons for the decline in R&D productivity are complex. They include changes to the regulatory land-scape and an increased aversion to risk, a challenging reimbursement and payer environment, escalating clinical trial costs and the fact that we are trying to find treat-ments for more complex and difficult-to-treat illnesses, and/or improve on existing

treatments that already have substantial effectiveness. However, with the develop-ment of high-throughput and ultra-high-throughput screening and combinatorial chemistry approaches during the 1980s and 1990s, as well as the perception that a wealth of new targets would emerge from genomics, part of this productivity issue can also be attributed to a shift of R&D organizations towards the ‘industrialization’ of R&D5,6. The aim was to drive efficiency while retaining quality, but in some organi-zations this led to the use of quantity-based metrics to drive productivity. The hypoth-esis was simple: if one drug was launched for every ten candidates entering clinical development, then doubling or tripling the number of candidates entering development should double or triple the number of drugs approved. However, this did not happen; consequently, R&D costs increased while output — as measured by launched drugs — remained static1–4.

This volume-based approach damaged not only the quality and sustainability of R&D pipelines but, more importantly, also the health of the R&D organizations and their underlying scientific curiosity. This is because the focus of scientists and clinicians moved away from the more demanding goal of thoroughly understanding disease pathophysiology and the therapeutic oppor-tunities, and instead moved towards meeting volume-based goals and identifying an unprecedented level of back-up and ‘me too’ drug candidates. In such an environment, ‘truth-seeking’ behaviours to understand disease biology may have been over-ridden by ‘progression-driven’ behaviours that rewarded scientists for meeting numerical volume-based goals7,8.

In 2011, AstraZeneca embarked on a major revision of its R&D strategy. As part of this strategic review, the company launched a systematic longitudinal analysis of its small-molecule drug project portfolio to examine the data and the decisions made on projects, as well as to understand the root causes of attrition. The goal was to determine the most critical factors that were likely to improve the health of the R&D organization, and increase the probability of successful transitions to Phase III trials and, ultimately, the launch of new medicines.

The portfolio reviewIn 2011, a comprehensive review was under-taken of 142 drug discovery and development projects at AstraZeneca. The review covered projects from all therapeutic areas that had been active during the 2005–2010 period, from the phases following the completion of preclinical research through to the end of clinical testing in Phase II. The key aims of the review were to understand the major reasons for project closure and to identify the features of projects that correlated with successful outcomes.

We did not expand the review to look at Phase III for two reasons. First, success-ful transition through proof of concept (Phase II) remains the area where the indus-try overall has the highest rate of attrition and which must be improved. Second, the number of projects in Phase III for a single company is too small to be able to draw valid

O U T LO O K

Lessons learned from the fate of AstraZeneca’s drug pipeline: a five-dimensional frameworkDavid Cook, Dearg Brown, Robert Alexander, Ruth March, Paul Morgan, Gemma Satterthwaite and Menelas N. Pangalos

Abstract | Maintaining research and development (R&D) productivity at a sustainable level is one of the main challenges currently facing the pharmaceutical industry. In this article, we discuss the results of a comprehensive longitudinal review of AstraZeneca’s small-molecule drug projects from 2005 to 2010. The analysis allowed us to establish a framework based on the five most important technical determinants of project success and pipeline quality, which we describe as the five ‘R’s: the right target, the right patient, the right tissue, the right safety and the right commercial potential. A sixth factor — the right culture — is also crucial in encouraging effective decision-making based on these technical determinants. AstraZeneca is currently applying this framework to guide its R&D teams, and although it is too early to demonstrate whether this has improved the company’s R&D productivity, we present our data and analysis here in the hope that it may assist the industry overall in addressing this key challenge.

PERSPECTIVES

NATURE REVIEWS | DRUG DISCOVERY VOLUME 13 | JUNE 2014 | 419

© 2014 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Drug Discovery

a Project success rates between 2005 and 2010

Perc

enta

ge

0

10

20

30

40

50

60

70

80

0

100

Preclinical Phase I Phase II Phase III Preclinical(33)

Phase I(27)

Phase IIa(26)

Phase IIb(8)

b Project closures

Perc

enta

ge

AstraZenecaIndustry median

SafetyEfficacy

PK/PDStrategy

6663

59

48

15

29

60*67

82

639

62

15

15

8

35

57

8

12

88

conclusions, and this number becomes even smaller if looking at successful transitions to launch of a medicine.

It should be noted that this is a single assessment from a single company, based on a limited number of projects, and over a lim-ited timeframe. Nevertheless, we think that insights from this work could help to guide future teams and improve R&D productivity.

Methods. The aim of the review was to identify key ‘lessons learned’ in projects that could be used to improve the R&D produc-tivity of the company. As such, the review was performed by a cross-functional group of scientists and clinicians drawn from the project team community, and conducted in a peer-to-peer manner. To be as objective as possible, structured questionnaires were used when interviewing teams and, where possible, contemporaneous documents were analysed to provide supporting evidence for assessments. In addition, to further avoid any potential bias, senior leaders who had been associated with the governance deci-sions over the assessment period were not involved in the review.

For this analysis, the drug development process was divided into four distinct phases: preclinical, Phase I, Phase IIa and Phase IIb. The preclinical phase was defined as the phase from the first good laboratory practice (GLP) toxicology dose of a candi-date drug through to an investigational new drug (IND) application or first clinical trial application (CTA) before first-in-human testing. Phase I was defined as the phase that included the first-in-human trials within a small trial population (typically <50 patients) and included safety, tolerability and dose-ranging studies. These studies were often conducted in healthy volunteers, but in some indications (for example, oncology) they could include patients. Phase II trials were defined as trials that were aimed at evaluating the candidate drug’s efficacy in a patient population, leading up to clinical proof of concept. Within our analysis, we sub-divided Phase II into Phase IIa and Phase IIb. Phase IIa studies were generally smaller (typically <200 patients) and designed to mainly address early evidence of drug activity, whereas Phase IIb studies included larger numbers of patients (typically <400 patients) and were designed to demonstrate clinical proof of concept and an understanding of dose response.

For each of these phases, projects were classified as being ‘active’ (still in that phase), ‘closed’ (shut during that phase) or ‘success-ful’ (transitioned from this phase to the next

one). Every project was analysed separately in each phase of its development path; so, for example, a project that had reached Phase III trials was analysed four times across the entire development process. Data were col-lected for each project, for each of the devel-opment phases that it had completed, using comprehensive surveys and questionnaires with over 200 questions covering all aspects of the project (for example, the scientific rationale, target validation and physico-chemical properties of the candidate drug). Questionnaires were adapted so that they were specific to each phase of the review to allow for the retrospective understanding of the data that were available for a project at that stage, and to analyse how the project knowledge and data developed as the project passed through different phases. Written surveys were supplemented with in-depth peer-to-peer interviews with project teams. Responses to the questionnaires and inter-views were subjected to rigorous peer review by a team of experienced scientists and clinicians to ensure consistent evaluation across all projects. In-depth ‘root-cause’ analysis was used to reach conclusions as to why projects had failed. Analyses that were based on answers to specific questions in

the questionnaires were only performed on projects that had provided a complete set of answers to the relevant questions.

Results. Overall, we gathered data from more than 80% of the 142 AstraZeneca projects within the scope of the review, and for 95% of projects in clinical phases. Of the pro-jects analysed, 94 closed during the period assessed; 33 closed before clinical testing and a further 61 closed during clinical testing. The remaining projects were still active at the time of this review.

We compared the success rates for our projects to pharmaceutical industry bench-marks, obtained from the Pharmaceutical Benchmarking Forum (FIG. 1a). Our success rate in the preclinical phase (defined as the percentage of projects completing this phase and moving to the next phase of development) was comparable with industry benchmarks (66% versus 63%; see the KMR Group website for further information on the Pharmaceutical Benchmarking Forum). Our data suggested that we had a higher suc-cess rate in completing Phase I (59% versus 48%) but a markedly lower success rate in completing Phase II (15% versus 29%), com-pared to industry benchmarks. In addition,

Figure 1 | Overview of project success rates and reasons for closure. a | Overall project success rates for the AstraZeneca portfolio during the 2005–2010 period compared to the Pharmaceutical Benchmarking Forum (PBF) data. Briefly, the PBF collects performance data from across the pharma-ceutical sector and measures it against a number of carefully defined research and development (R&D) performance metrics. Data are aggregated in an anonymized form and provide a benchmark for performance across the industry. Success was defined as the percentage of projects that moved from the indicated phase to the next phase, and this percentage was compared to the industry median. The percentage success rate is shown above each respective bar. *Only five projects reached Phase III during this period. b | Stacked column plot showing primary reasons for project closure. Project closures were classified as failing because of safety (toxicology or clinical safety), efficacy (failure to achieve sufficient efficacy), pharmacokinetics/pharmacodynamics (PK/PD) or the strategy. The percentage of project closures at each of the indicated phases of development is shown in the bars and the number of projects assessed is shown in brackets below the bars.

P E R S P E C T I V E S

420 | JUNE 2014 | VOLUME 13 www.nature.com/reviews/drugdisc


http://kmrgroup.com/ForumsPharma.html

Phase III success rates were lower than the industry overall (60% versus 67%), although the number of projects in this phase was very low. Therefore, AstraZeneca was allowing more projects to enter later-stage development, only to have them subse-quently fail. Overall, AstraZeneca’s success in bringing candidate drugs to market during the 2005–2010 period was significantly lower than the industry median (2% versus 6%).

Unacceptable safety was the most impor-tant reason for failure, accounting for more than half of all project closures (FIG. 1b). The majority of these failures occurred before clinical testing (primarily during reg-ulatory GLP toxicology testing), with safety issues being the reason for 82% of preclinical project closures (it should be noted that this is the percentage of project closures, not all projects; two-thirds of projects successfully completed the preclinical phases and entered clinical testing).

Safety failures also dominated in Phase I, with 62% of closed projects at this stage fail-ing owing to safety issues. This remained an important reason for project closure in later development: 35% of projects failed owing to safety issues in Phase IIa, and 12% failed in Phase IIb (FIG. 1b). Overall, 30% of projects failed in Phase II as a result of safety issues — a figure that is apparently substantially higher than that recently reported for the industry overall9 (19% of project failures between 2008 and 2010, and 22% between 2011 and 2012). In addition, safety indirectly contributed to project closure by limiting the dose at which compounds could be tested in humans, thereby preventing adequate drug exposure and limiting target engagement (see the case study in BOX 1).

The majority of preclinical safety closures could be attributed to specific organ toxici-ties (FIG. 2a). Cardiovascular toxicity was the most common cause for such closure (17%), followed by hepatotoxicity (14%), renal toxicity (8%) and central nervous system (CNS) toxicity (7%). There was considerable attrition owing to musculoskeletal toxicities (12%), although these were predominantly due to research focused on matrix metallo-proteinase inhibitors; this research has been subsequently discontinued. We also observed a number of genotoxicity liabilities (in 10% of safety closures) attributed to a small number of chemical series that were subsequently abandoned. Overall, the rea-sons for safety failure were consistent with those observed by others in industry10.

During preclinical testing, 75% of safety closures were compound-related (that is, they were due to ‘off-target’ or other

properties of the compound other than its action at the primary pharmacological target) as opposed to being due to the primary pharmacology of the target (FIG. 2b). By contrast, the proportion of target-related safety closures rose substantially in the clini-cal phase and was responsible for almost half of the safety-related project closures. Such failures were often due to a collapse in the

predicted margins between efficacious doses and safety outcomes, meaning it was not possible to achieve target engagement or patient benefit without incurring an unacceptable safety risk.

In addition, we found that projects with preclinical safety signals often closed owing to safety issues in the clinic, whereas projects with minimal preclinical safety

Box 1 | Case study 1: the mGluR2 modulator AZD8529 for schizophrenia

AZD8529 is a positive allosteric modulator (PAM) of the presynaptic autoreceptor metabotropic glutamate receptor 2 (mGluR2). The rationale for mGluR2 as a target for schizophrenia was based on the hypothesis that failure of the cortical glutamatergic drive results in disinhibition of subcortical dopamine function, which in turn leads to psychosis61. The hypothesis predicts that normalization of glutamatergic function should have antipsychotic effects, with stimulation of mGluR2 proposed to provide such normalization62.

Before initiating a Phase II study, Lilly had disclosed that its dual mGluR2 and mGluR3 agonist demonstrated a statistically significant improvement in patients with acute psychotic schizophrenia63, although the results of a subsequent dose-finding study were negative. AZD8529 was active in seven preclinical models that were used to predict antipsychotic activity as well as in two preclinical models of anxiety. Effective plasma exposures for these models covered a broad range, from 2 nM to over 1,000 nM. Toxicology studies revealed testicular lesions in the dog models and cataracts in the rat models, which limited the maximum dose that could be given to humans. The dose for the Phase II study, 40 mg given every other day, was chosen based on the plasma exposure associated with the ‘collective activity’ in animal models and the need to provide a sufficient margin bearing in mind the toxicology findings. Limited cerebrospinal fluid (CSF) sampling conducted during the multiple-dose study indicated that free drug exposure in the CSF was roughly 50% of the free plasma exposure in humans. Efforts to develop a positron emission tomography (PET) ligand had been unsuccessful and there was no direct way to assess target engagement.

The Phase II study was a 4-week placebo-controlled study in 152 patients with symptomatic schizophrenia who were diagnosed and assessed using standard methods. Patients were randomized in a 2:2:1 ratio to receive AZD8529, placebo or the marketed antipsychotic risperidone. Risperidone had a robust effect on positive and negative symptoms relative to placebo, but no effect was observed with AZD8529. Following the negative Phase II study, further development in schizophrenia was stopped. Lilly also ceased development of its dual mGluR2 and mGluR3 agonist following the failure of two Phase III trials (see the 29 August 2012 press release on the Lilly website).

Applying the ‘5R’sThis programme failed before we implemented the ‘5R’ strategy, but can be analysed retrospectively in this framework. Before the initiation of the Phase II study, we only had low confidence that mGluR2 represented the ‘right target’ for the treatment of schizophrenia, on the basis of the well-documented limitations of preclinical models of efficacy in this indication, particularly with novel mechanisms64. Moreover, there was no clear understanding of how the underlying biology was linked to the disease. Similarly, there was low confidence in the ability to select the ‘right patients’ for this study because no information was available to identify the most appropriate patients for this study beyond the phenomenological diagnosis of schizophrenia.

However, the unmet medical need drove high confidence in the ‘right commercial potential’ dimension. The low confidence in the ‘right target’ and the ‘right patient’ dimensions focused attention on the remaining two ‘R’s; the ‘right tissue’ and the ‘right safety’. Confidence in the ‘right tissue’ dimension was low owing to the variable exposure observed as well as our failure to develop a PET ligand or other biomarker by which to measure target engagement. The toxicity, a side effect that was unacceptable for this indication, also raised concerns with regard to the ‘right safety’ and consequently limited dose selection in the clinic and the ability to test exposures that were likely to provide adequate target engagement.

Taken together, the inability to directly measure drug exposure at the target tissue at the appropriately predicted doses, coupled with clear gaps in our knowledge of the target and of patient selection, made it highly unlikely that this project would be successful. Applying the 5R framework would have identified many of these issues and gaps early on, and if they had not been addressed and solved by the team they would have prevented the programme from progressing into expensive proof-of-concept studies.




https://investor.lilly.com/releasedetail.cfm?ReleaseID=703018

High Medium Low

13

4938

Number of projects

Con

fiden

ce in

prec

linic

alsa

fety

pro

file


a Organ systems involved in safety failures

Perc

enta

ge

0

10

20

30

40

0

100

Cardiovascular Liver Musculoskeletal Genetic Preclinical(27)

Clinical(27)

b Safety failures

Perc

enta

ge

PreclinicalClinical

Compound-relatedTarget-related

17

24

1412 12

3

10

3

Respiratory Renal CNS Gastrointestinal

8

3

8 97

34

3

9

Other(for example,

immunity)

Respiratory and inflammation

CNS and pain

Cardiovascular and gastrointestinal

Oncology

Infection

Car

diov

ascu

lar

Live

rM

uscu

losk

elet

alG

enet

icRe

spira

tory

Rena

lC

NS

Gas

troi

ntes

tinal

21

3

75

25

52

48

Organ system

c Ratio of projects

d Number of projects

OtherSafetyEfficacy 0

1–23–45+

14

86

22

36

42

7

73

20

Organ system

Ther

apeu

tic

area

3 5* 5 2 4

2 7 1 1 2 6

6 5 1 2 5‡ 4

3 1

2

0

100

7 36 15

Rea

son

for c

linic

alpr

ojec

t clo

sure

(%)

signals rarely failed as a result of clinical safety issues (FIG. 2c). It should, however, be noted that a lack of preclinical safety signals was not sufficient to be a predictor of project success in later stages of development.

Overall, our analysis demonstrated an intuitive but crucial need for teams to pay attention to preclinical safety signals, and also highlighted that safety signals become more problematic as a project progresses, resulting in project delays. It also indicated the progression of molecules that could have been stopped earlier through the application of more robust criteria.

The next highest cause of project closure overall was a lack of efficacy in the chosen dis-ease indication. Over half of the project failures in Phase IIa, and 88% in Phase IIb, were due to lack of efficacy (FIG. 1b). Overall, 65% of pro-jects failed owing to efficacy issues in Phase II: a level that was, like safety failures, slightly higher than that published for the industry overall9 (51% of project failures between 2008 and 2010; 59% between 2011 and 2012).

We surveyed teams to understand the reasons why projects did not achieve clinical efficacy. The reasons were complex, with teams often reporting more than one contrib-uting factor (FIG. 3a). The ‘cleanest’ reason for

failure was when a project was able to clearly demonstrate that pharmacological engage-ment of a proposed mechanism of action did not result in clinical benefit in the patient population tested. In these instances, 40% of project responses indicated that teams lacked data demonstrating a clear linkage of the tar-get to the disease or access to a well-validated animal model of the disease. Programmes in this category were considered to have a poor understanding of the role of the target in the underlying disease pathophysiology; however, they were also thought of as ‘good’ failures as there was little doubt that the hypothesis was tested and that it was proven to be wrong.

Figure 2 | Analysis of project closures due to safety issues. Preclinical and clinical projects that had been closed because of safety issues were analysed to understand the principal causes of failure. a | Major organ systems involved in preclinical (red bars) and clinical (light brown bars) safety closures. b | Safety closures in the preclinical and clinical phases were categorized as either being related to the action of the drug at the primary biological target (red bars) or to other properties of the candidate drug (light brown bars). Percentages are shown within the bars and total numbers of projects shown under each bar. c | The level of confidence that teams had in their preclinical safety profile (lower graph) was compared to the reasons

for project closure in the clinical phase (upper graph). Percentages of projects in each category are shown within bars, and numbers of project closures analysed are shown underneath each bar. d | Table showing the breakdown of major organ systems associated with safety closures in pro-jects from different therapeutic areas. Only major organ toxicities are shown. *All musculoskeletal closures in the respiratory and inflammation area were associated with projects involving matrix metalloproteinase inhibitors.‡60% of nervous system failures in the cardiovascular and gastro-intestinal area were associated with projects involving cannabinoid receptor antagonists. CNS, central nervous system.





a Reasons for lack of clinical efficacy

b Phase II projects

0

100

Yes(15)

No(21)

Rat

io o

f pro

ject

s (%

)

Active or successfulClosed

27

73

57

43

40 (18)Target linkage to disease not establishedor no validated models available

29 (13)Dose limited by compound characteristicsor tissue exposure not established

20 (9)Indication selected does not fitstrongest preclinical evidence

11 (5)Evidence from previousphase not robust enough

0 10 20 30 40 50

Percentage of all reported reasons (total number of projects: 28)

Phase IIa projects

Yes(17)

No(7)

18

82

71

29

Projects with human geneticlinkage of the target tothe disease indication

Projects with efficacybiomarkers availableat start of phase

c Confidence in target

0

100

Preclinical(142)

Phase I(73)

Rat

io o

f pro

ject

s (%

)HighMediumLow

17

39

44

20

25

55

Phase IIa(51)

Phase IIb(19)

11

46

43

11

44

45

Number of projects

By contrast, 29% of project responses indicated that the properties of the com-pound either limited the dose and/or could not establish compound exposure in the relevant target tissue (FIG. 3a). For example, AZD3778, a novel chemokine receptor antagonist that was being developed for the treatment of asthma, had undesirable phar-macokinetic properties with high protein binding and a much shorter half-life than expected in humans. In addition, safety con-cerns associated with the compound placed limits on human dosing. AZD3778 failed to show efficacy in a clinical proof-of-principle study, and because of the compound’s properties it was unclear whether adequate receptor exposure had been achieved, so the team was no closer to validating or invalidat-ing the approach or the target (see also the case study in BOX 1). For 21% of projects that were classified as efficacy failures in Phase II (notably in neuroscience), there was no way to determine target engagement or pharma-codynamic activity. In all such cases, it was unclear whether a compound had tested the mechanistic hypothesis, as there was no demonstration of compound exposure and/or evidence of pharmacological engagement at the target tissue. Such projects were clas-sified as ‘poor’ failures as they failed to vali-date or invalidate the target hypothesis and thus left teams with no improvement in the understanding of their target hypothesis.

Projects that showed greater confidence in target validation, genetic target linkage to the disease or a stronger understanding of the role of the target in the disease aetiology were less likely to fail owing to a lack of effi-cacy. 73% of projects with some genetic link-age of the target to the disease were active or successful in Phase II compared with 43% of projects without such data (FIG. 3b). Patient selection was also a contributing factor in project success; high confidence in patient selection positively correlated with active projects in Phase IIb, whereas low confi-dence in patient selection correlated with project closures in the same phase owing to a lack of clinical efficacy (FIG. 4a).

Surprisingly, for a number of projects (9 out of 28) it was indicated that their development plans had not targeted the optimal patient population based on the scientific understanding of the disease at the time (FIG. 3a). To better understand this, we examined how confidence levels changed in teams with regard to their target in the disease indication and population, based on the scientific evidence, compared to the confidence in the perceived commercial value (FIG. 4b). Early in development, project

teams demonstrated greater confidence that they were studying the most appropriate indication and patient population for their target, whereas commercial confidence was low. Conversely, in later phases there was high commercial confidence in the projects but low confidence that the projects were in the indication and patient population for which the scientific rationale was strongest. This suggests that some teams were driven

by commercial value in addition to studying the most appropriate patient population based on the scientific understanding of the disease at the time.

A surprising factor contributing to pro-ject failure was the transitioning of projects to the next phase in the absence of suf-ficiently robust data (FIG. 3a). For example, 18% of projects that failed in Phase II owing to a lack of clinical efficacy (5 out of 28)

Figure 3 | Analysis of project closures due to efficacy issues. Projects that were terminated owing to efficacy-related issues were analysed to understand the causes for failure and identify potential predictors of success. a | Project teams were surveyed for the reasons why their projects failed, and their responses were categorized as shown. Project teams could report more than one contributing factor for the lack of clinical efficacy observed. Percentages of project responses for each category are given in bars. The numbers in brackets represent the number of projects that reported each category as a contributing factor for the failure of the project. b | Left panel: projects in Phase II were analysed for evidence of genetic linkage of their biological target to the disease indication. Responses were compared against the project status — whether it was still active or closed. Right panel: projects in Phase IIa were analysed to determine whether efficacy biomarkers were available before these pro-jects entered Phase IIa development. Responses were compared against the project status — whether it was still active or closed. Percentages for each category are shown within the bars and the number of projects are shown in brackets on the x axis. c | Team confidence in the target biology was assessed and categorized as ‘high’, ‘medium’ or ‘low’ based on the scientific evidence supporting a role for the target or pathway in the indication being studied at that particular phase.





Confidence in patient selectionin Phase IIb

a

0

100

High(27)

Low(27)

Perc

enta

ge o

f pro

ject

s

10

90

78

22

b Ratio of projects

0

100

Rig

ht ta

rget

inri

ght p

atie

nt (%

)

Low confidenceHigh confidence

60

40

50

50

11

89

0

100

Preclinical andPhase I (142)

Phase IIa(51)

Perc

eive

d co

mm

erci

alva

lue

(%)

5

95

71

29

Phase IIb(19)

67

33ClosedActive

were identified as having transitioned into this phase based on weak clinical evidence, which is indicative of inadequate project governance. One potential reason for this observation might be that, as noted previ-ously, the use of volume-based metrics encouraged project teams and leadership groups to progress projects to the next phase in order to meet yearly goals.

Only 5% of projects failed because the candidate molecules did not have the neces-sary pharmaceutical properties (for example, properties relating to drug metabolism, pharmacokinetics, bioavailability, and so on). This is in line with previous analyses11 demonstrating the reduction in attrition for such compound-related reasons over the past two decades.

The remaining project failures were categorized as ‘strategic’ and represented decisions to close projects for non-technical or non-scientific reasons. For example, AstraZeneca exited osteoarthritis R&D during the period analysed, and conse-quently closed a number of projects in this indication (mainly before GLP toxicology

testing). Strategic closures accounted for 7% (seven projects) of project closures, and four clinical projects were terminated for strate-gic reasons during this assessment period. In contrast to failures that were either due to safety or to lack of efficacy, this figure is much lower than that published for the industry overall9 (29% of project closures in Phase II between 2008 and 2010; 16% between 2011 and 2012). This may, in part, be because in our analysis we made a con-siderable effort to uncover the underlying reasons for ‘strategic’ closures and often sub-sequently categorized them as either ‘safety’ or ‘efficacy’ failures. This may also be why our failures in these two categories are also apparently higher than the published cross-industry metrics. However, it should be noted that repeated strategic shifts in disease area focus can also contribute to increased attrition and reduced productivity.

Overall, these data highlight that, throughout every phase of early R&D, it is crucial for scientists and clinicians to gain an understanding of, and confidence in, the disease biology, the relationship of the target

to the disease indication, and the proposed mechanism of action of a potential drug in the context of the right patient.

Key factors underlying project failuresUsing the data from our analyses, we identi-fied five key technical factors as substantial contributors to project failures. These were as follows: the strength and quality of target validation (the right target), demonstration of target engagement (the right tissue), safety margins (the right safety), patient stratifica-tion plans (the right patient) and the medical value proposition (the right commercial potential). We called these five categories the five ‘R’s (FIG. 5).

Right target: the importance of solid biological and disease understanding. Lack of efficacy was the most important cause of project failure in clinical trials (FIG. 1b). It is obvious that selecting the biolog-ical target for a drug discovery programme is one of the most important decisions a team will make. The safest and most potent molecule will still fail if a team is working on the wrong target for the disease of interest.

Given the importance of target selection, one would expect that confidence in the target would increase as a project progresses through the pipeline stages, and this would be driven by a range of evidence. This can include direct evidence of target linkage to human disease, genetic evidence from animal models, an understanding of the biology underpinning the target and/or disease aetiology, confidence in preclinical and clinical data generated using animal models, data generated with tool com-pounds in the preclinical or clinical setting, and validated efficacy biomarkers.

In our analysis, a high level of confidence in the biological role of the target in human disease was a predictor of successful pro-jects. For example, the availability of human genetic data linking the target to the disease before candidate drug nomination was more common in projects that remained active in Phase II (73% of projects for which there was evidence of genetic linkage of the target to the disease were still active in, or successfully completed, Phase II), whereas the absence of such data was more common in projects that failed (57% of projects that failed owing to efficacy did not have any evidence of genetic validation; see FIG. 3b, left-hand panel). The availability of efficacy biomarkers was simi-larly associated with active projects; 82% of projects with an efficacy biomarker were active or successful in Phase IIa compared to less than 30% of projects without such biomarkers (see FIG. 3b, right-hand panel).

Figure 4 | Analysis of patient selection and commercial positioning. a | Confidence in patient selection was analysed for all projects reaching Phase IIb. Confidence was categorized as either ‘high’ or ‘low’ based on whether teams were able to confidently select a stratified or enriched patient popu-lation during their trial. Responses were compared against projects that were still active (blue bars) or closed (light brown bars). Percentages are shown within the bars and the total numbers of projects shown under each bar. b | Upper panel: confidence that the teams were studying the right target in the right patient population was assessed based on the scientific evidence available at the time in each phase and categorized as being either ‘high’ (high scientific confidence that the team was pursuing the right target in the right patient population) or ‘low’ (low scientific confidence the team was pursuing the right target in the right patient population; that is, the study was probably being carried out in the wrong patient population). Lower panel: confidence in the perceived commercial value was assessed by teams and categorized as follows: either there was ‘high’ (dark grey bars) or ‘low’ (dark red bars) confidence that the indication being pursued was commercially attractive.





Right target

• Strong link between target and disease• Differentiated efficacy• Available and predictive biomarkers

Right safety

• Differentiated and clear safety margins• Understanding of secondary pharmacology risk• Understanding of reactive metabolites, genotoxicity, drug–drug interactions• Understanding of target liability

Right tissue

• Adequate bioavailability and tissue exposure• Definition of PD biomarkers• Clear understanding of preclinical and clinical PK/PD• Understanding of drug–drug interactions

Right patients

• Identification of the most responsive patient population• Definition of risk–benefit for given population

Right commercial potential

• Differentiated value proposition versus future standard of care• Focus on market access, payer and provider • Personalized health-care strategy, including diagnostic and biomarkers

In addition to strong human validation, confidence in the ‘right target’ was based on developing a good platform of evidence in preclinical studies, and in particular on the level of confidence that data from such studies are translatable to human disease. For example, based on previous clinical experience and the known validity and translation of the models, the level of confi-dence in the data generated from preclinical models to test anticoagulants intended to treat thrombosis or acute coronary syndrome was deemed to be high12.

By contrast, target confidence in some disease areas (for example, oncology, respira-tory and CNS) can be low because it is based on screens with unknown or poor transla-tion to clinical outcomes13–16. For example, the use of subcutaneous tumour xenograft models has dominated oncology research14, but for various reasons these models do not accurately replicate the human disease: they often use immunocompromised animals; the human tumour material is not introduced at the site of its primary source; and, for many patients with cancer, morbidity is due to metastatic disease rather than the primary tumour. It is possible to address some of these problems, for example, by introduc-ing the primary tumour in the appropriate site rather than subcutaneously, but this increases the complexity of the models, which has consequences on throughput and interpretation14. Modelling complex neuro-degenerative diseases such as Alzheimer’s disease in mice is also challenging16,17. Mice overexpressing mutant forms of the gene encoding human amyloid precursor protein (APP), which is known to be a causative factor in early-onset familial Alzheimer’s disease, do not develop a murine equivalent of human Alzheimer’s disease but instead show robust and rapid accumulation of cerebral amyloid-β. Genetic animal models can be useful for measuring target engage-ment, but they should not be used as faithful representations of the disease16.

Of course, human linkage data (even if functional) may not be fully predictive of the validity of a target. For example, genetic data supported the potential of CC-chemokine receptor 5 (CCR5) antagonism in the treat-ment of rheumatoid arthritis; a naturally occurring human variant of the CCR5 gene (CCR5-Δ32), producing a non-functional receptor, was negatively associated with rheumatoid arthritis18. Preclinical models also supported the therapeutic potential of this approach19,20. Nevertheless, several CCR5 antagonists (AZD5672, aplavoiroc and maraviroc) failed to show clinical

benefit in patients with rheumatoid arthritis despite the dosing being well tolerated and despite achieving exposure at the target21–23.

Overall, understanding the biological and clinical evidence supporting target selection is crucial and provides a basis to direct fur-ther studies to strengthen or invalidate the scientific hypothesis24 (see the case study in BOX 2). A striking observation in our analysis was that, for most projects, confidence in the biological role of the target did not improve as the project progressed from preclinical candidate selection through to Phase II (FIG. 3c).This suggests a failure to rigorously ‘pressure test’ the biological hypothesis of projects. In fact, as previously noted, some projects still progressed despite only having limited knowledge of the target biology (FIG. 3a).

Right tissue: demonstrating candidate drug exposure and pharmacological activity in the target organ. In addition to the choice of the ‘right target’, the probability of success is increased if it can be demonstrated, through appropriate pharmacokinetics/pharmaco-dynamics (PK/PD) modelling in both preclinical and clinical models, that the can-didate drug achieved exposure in the target organ and achieved sufficient pharmacologi-cal activity. Within the AstraZeneca analyses, this is referred to as the ‘right tissue’ and covers pharmacokinetic properties, target

engagement and pharmacological activity together with the appropriate understanding of PK/PD, which draws together these prop-erties relative to the target organ. Our analysis demonstrated that only a small number of projects failed directly as a consequence of poor PK/PD (that is, the candidate did not have the required pharmaceutical properties; see FIG. 1b). However, the project question-naires illustrated that <10% of projects had a high level of confidence that there was a combination of good drug properties (for example, bioavailability, PK/PD, formulation, and so on) and good pharmacological end points (for example, target occupancy and evidence of pharmacological modulation).

A separate analysis of the success factors for Pfizer’s development candidates (over the 2005–2009 period) concurs with the impor-tance of the ‘right tissue’ concept by showing that the candidates that were most likely to test the mechanism of action and that had an increased likelihood of Phase II survival were those for which there was an integrated understanding of three key principles: expo-sure at the site of action, target occupancy, and expression of pharmacological activity in the target organ. These principles were referred to as the ‘three pillars of survival’25.

Within the AstraZeneca data set, the development of the NMDA (N-methyl-d-aspartate) receptor channel blocker AZD6765 is an example of good alignment

Figure 5 | The 5R framework. Summary of the key features of the five-dimensional framework that can be used to describe a drug discovery and development project. PK/PD, pharmacokinetics/pharmacodynamics.




with the ‘right tissue’ (and the ‘three pillars’) concept. AZD6765 was a compound of interest for major depressive disorder, given that the NMDA receptor channel blocker ketamine had been shown to exert anti-depressant effects within a few hours26,27. However, it was postulated that differenti-ated modulation of the NMDA receptor complex relative to ketamine would result in an improved tolerability profile compared with ketamine (which has psychotomimetic effects)28,29.

Indeed, AZD6765 demonstrated differ-ential modulation of the NMDA receptor complex in studies in cultured neurons, showing lower trapping of NMDA-induced currents than ketamine30. Before under-taking clinical trials in patients with depres-sion, it was important to test, in preclinical models, the hypothesis that AZD6765 has an appropriate PK/PD relationship in the target organ and a different pharmaco-logical mode of action to ketamine. These factors were key to establishing a good understanding of the ‘right tissue’ concept to increase confidence in the therapeutic potential of AZD6765. Using modulation of electroencephalography (EEG) gamma band power in the frontal cortex as a pharmaco-dynamic biomarker, a concentration–effect relationship was established for AZD6765 and ketamine in mouse and rat models. For AZD6765, the mode of action was best described by inhibition of the turnover rate of the EEG gamma band power signal, which is consistent with the low trapping hypothesis. However, it was not possible to describe the data for ketamine using this model, which illustrates that ketamine has a different mode of action to AZD6765. Assuming that a change of 20% in the EEG gamma band power from the baseline is a statistically significant threshold, thera-peutically relevant concentrations could be estimated in humans, which could be subsequently tested in clinical studies.

Through these studies, an understanding of the human PK/PD relationship of AZD6765 in the target organ was estab-lished, and therapeutically relevant concen-trations were determined to guide clinical studies; moreover, there was increased confidence that the pharmacological mode of action of AZD6765 was different to that of ketamine. Encouragingly, a proof-of-concept study of AZD6765, in which a single intra-venous infusion of AZD6765 was given to patients with treatment-resistant major depressive disorder, showed that the com-pound rapidly improved depressive symp-toms without inducing psychotomimetic

Box 2 | Case study 2: the CRTH2 antagonist AZD1981 for asthma

AZD1981 was an antagonist of a G protein-coupled receptor known as chemoattractant receptor-homologous molecule expressed on T

H2 cells (CRTH2; also known as PTGDR2) and it blocked the

action of prostaglandin D2 (PGD2) with the aim of reducing cellular infiltration and thus decreasing

the airway inflammation implicated in asthma pathogenesis. Preclinical studies carried out in CRTH2-deficient mice using ovalbumin (OVA) challenge and using the PGD2 agonist (BW245C)65, as well as preclinical studies carried out using an oral CRTH2 antagonist (S-5751) in guinea pig and sheep models66,67, demonstrated that PGD2 mediates asthma-related inflammation. Antagonism of CRTH2 in animal models of asthma68,69 and in patients with moderate persistent asthma70 demonstrated reduced inflammation. In vitro data in both human eosinophils and T helper 2 (T

H2)

cells supported the predicted efficacy of AZD1981 in the treatment of allergic asthma; however, the lack of CRTH2-mediated functional responses in cells from rats or mice precluded investigation of allergic airway inflammatory responses in preclinical models.

AZD1981 did not reveal any significant preclinical safety concerns following acute dosing, although a low level of acyl-glucuronidation was observed that could limit long-term treatment. In addition, minor increases in the levels of liver enzymes were seen in preclinical test species at higher doses; although these were within acceptable ranges, they placed limits on human exposure.

Two Phase IIa studies were carried out using AZD1981 in patients with asthma, but neither study showed any safety concerns. The first Phase IIa study in patients with mild to moderate asthma who had a positive skin prick test (atopic/allergic asthma) demonstrated two key findings. First, it showed that AZD1981 maintained the morning peak expiratory flow (mPEF; the primary end point) at similar levels to pre run-in before inhaled corticosteroid (ICS) withdrawal. By contrast, patients receiving placebo had a reduction in mPEF. Second, ICS withdrawal during run-in resulted in a large decrease in the forced expiratory volume in 1 second (FEV

1), which was only partially restored

(~75%) by treatment with AZD1981. In the second Phase IIa study, patients with moderate to severe asthma were placed on a 2-week ICS run-in before treatment with one of three doses of AZD1981 or placebo. The study failed its primary mPEF end point at all three doses. During the run-in phase, all treatment groups had a marked improvement in FEV

1, and only patients receiving

the middle dose showed a statistically significant improvement in FEV1 compared to placebo.

By contrast, patient-reported outcomes from the Asthma Control Questionnaire (ACQ) showed a statistically significant improvement above the placebo response at all three doses. A post-hoc analysis in which the population was stratified into atopic (72%) and non-atopic (28%) patient groups indicated that the FEV

1 lung function and ACQ responders fell within the atopic cohort.

The team ran a large Phase IIb study, using atopy as a criterion for patient selection (PhadiaTop positive only). This study was carried out in patients with moderate to severe asthma who were stabilized on a background of ICS and inhaled long-acting β-adrenergic receptor agonists during a 2-week run-in period. Of the six treatment arms in the study, only patients on the 40 mg twice daily dose (the second lowest dose) demonstrated a statistically significant improvement in FEV

1.

Overall, the study failed to meet its primary end point. Retrospective analysis of efficacy data did not reveal any additional subpopulation of patients who consistently demonstrated improved efficacy with AZD1981 compared to placebo. Amgen recently reported the failure of a CRTH2 antagonist in improving asthma symptoms or lung function in patients with moderate to severe asthma71.

Applying the ‘5R’sThis programme failed before we implemented the 5R strategy, but as with the case study in BOX 1 it can be analysed retrospectively in this framework.There was low confidence in the ‘right target’ dimension for AZD1981 in asthma; the underlying biology was insufficiently understood and the linkage of the target to the disease biology was not convincing. Moreover, owing to the inability to utilize animal models to help support the progression of the molecule, AZD1981 entered clinical testing essentially based on in vitro data.

The large unmet medical need and consequent potential market (representing the ‘right commercial potential’) for asthma formed part of the investment decision to progress AZD1981. Although clinical validation was not confirmed in the Phase IIa studies, retrospective analysis suggested that AZD1981 showed limited efficacy in an atopic/allergic asthma population selected by a positive skin prick test. Although a small signal was identified, the weight of evidence of the clinical benefit of patient selection (for the ‘right patient’ dimension of the 5R framework) was not compelling enough to pursue the compound further. In addition, no evidence of target engagement or an understanding of pharmacokinetics/pharmacodynamics (PK/PD) was generated during Phase I or IIa, which resulted in low confidence in the ‘right tissue’ dimension.

Finally, although the doses tested in these studies did not exhibit any safety issues that would place patients at risk, the limits placed on dosing owing to the probable hepatotoxic risk meant that ‘right safety’ was an issue that was likely to become worse with expanded clinical testing and longer exposure to the study drug. Taken together, the 5R analysis would have probably led to the closure of this project following the initial Phase IIa trials.




effects31. It should be noted, however, that a recent Phase IIb clinical study with AZD6765 reported negative results; analysis of this trial is ongoing but initial data suggest that the study may have failed owing to an abnormally high placebo response rate.

The power of PK/PD modelling was illustrated by its successful application in the characterization of the therapeutic concentration range for the direct thrombin inhibitor AZD0837, which was developed for the prevention of stroke and systemic embolic events in patients with atrial fibrillation. Blood is the biophase for anti-coagulation, and a close exposure–response relationship was expected. Relative potency was first investigated in vitro and then in animal and human disease models, and this supported dose selection for Phase IIa studies32–35.

However, because of the low frequency of clinical events, large dose-finding studies would have been needed to estab-lish the dose regimen that would achieve a positive risk–benefit balance in terms of adequate antithrombotic efficacy and low bleeding risk. Instead, the dose selection strategy for AZD0837 was based on PK/PD modelling data for D-dimer (a degrada-tion product of fibrin that was used as a biomarker of thrombogenesis) and using clinical data available for ximelagatran, a previously developed thrombin inhibitor, which would dramatically shorten the clini-cal programme. The US Food and Drug Administration (FDA), in a scientific con-sultation meeting (type C, end-of-Phase IIa meeting), provided valuable discussions and contributions to this approach. Exposure–response data from the Phase IIa studies, which included approximately 1,200 patients with atrial fibrillation, were com-bined with data from the previous develop-ment programme of ximelagatran36,37 (in which more than 7,000 patients with atrial fibrillation were treated with ximelagatran or a vitamin K antagonist) to model the bleeding risk and define the maximum exposure for treatment with AZD0837.

The greatest challenge in gaining confi-dence about the ‘right tissue’ occurs when the pharmacological target and blood are separated by a barrier, such as the blood–brain barrier, as illustrated by the dopamine D3 receptor agonist PF-592379 in the study by Morgan et al.25, and by the AstraZeneca case study of AZD8529 in BOX 1. The use of imaging techniques, particularly position emission tomography (PET), to understand the relationship between blood exposure,

brain receptor occupancy and efficacy has been effective in the development of CNS-active drugs38. Added confidence in the relationship between blood exposure and the pharmacological target is provided when the same biomarker of pharmacologi-cal activity can be assessed in both the brain and blood39,40. More recently, elegant PK/PD relationships have been developed for CNS targets, which allow complex receptor kinetics to be understood and translated to the clinical situation41.

The ongoing challenge is to build PK/PD relationships in settings where direct measurement of drug exposure at the target site may not be attainable experimentally. The evolution of physiologically based pharmacokinetic modelling combined with non-invasive techniques to determine target engagement will lead to the generation of improved translatable PK/PD relationships for ‘right tissue’ alignment between animal models and humans.

Right safety: having an appropriate safety profile. Safety was the single most important reason for project closures in our analysis. Over 50% of the failures in preclinical studies and Phase I trials were due to safety issues that were incompatible with the pro-posed lead indication, and safety was also responsible for 30% of the Phase II project failures (FIG. 1b). So, building confidence in the ‘right safety’ profile of a given molecule represents a key opportunity for improving overall success rates. However, this is not a trivial issue, because it depends on a number of factors, such as the disease indication and the unmet medical need therein. There is no such thing as ‘absolute drug safety’; what would be considered an acceptable safety pro-file in one indication would be unacceptable in another.

Safety assessment requires dealing with both hypothesis-driven components (for example, predicted or potential risks asso-ciated with modulation of the target) and hypothesis-free components (for example, observed toxicology signals or adverse events caused by the compound). The latter can only be identified once a candidate or lead series enters extensive preclinical or clinical testing; furthermore, these compo-nents may only emerge late in development or at the post-marketing stage42. Given that not all safety signals can be predicted, it is not surprising that safety is a major reason for project closures between preclinical and clinical phases. However, a situation in which over half of all projects fail owing to safety issues seems unsustainable.

Analysis of our safety failures suggested that the reasons for failure could be divided into two broad categories: either a failure to detect a safety signal, or a failure to appro-priately assess the risk of a safety signal. A failure to detect a safety signal in preclinical screens (for example, in vitro, in vivo or in silico) was responsible for ~40% of safety failures (data not shown). Such failures high-light the limits of preclinical safety screening and the need for continued R&D on new assays to identify safety risks.

The second category of failures is more complex. Projects that had low confidence in preclinical safety and subsequently failed in clinical development usually failed because of safety and often for reasons that were directly related to the preclinical safety observation (FIG. 2c). There were three major reasons for such failures. The first reason was that the predicted exposure margins between safety and the predicted efficacious dose were over-estimated and did not trans-late to human dose escalation. The second reason was that safety signals were unclear or difficult to translate from preclinical data to patients; for example, the cannabinoid receptor (CB1) antagonist programme was closed following the withdrawal of rimona-bant owing to reports of suicidal ideation43 — an effect that was impossible to predict using preclinical models. The third reason was that safety signals were observed and considered by the team, yet the project still progressed because there was a rationale to believe that these signals would be accept-able and manageable within the disease indi-cation. In this instance, 11 out of 14 projects with a preclinical safety signal subsequently closed in the clinic owing to safety issues, which suggests that teams were too optimis-tic and did not pay enough attention to the preclinical signals (FIG. 2c).

We also observed a ‘disease area bias’ in the reasons for safety closure. For example, we found that 24% of the safety closures in cardiovascular projects were for cardiovas-cular safety, 24% of safety closures for CNS projects were due to CNS liabilities, and 31% of respiratory projects failed because of lung safety (FIG. 2d). This, in part, reflects the rela-tive distribution of targets between different disease areas but also illustrates the difficul-ties in predicting the risk–benefit profile of targets. This highlights the need (and oppor-tunity) to tailor safety assessment packages to different therapy areas.

Although considerable improvements have been made in the use of preclinical safety studies to assess and reduce the risk of subsequent failure, we are still a long way




from being able to predict the complete safety profile of a molecule before clinical testing44. It is therefore crucial to at least have robust predictions of efficacious exposures, which can then be explored in clinical trials relative to the safety exposure margins (see the case studies in BOXES 1,2). Finally, it is important to continue to develop predictive safety biomarkers to provide early insight into potential risks.

Right patient: testing drugs in the correct patient population. In our analysis, for pro-jects that failed in Phase II there was often low confidence that they were being tested in the correct patient population; projects that had clear patient stratification plans demonstrated a greater likelihood of suc-cess (FIG. 4a). We believe that confidence in identifying and studying the ‘right patient’ population is another crucial success factor that affects not only the probability of suc-cessful transition to later-stage trials but also, more importantly, the ability to demonstrate clinical benefit.

Ideally, patients can be selected through-out clinical development using a predictive biomarker. This requires that the target biology is sufficiently well understood, as was the case for the poly(ADP-ribose) poly-merase (PARP) inhibitor olaparib, which targets tumour cells with defects in DNA repair, such as the BRCA1 (breast cancer sus-ceptibility 1) and BRCA2 genes. Preclinical experiments showed that cell lines that are deficient in BRCA1 or BRCA2 are extremely sensitive to PARP inhibition45. Olaparib showed encouraging results in Phase I trials, in which the presence of BRCA1 and/or BRCA2 germline mutations was used to enrich patients for study46. Phase II studies in serous ovarian cancer, however, did not select patients on the basis of mutations in BRCA1 or BRCA2 (as this was initially viewed as too narrow a patient population) and the results were disappointing. A pre-planned subgroup analysis, however, suggested that olaparib may lead to greater clinical benefit in patients with BRCA1 or BRCA2 germline mutations; these patients had the greatest progression-free survival benefit versus placebo (median: 11.2 months versus 4.1 months; hazard ratio (HR): 0.17; P <0.001)47. The drug has recently entered Phase III trials in such patients.

Although oncology currently leads the way in patient stratification, there are similar examples in other disease areas. For example, in Alzheimer’s disease, Jack et al.48 have proposed a model that utilizes a series of imaging (adjusted hippocampal volume)

and biochemical biomarkers (changes in levels of tau and amyloid-β in the cerebral spinal fluid) to identify patients at different stages of the disease. In addition, the apo-lipoprotein E4 (APOE4) variant has been associated with an increased risk of devel-oping sporadic Alzheimer’s disease49. These insights allow the use of biomarkers to identify and stratify patients with Alzheimer’s disease during drug develop-ment. The importance of such biomarkers is highlighted by the recent Phase III failure of bapineuzumab, a monoclonal antibody targeting amyloid-β, for which retrospec-tive analysis of patients undergoing amyloid PET scans indicated that at least one-third of the APOE4 non-carriers fell below the pre-determined threshold for amyloid positivity. Similar results were reported for studies with solenezumab, which suggests that a substantial number of patients in both of these Phase III programmes were probably not suffering from Alzheimer’s disease (for further information, see the abstracts from the 5th Conference of the Clinical Trials on Alzheimer’s Disease (CTAD)).

For diseases in which the understanding of target biology is not sufficiently well advanced to identify candidate biomarkers based on the proposed mechanism of action of the drug, another approach is to base patient selection on known markers of disease stratification. For example, new approaches are being developed to treat patients with severe asthma that is not effec-tively controlled with standard therapies such as inhaled long-acting β-adrenergic receptor agonists and corticosteroids. Central to these approaches is the under-standing that severe asthma is a hetero-geneous mixture of syndromes in which various clinical, physiological and inflamma-tory markers determine disease severity and progression50. Periostin, whose expression is induced by interleukin-13 (IL-13), is an example of a candidate marker for patients who do not respond to corticosteroids and who may respond to anti-IL-13 therapy51–54. Clinical trials of the IL-13-specific antibody lebrikizumab in patients with asthma who are inadequately controlled on inhaled corticosteroids showed that lebrikizumab treatment was associated with improved lung function as measured by the forced expiratory volume in 1 second (FEV1)

55,56. Patients with high pre-treatment levels of serum periostin had greater improvement in lung function with lebrikizumab than did patients with low periostin levels; T helper 2 (TH2) status and fractional exhaled nitric oxide

also provided good response separation55,56. Trials of monoclonal antibodies that target the IL-5 pathway (for example, the IL-5-specific mepolizumab and the IL-5 receptor (IL-5R)-specific benrali-zumab) have used eosinophil counts to stratify reponse57,58. Mepolizumab signifi-cantly lowered eosinophil numbers59 and demonstrated a reduction in the rate of exacerbations, which varied according to the baseline blood eosinophil count57.

In all of these examples, the development programme benefited (or could have ben-efited) from the identification of appropriate clinical biomarkers for the selection of the ‘right patients’ in clinical development. To achieve success, teams should plan and implement strategies for patient selection as early as possible, and this should be an integral part of the R&D programme.

Right commercial potential: ensuring align-ment of scientific opportunity with commercial insight and position. The final theme identi-fied in our analysis was an understanding of the commercial potential of a project and how this was used to guide the project’s development. We use the term ‘right com-mercial potential’ to describe confidence that a project would ultimately deliver a medically differentiated and commercially viable product. Confidence in the ‘right commercial potential’ captures many of the features used to define confidence in the other project dimensions because it is largely a description of the desired profile of the target product. For example, an understand-ing of current and future standards of care dictates what is required for a new medicine with respect to predicted efficacy and safety in the right patient population.

In our analysis, we found that for many projects in the early-stage portfolio there was low confidence in the ‘right commercial potential’ dimension either because these projects were not competitive with respect to other drugs being developed or because they were not clearly differentiated in terms of defining a clear unmet medical need. In these instances, projects were driven by a ‘volume-based’ culture (see below for further discussion), and termination of the projects was straightforward.

Interestingly, for projects in the preclini-cal and Phase I stages, the teams had a high level of confidence that they were pursuing the right target in the right patient popula-tion, whereas confidence in the commercial potential was very low. By contrast, as pro-jects progressed into Phase II, confidence that the team was pursuing the right target




http://www.ctad.fr/07-download/Congres2012/Abstracts.pdf

Nature Reviews | Drug DiscoverySu

cces

s ra

te (%

)

0

10

20

30

40

50

60

70

80

Preclinical

All candidate drugsCD1CD2+

64 71 58

Phase I

56 56 62

Phase II

42 44 40

in the right patient population markedly decreased, whereas confidence in the commercial potential markedly increased (FIG. 4b). The reason for this was largely due to the fact that the projects were pushed to investigate what were perceived as higher-value, more commercially attractive indica-tions, but the scientific and medical basis for these indications was weak. Three-quarters of these projects failed owing to efficacy issues later in development.

Ultimately, the ‘right commercial potential’ dimension emphasizes the need to deliver a differentiated medicine. It covers a number of important elements, including the market opportunity based on the unmet medical need, the size and geographical distribution of the patient population, the competitive position of the programme, the reimbursement environment, the cost to reach specific decision points, the cost to deliver a dossier for regulatory submission for a medicine that is differentiated and reimbursable, the time to launch and, of course, the overall programme risk and probability of technical and regulatory success. Evaluation of these factors requires a clear dialogue between the commercial and scientific groups in the company that must always be underpinned by scientific understanding and medical need.

Right culture: promoting truth-seeking rather than progression-seeking behaviour. Throughout our review and analyses, a key contributor to poor pipeline outcomes appears to be the fact that team behaviour and decision-making were encouraged and rewarded by volume-based goals. As discussed earlier, many companies imple-mented research goals that were focused on volume, such as the number of candidates or numbers of INDs delivered into clinical development. In practice, this resulted in projects that focused on candidate mile-stone delivery and placed less emphasis on understanding the target biology, on the pathophysiology of the disease, on the right patient population and on maintaining a clear vision of what was needed to deliver a differentiated (and ultimately reimbursable) medicine. Volume-based goals valued and rewarded candidate transitions from phase to phase, placing less focus on truth-seeking behaviour or on asking the ‘killer questions’ to determine the validity of a hypothesis or the likelihood of delivering a differentiated medicine.

In our analysis, another example of the impact of volume-based goals could be seen in the strategy used to select back-up drug

candidates. Back-up molecules are often developed for important projects where biological confidence is high. They should be structurally diverse to mitigate the risk for the programme against compound-related issues in preclinical or early devel-opment, and/or they should confer some substantial advantage over the lead mol-ecule. When used well, this strategy can save time and maintain the momentum of a project. However, with scientists being rewarded for the numbers of candidates coming out of the research organization, we observed multiple projects for which back-up molecules were not structurally diverse or a substantial improvement over the lead molecule. Although all back-up candidates met the chemical criteria for progression into clinical testing, and research teams were considered to have met their volume-based goals, these mol-ecules did not contribute to the de-risking of a programme or increase project suc-cess rates. As a consequence, all back-up candidates from a ‘compound family’ could end up failing for the same reason as the lead compound and indeed had no higher probability of a successful outcome than the original lead molecule (FIG. 6). In one extreme case, we identified a project with seven back-up molecules in the family, all of which were regarded as a successful candidate delivery yet they all failed owing to the same preclinical toxicology finding. This overuse of back-up compounds resulted in a highly disproportionate number of back-up candidates in the portfolio. At the time of writing, approxi-mately 50% of the AstraZeneca portfolio was composed of back-up molecules.

Conclusions: what AstraZeneca did nextThe analysis we describe here was performed 3 years ago, and since then AstraZeneca has undergone considerable changes to implement the lessons learned. Although we developed the ‘5R’ framework as part of a retrospective analysis to understand the reasons for our past portfolio performance, we have also adopted it as a core part of our project framework and operating model. This has meant a conscious shift away from the high-volume-based strategy previously used in R&D to one where we focus on pro-ject quality and depth of understanding as a key driver of success.

The immediate impact of adopting the 5R framework was a substantial reduction in the size and shape of the research and early development portfolio. For the remaining projects or those that are due to commence, each project team is now expected to clearly articulate its confidence in each component of the 5R framework, to identify gaps in knowledge and understanding, and to work to fill those gaps during the duration of the programme. This, in turn, has driven better decision-making and transparency with regard to the strengths and risks of projects.

In addition to adopting the framework, we have made substantial investments in certain key capabilities. For example, we have built a capability in personalized health-care and biomarker support to help in both defining and delivering data in the ‘right patient’ dimension. As a consequence, 85% of all projects now include a personal-ized health-care strategy, and our initial analysis shows that projects with a prospec-tive personalized health-care approach (for example, a companion diagnostic) are four

Figure 6 | Analysis of back‑up candidate drugs. The success rate for the first candidate drug nominated against a specific target (CD1) was compared to that of the back-up or follow-up candidate (CD2+) drugs against the same target. The percentage of successful project transitions to the next phase is shown for each of the indicated phases for all candidate drugs, CD1 and CD2+. Percentages are shown inside the respective bars.




times more likely to be successful (data not shown). In addition, we have implemented a ‘model-based drug discovery’ platform across AstraZeneca that uses state-of-the-art modelling and simulation approaches to better understand and predict the PK/PD properties of molecules and enhance our understanding not just of the ‘right tissue’ dimensions but also of the ‘right target’ and ‘right safety’60. Finally, our understanding of the payer and reimbursement environment has been improved by the creation of a real-world payer evidence capability, which has consequently improved our understanding of the ‘right commercial potential’ within the 5R framework.

Ultimately, the 5R framework is not about having total confidence in every dimension; by definition, this can only be achieved once a project successfully reaches the market. However, by using this framework, teams can identify critical areas of risk that need to be addressed dur-ing development. They can provide a more objective assessment of a project and have a more transparent discussion with govern-ance bodies. The 5Rs capture what might be considered the key technical dimensions in a project and can help in supporting teams to make the best decision at the right stage. However, teams also need to be working in the ‘right culture’. It is vital to ensure that teams are encouraged and rewarded to ask the ‘killer question’, are recognized for the quality of their science, and are well con-nected to the external scientific community and supported by experienced leaders with a record of good judgment8. An underlying theme that ran through the interviews with our project teams was how the need to main-tain portfolio volume led to individual and team rewards being tied to project progres-sion rather than ‘truth-seeking’ behaviour. The scientists and clinicians within the pro-ject teams need to believe that their personal success and careers are not intrinsically linked to project progression but to scientific quality, smart risk-taking and good decision-making. In this context, the 5R framework promotes more informed decision-making, enabling either the early termination of projects based on clearly defined go/no-go criteria or further investment in those pro-jects that demonstrate the attributes and clinical signals that lead to increased confi-dence and probability that a project will successfully deliver a differentiated medicine.

Although it is too early to see tangible benefits to the AstraZeneca pipeline, since the implementation of the 5R framework we have seen directional improvements in

the quality and success rates of our current cohort of programmes. Our ambition is to achieve an overall success rate of at least 8% from preclinical GLP toxicology to launch. In the preclinical to Phase I transition, our success rates have risen to over 75% com-pared to 66% in our original analysis, which suggests that more of our molecules are successfully clearing GLP toxicology and entering clinical testing. By contrast, our Phase I success rates have come down to a level closer to the industry benchmarks of 40–50%, which suggests that more rigor-ous hurdles are being set for transition into Phase II; and in Phase II our success rates are moving upwards from 15% in the original cohort to over 20% in our current cohort of programmes. Of course, monitoring the launch of innovative new medicines over the coming years will be the only way to determine whether our strategy is successful overall. Clearly, the 5R framework is still in its infancy within AstraZeneca, but although it is too early to draw any conclusions about successes or failures, it is encouraging that our project success rates seem to be moving in the right direction.

David Cook, Dearg Brown, Ruth March, Paul Morgan, Gemma Satterthwaite and Menelas N. Pangalos

are at AstraZeneca, Innovative Medicines and Early Development, Alderley Park, Macclesfield

SK10 4TG, UK.

Robert Alexander is at AstraZeneca, Innovative Medicines & Early Development, 141 Portland Street, 10th Floor Cambridge, Massachusetts 02139, USA.

Correspondence to M.N.P. e‑mail: [email protected]

doi:10.1038/nrd4309 Published online 16 May 2014

1. Pammolli, F., Magazzini, L. & Riccaboni, M. The productivity crisis in pharmaceutical R&D. Nature Rev. Drug Discov. 10, 428–438 (2011).

2. Paul, S. M. et al. How to improve R&D productivity: the pharmaceutical industry’s grand challenge. Nature Rev. Drug Discov. 9, 203–214 (2010).

3. Herper, M. The truly staggering cost of inventing new drugs. Forbes [online], http://www.forbes.com/sites/matthewherper/2012/02/10/the-truly-staggering-cost-of-inventing-new-drugs/ (10 Feb 2012).

4. Munros, B. Lessons from 60 years of pharmaceutical innovation. Nature Rev. Drug Discov. 8, 959–968 (2009).

5. Scannell, J. W., Blanckley, A., Boldon, H. & Warrington, B. Diagnosing the decline in pharmaceutical R&D efficiency. Nature Rev. Drug Discov. 11, 191–200 (2012).

6. Handen, J. S. The industrialization of drug discovery. Drug Discov. Today 7, 83–85 (2002).

7. Linder, M. D. Clinical attrition due to biased preclinical assessments of potential efficacy. Pharmacol. Ther. 115, 148–175 (2007).

8. Ringel, M., Tollman, P., Hersch, G. & Schulze, U. Does size matter in R&D productivity? If not, what does? Nature Rev. Drug Discov. 12, 901–902 (2013).

9. Arrowsmith, J. & Miller, P. Trial Watch: Phase II and Phase III attrition rates 2011–2012. Nature Rev. Drug Discov. 12, 569 (2013).

10. Redfern, W. et al. Impact and frequency of different toxicities throughout the pharmaceutical life cycle. The Toxicologist 114, 231 (2010).

11. Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates? Nature Rev. Drug Discov. 3, 711–716 (2004).

12. Picker, S. M. In-vitro assessment of platelet function. Transfus. Apher. Sci. 44, 305–319 (2011).

13. Seok, J. et al. Genomic responses in mouse models poorly mimic human inflammatory diseases. Proc. Natl Acad. Sci. USA 110, 3507–3512 (2013).

14. de Jong, M. & Maina, T. Of mice and humans: are they the same? — Implications in cancer translational research. J. Nucl. Med. 51, 501–504 (2010).

15. Wendler, A. & Wehling, M. The translatability of animal models for clinical development: biomarkers and disease models. Curr. Opin. Pharmacol. 10, 601–606 (2010).

16. Jucker, M. The benefits and limitations of animal models for translational research in neurodegenerative diseases. Nature Med. 16, 1210–1214 (2010).

17. Pangalos, M. N., Schechter, L. E. & Hurko, O. Drug development for CNS disorders: strategies for balancing risk and reducing attrition. Nature Rev. Drug Discov. 6, 521–532 (2007).

18. Prahalad, S. Negative association between the chemokine receptor CCR5-Δ32 polymorphism and rheumatoid arthritis: a meta-analysis. Genes Immun. 7, 264–268 (2006).

19. Vierboom, M. P. et al. Inhibition of the development of collagen-induced arthritis in rhesus monkeys by a small molecular weight antagonist of CCR5. Arthritis Rheum. 52, 627–636 (2005).

20. Okamoto, H. & Kamatani, N. CCR-5 antagonist inhibits the development of adjuvant arthritis in rats. Rheumatology 45, 230–232 (2006).

21. Fleishaker, D. L. et al. Maraviroc, a chemokine receptor-5 antagonist, fails to demonstrate efficacy in the treatment of patients with rheumatoid arthritis in a randomized, double-blind placebo-controlled trial. Arthritis Res. Ther. 14, R11 (2012).

22. van Kuijk, A. W. et al. CCR5 blockade in rheumatoid arthritis: a randomised, double-blind, placebo-controlled clinical trial. Ann. Rheum. Dis. 69, 2013–2016 (2010).

23. Gerlag, D. M. et al. Preclinical and clinical investigation of a CCR5 antagonist, AZD5672, in patients with rheumatoid arthritis receiving methotrexate. Arthritis Rheum. 62, 3154–3160 (2010).

24. Wehling, M. Assessing the translatability of drug projects: what needs to be scored to predict success? Nature Rev. Drug Discov. 8, 541–546 (2009).

25. Morgan, P. et al. Can the flow of medicines be improved? Fundamental pharmacokinetic and pharmacological principles toward improving Phase II survival. Drug Discov. Today 17, 419–424 (2012).

26. Berman, R. M. et al. Antidepressant effects of ketamine in depressed patients. Biol. Psychiatry 47, 351–354 (2000).

27. aan het Rot, M. et al. Safety and efficacy of repeated-dose intravenous ketamine for treatment-resistant depression. Biol. Psychiatry 67, 139–145 (2010).

28. Quirk, M. et al. Abstract P-09-045. Effects of low-trapping NMDA channel blocker AZD6765 on gamma-band EEG and psychotomimetic liability: a comparison to ketamine in freely behaving rats. Int. J. Neuropsychopharmacol. 15 (Suppl. S1), 153 (2012).

29. Skolnick, P., Popik, P. & Trullas, R. Glutamate-based antidepressants: 20 years on. Trends Pharmacol. Sci. 30, 563–569 (2009).

30. Mealing, G., Lanthorn, T. H., Murray, C. L., Small, D. L. & Morley, P. Differences in degree of trapping of low-affinity uncompetitve N-methyl-d-aspartic acid receptor antagonists with similar kinetics of block. J. Pharmacol. Exp. Ther. 288, 204–210 (1999).

31. Zarate, C. A. et al. Replication of ketamine’s antidepressant efficacy in bipolar depression: a randomized controlled add-on trial. Biol. Psychiatry 71, 939–946 (2012).

32. Lip, G. Y. et al. Oral direct thrombin inhibitor AZD0837 for the prevention of stroke and systemic embolism in patients with non-valvular atrial fibrillation: a randomised doseguiding, safety and tolerability study of four doses of AZD0837 versus vitamin K antagonists. Eur. Heart J. 30, 2897–2907 (2009).

33. Olsson, S. B. et al. Safety and tolerability of an immediate-release formulation of the oral direct thrombin inhibitor AZD0837 in the prevention of stroke and systemic embolism in patients with atrial fibrillation. Thromb. Haemost. 103, 604–612 (2010).




mailto:Mene.Pangalos%40astrazeneca.com?subject=

http://www.forbes.com/sites/matthewherper/2012/02/10/the-truly-staggering-cost-of-inventing-new-drugs/



34. Pehrsson, S., Johansson K., Kjaer, M. & Elg, M. Evaluation of AR-H067637, the active metabolite of the new direct thrombin inhibitor AZD0837, in models of venous and arterial thrombosis and bleeding in anaesthetised rats. Thromb. Haemost. 104, 1242–1249 (2010).

35. Wolzt, M. et al. Effect on perfusion chamber thrombus size in patients with atrial fibrillation during anticoagulant treatment with oral direct thrombin inhibitors, AZD0837 or ximelagatran, or with vitamin K antagonists. Thromb. Res. 129, e83–e91 (2012).

36. Olsson, S. et al. Stroke preventions with the oral direct thrombin inhibitor ximelagatran compared with warfarin in patients with non-valvular atrial fibrillation (SPORTIF III): randomised clinical trial. Lancet 362, 1691–1698 (2003).

37. SPORTIF Executive Steering Committee for the SPORTIF V Investigators. Ximelagatran versus warfarin for stroke prevention in patients with nonvalvular atrial fibrillation: a randomized trial. JAMA 293, 690–698 (2005).

38. Uppoor, R. S. et al. The use of imaging in the early development of neuropharmacological drugs: a survey of approved NDAs. Clin. Pharmacol. Ther. 84, 69–74 (2008).

39. Freedman, N. M. et al. In vivo measurement of brain monoamine oxidase B occupancy by rasagiline, using 11C-l-Deprenyl and PET. J. Nucl. Med. 46, 1618–1624 (2005).

40. Thebault, J. J., Guillaume, M. & Levy, R. Tolerability, safety and pharmacodynamics and pharmacokinetics of rasagiline: a potent, selective and irreversible monoamine oxidase type B inhibitor. Pharmacotherapy 24, 1295–1305 (2004).

41. Ramsey, S. J., Attkins, N. J., Fish, R. & van der Graaf, P. H. Quantitative pharmacological analysis of antagonist binding kinetics at CRF1 receptors in vitro and in vivo. Br. J. Pharmacol. 164, 992–1007 (2011).

42. Qureshi, Z. P., Seoane-Vazquez, E., Rodriguez-Monguio, R., Stevenson, K. B. & Szeinbach, S. L. Market withdrawal of new molecular entities approved in the United States from 1980 to 2009. Pharmacoepidemiol. Drug Safety 20, 772–777 (2011).

43. King, A. Prevention: neuropsychiatric adverse effects signal the end of the line for rimonabant. Nature Rev. Cardiol. 7, 602 (2010).

44. Tralau, T. & Luch, A. Drug-mediated toxicity: illuminating the ‘bad’ in the test tube by means of cellular assays? Trends Pharmacol. Sci. 33, 353–364 (2012).

45. Farmer, H. et al. Targeting the DNA repair defect in BRCA mutant cells as a therapeutic strategy. Nature. 434, 917–921 (2005).

46. Fong, P. C. et al. Inhibition of poly(ADP-ribose) polymerase in tumors from BRCA mutation carriers. N. Engl. J. Med. 361, 123–134 (2009).

47. Ledermann, J. et al. Olaparib maintenance therapy in patients with platinum-sensitive relapsed serous ovarian cancer (SOC) and a BRCA mutation (BRCAm). J. Clin. Oncol. Abstr. 31, 5505 (2013).

48. Jack, C. R. et al. Evidence for ordering of Alzheimer disease biomarkers. JAMA Neurol. 68, 1526–1535 (2011).

49. Poirier, J. et al. Apolipoprotein E polymorphism and Alzheimer’s disease. Lancet. 2, 697–699 (1993).

50. Holgate, S. Pathophysiology of asthma: what has our current understanding taught us about new therapeutic approaches? J. Allergy Clin. Immunol. 128, 495–505 (2011).

51. Izuhara, K. et al. [Clarification of the pathogenesis and development of clinical examination for allergic disease]. Rinsho Byori 55, 369–374 (in Japanese) (2007).

52. Izuhara, K. & Saito, H. Microarray-based identification of novel biomarkers in asthma. Allergol. Int. 55, 361–367 (2006).

53. Izuhara, K. et al. IL-13: a promising therapeutic target for bronchial asthma. Curr. Med. Chem. 13, 2291–2298 (2006).

54. Woodruff, P. G. et al. Genome-wide profiling identifies epithelial cell genes associated with asthma and with treatment response to corticosteroids. Proc. Natl Acad. Sci. USA. 104, 15858–15863 (2007).

55. Corren, J. et al. Lebrikizumab treatment in adults with asthma. N. Engl. J. Med. 365, 1088–1098 (2011).

56. Thomson, N. C., Patel, M. & Smith, A. D. Lebrikizumab in the personalized management of asthma. Biologics 6, 329–335 (2012).

57. Wenzel, S. E. et al. Evidence that severe asthma can be divided pathologically into two inflammatory subtypes with distinct physiologic and clinical characteristics. Am. J. Respir. Crit. Care Med. 160, 1001–1008 (1999).

58. Pavord, I. D. et al. Mepolizumab for severe eosinophilic asthma (DREAM): a multicentre, double-blind, placebo-controlled trial. Lancet 380, 651–659 (2012).

59. Leckie, M. J. et al. Effects of an interleukin-5 blocking monoclonal antibody on eosinophils, airway hyper-responsiveness, and the late asthmatic response. Lancet 356, 2144–2148 (2000).

60. Visser, S. et al. Model-based drug discovery: implementation and impact. Drug Discov. Today 18, 746–775 (2013).

61. Lisman, J. E., Raghavachari, S. & Tsien, R. W. The sequence of events that underlie quantal transmission at central glutamatergic synapses. Nature Rev. Neurosci. 8, 597–609 (2007).

62. Moghaddam, B. & Javitt, D. From revolution to evolution: the glutamate hypothesis of schizophrenia and its implication for treatment. Neuropsychopharmacology 37, 4–15 (2012).

63. Patil, S. T. et al. Activation of mGlu2/3 receptors as a new approach to treat schizophrenia: a randomized Phase 2 clinical trial. Nature Med. 13, 1102–1107 (2007).

64. Geyer, M. A. & Moghaddam, B. in Neuropsycho‑pharmacology: The Fifth Generation of Progress (eds Davies, K., Charney, D., Coyle, J. T. & Nemeroff, C.) 689–701 (Lippincott Williams & Wilkins, 2002).

65. Matsuoka, T. et al. Prostaglandin D2 as a mediator of allergic astma. Science 287, 2013–2017 (2000).

66. Arimura, A. et al. Prevention of allergic inflammation by a novel prostaglandin receptor antagonist, S-5751. J. Pharmacol. Exp. Ther. 298, 411–419 (2001).

67. Shichijo, M. et al. A prostaglandin D2 receptor antagonist modifies experiemental asthma in sheep. Clin. Exp. Allergy 39, 1404–1414 (2009).

68. Lukacs, N. W. et al. CRTH2 antagonism significantly ameliorates airway hyperreactivity and downregulates inflammation-induced genes in a mouse model of airway inflammation. Am. J. Physiol. Lung Cell. Mol. Physiol. 295, L767–L779 (2008).

69. Gervais, F. G. et al. Pharmacological characterization of MK-7246, a potent and selective CRTH2 (chemoattractant receptor-homologous molecule expressed on T-helper type 2 cells) antagonist. Mol. Pharmacol. 79, 69–76 (2011).

70. Barnes, N. et al. A randomized, double-blind, placebo-controlled strudy of the CRTH2 antagonist OC0000459 in moderate persistent asthma. Clin. Exp. Allergy. 42, 38–48 (2012).

71. Busse, W. W. et al. Safety and efficacy of the prostaglandin D2 receptor antagonist AMG 853 in asthmatic patients. J. Allergy Clin. Immunol. 131, 339–345 (2013).

AcknowledgementsThe authors would like to thank S. Curran, J. Curwen, U. Eriksson, G. Johnston, R. Maciewicz, J. Munroe, M. Needham, P. Newham, W. Redfern, M. Snowden, F. Tierney, J.-P. Valentin and J. Waterton for their help in producing and reviewing this manuscript.

Competing interests statementThe authors declare competing interests: see Web version for details.

FURTHER INFORMATIONClinical Trials on Alzheimer’s Disease (CTAD) 5th Conference (Grimaldi Forum, Convention Center, Monte Carlo; 29–31 October 2012): http://www.ctad.fr/07-download/Congres2012/Abstracts.pdfLilly website — 29 August 2012 press release (‘Lilly Stops Phase III Development of Pomaglumetad Methionil For the Treatment of Schizophrenia Based on Efficacy Results’): https://investor.lilly.com/releasedetail.cfm?ReleaseID=703018Pharmaceutical Benchmarking Forum — KMR Group website: http://kmrgroup.com/ForumsPharma.html

ALL LINKS ARE ACTIVE IN THE ONLINE PDF




http://www.nature.com/nrd/journal/v13/n6/full/nrd4309.html#affil-auth

lessons learned from the fate of astrazeneca's drug pipeline: a...

Documents