F E AT U R E
`
`Clinical development success rates for
`investigational drugs
`Michael Hay, David W Thomas, John L Craighead, Celia Economides & Jesse Rosenthal
`
`The most comprehensive survey of clinical success rates across the drug industry to date shows productivity may be
`even lower than previous estimates.
`
`Since the human genome was sequenced
`
`ten years ago, the number of compounds
`in development has increased 62% and total
`R&D expenditures have doubled1–3. And yet,
`the average number of new drugs approved
`by the US Food and Drug Administration
`(FDA) per year has declined since the 1990s.
`In 2012, 39 novel drugs classified as new
`molecular entities (NMEs) and biologic
`license applications (BLAs) were approved
`by the FDA4. Although this represents the
`highest number of approvals since 1997 and
`is nearly 50% above the average of 26 approv-
`als per year over the past decade, 25% fewer
`NME and BLA drugs were approved on aver-
`age in the past 10 years compared with the
`1990s5. Several possible explanations for the
`divergence of R&D spending and new product
`approvals have been offered by professionals
`in the industry, such as unbalanced regulatory
`risk-benefit assessments, higher regulatory
`efficacy hurdles, commercial and financial
`decisions driving project termination, and
`the increased complexity and cost of clinical
`trials6,7.
`This article aims to measure clinical devel-
`opment success rates across the industry with
`a view to strengthening benchmarking met-
`rics for drug development. The study is the
`largest and most recent of its kind, examining
`success rates of 835 drug developers, includ-
`ing biotech companies as well as specialty and
`
`Michael Hay and Jesse Rosenthal are at
`BioMedTracker, Sagient Research Systems, San
`Diego, California, USA; David W. Thomas
`and Celia Economides are at the Biotechnology
`Industry Organization (BIO), Washington,
`DC, USA; and John L. Craighead is at Biotech
`Strategy & Analytics, Rockville, Maryland, USA.
`e-mail: mhay@sagientresearch.com
`
`large pharmaceutical firms from 2003 to 2011.
`Success rates for over 7,300 independent drug
`development paths are analyzed by clinical
`phase, molecule type, disease area and lead
`versus nonlead indication status.
`Our results pinpoint weaknesses along the
`capital-intensive pathway to drug approval.
`Our hope is that they will prove useful in
`informing policy makers where to focus
`changes in regulation and strengthen valua-
`tion models used by industry and the invest-
`ment community.
`
`Analyzing success
`To measure clinical development success rates
`for investigational drugs, we analyzed phase
`transitions from January 1, 2003 to December
`31, 2011, in the BioMedTracker database. The
`BioMedTracker data set contained 4,451 drugs
`with 7,372 independent clinical development
`paths from 835 companies and included 5,820
`phase transitions. The development paths
`comprised lead (primary) and nonlead (sec-
`ondary) indications, with roughly 38% desig-
`nated as nonlead. A more detailed description
`of the data collection, composition and analy-
`sis methodology is described in Boxes 1–3 (see
`also Tables 1 and 2).
`Unlike many previous studies that reported
`clinical development success rates for large
`pharmaceutical companies, this study pro-
`vides a benchmark for the broader drug devel-
`opment industry by including small public and
`private biotech companies and specialty phar-
`maceutical firms. The aim is to incorporate
`data from a wider range of clinical develop-
`ment organizations, as well as drug modalities
`and targets. Two landmark publications on the
`subject, DiMasi et al.6 and Kola et al.8 use 50
`and 10 pharmaceutical company pipelines,
`respectively, to arrive at their conclusions. An
`important study published by the US Federal
`
`Trade Commission Bureau of Economics,
`Abrantes-Metz et al.9 covered a wide num-
`ber of drugs over a 14 year period from 1989
`to 2002, but did not provide the number or
`type of companies investigated. Although the
`impact of company size and experience on
`R&D productivity has been studied exten-
`sively10–13, success rates established by DiMasi
`et al.6, Kola et al.8 and Abrantes-Metz et al.9
`remain the primary benchmarks for the drug
`development industry.
`We believe it is of great value to report
`updated success rates that capture the diver-
`sity in drug development sponsor types as
`experience and technology vary widely out-
`side of traditional, large pharmaceutical cor-
`porations. Furthermore, the more recent time
`frame for this study provides insight into the
`latest industry productivity. A comparison of
`previously published reports with the current
`study is summarized in Table 3 and is dis-
`cussed below.
`One key distinction of the study pre-
`sented here is our ability to evaluate all of
`a drug’s indications to determine success
`rates. Danzon et al.12 first considered suc-
`cess rates at the indication level, recognizing
`that FDA requires clinical trial evidence to
`establish efficacy for each approved indi-
`cation. Although these authors included
`data from 1988 to 2000, an observation
`period similar to Kola et al.8 and Abrantes-
`Metz et al.9, their success rates were sig-
`nificantly higher and lacked a characteristic
`decrease in phase 2 probability reported in
`previous studies as well as here. Danzon et
`al.12 concluded that higher clinical develop-
`ment success rates resulted from the analysis
`of all indications. Even so, evidence presented
`here strongly suggests that evaluating all
`indications results in lower probabilities of
`success across all phases of drug development.
`
`40
`
`volume 32 NumBeR 1 JANuARY 2014 n ature biotechnology
`
`© 2014 Nature America, Inc. All rights reserved.
`
`npg
`
`Abraxis EX2079
`Apotex Inc. and Apotex Corp. v. Abraxis Bioscience, LLC
`IPR2018-00151; IPR2018-00152; IPR2018-00153
`
`

`

`To illustrate the importance of using all
`indications to determine success rates, con-
`sider this scenario. An antibody is developed
`in four cancer indications, and all four indi-
`cations transition successfully from phase 1
`to phase 3, but three fail in phase 3 and only
`one succeeds in gaining FDA approval. Many
`prior studies reported this as 100% success,
`whereas our study differentiates the results as
`25% success for all indications, and 100% suc-
`cess for the lead indication. Considering the
`cost and time spent on the three failed phase 3
`indications, we believe including all ‘develop-
`ment paths’ more accurately reflects success
`and R&D productivity in drug development.
`Examining individual drug indications
`allows us to answer the question: “what is the
`probability that a drug developed for a specific
`indication will reach approval?” Whereas,
`using only the lead or most advanced indi-
`cation seeks to answer the question: “what is
`the probability that a drug will reach approval
`for any indication?” This study addresses both
`questions with emphasis on the findings of
`the former. In the following sections, we pres-
`ent the results of our analysis as they relate
`to overall phase success and likelihood of
`approval (LOA; see Box 2), to the type of ther-
`apeutic modality, to the disease being treated
`and to the type of drug application (whether
`orphan or Special Protocol Assessment (SPA)
`pathways).
`
`Phase success and likelihood of approval
`We found that approximately one in ten
`(10.4%, n = 5,820) of all indication develop-
`ment paths in phase 1 were approved by FDA
`(Fig. 1 and Table 4). Examining the individual
`phase components of this compound prob-
`ability, phase I success (the number of phase 1
`drugs that successfully transitioned to phase 2
`divided by the total transitions in phase 1) was
`64.5% (n = 1,918). Success in phase 2 (32.4%,
`n = 2,268) was substantially lower than in
`phase 1, but subsequently increased in phase 3
`(60.1%, n = 975). The probability of FDA
`approval after submitting a new drug appli-
`cation (NDA) or biologic license application
`(BLA) was 83.2% (n = 659).
`Success rates for lead indication develop-
`ment paths were higher than for all indica-
`tion development paths in every phase. Lead
`indications had a LOA from phase 1 of 15.3%
`(n = 3,688).
`
`Success rates by drug classification
`Drugs in the BioMedTracker data set were
`annotated by their FDA classification: new
`molecular entity (NME), non-NME, biologic
`and vaccine. However, owing to inconsistency
`in the FDA classifications, we also used our
`
`F E AT U R E
`
`Box 1 Data collection and composition
`
`BioMedTracker, a subscription-based product of Sagient Research Systems (San Diego)
`introduced in 2002, tracks the clinical development and regulatory history of novel
`investigational drugs in the United States. Analysts with advanced degrees in the life
`sciences and medicine maintain the database using information from company press
`releases, analyst conference calls, and presentations at investor and medical meetings.
`BioMedTracker also uses other sources, including regular communication with companies
`conducting clinical trials, to ensure the accuracy and timeliness of the data.
`Data included in this study were selected using BioMedTracker’s Probability of Technical
`Success (PTS) calculator, which identified 5,820 phase transitions from January 1, 2003,
`to December 31, 2011. Transitions in all phases of development were recorded in the early
`years of observation and resulted from clinical studies initiated before 2003. The data set
`contained 4,451 drugs from 835 companies and 7,372 independent clinical development
`paths in 417 unique indications.
`The composition of these novel drug development sponsors included a wide range of
`company sizes and types (Table 1). Emerging biotech represented 85% (712) of the
`companies, whereas a small number (33) of large firms (4% of total) were responsible for
`48% (3,573) of indications and 47% (2,075) of drugs in development. Similarly, private
`firms represented 49% (412) of the companies and fewer than 20% of indications and
`drugs included in the study.
`These ownership classifications were recorded at the end of the analysis time period
`and underestimate the number of drugs and indications developed by biotech companies
`due to licensing and acquisitions during the study time frame. In addition, ownership was
`assigned to the licensee controlling and funding the majority of development. In cases
`where development and economics were shared equally, ownership was generally assigned
`to the larger organization, further contributing to the conservative estimate of drugs
`developed by small and private biotech companies. Although generic products were not
`included, generic manufacturers developing novel investigational drugs were represented.
`The study also likely tracked a larger percentage of late-stage studies as these programs
`are more often in the public domain. Even so, small biotech companies often disclose
`ongoing phase 1 studies and we would expect their substantial representation in this
`study to partially offset the under-representation of early-stage discontinuation rates.
`Only company sponsored development paths designed for FDA approval were considered;
`investigator sponsored studies and combinations with other investigational drugs were
`excluded in this analysis.
`In addition, this study analyzed development paths organized by disease area,
`biochemical composition, molecular size, FDA classification and regulatory status (SPA and
`orphan drug status). Given the increasing complexity of ownership and diversity of invention
`in the drug development industry, the study did not further classify the database on the
`discovery origin or licensing status of the drug.
`
`Table 1 Analysis of company size and type
`Companies
`Number
`Percentage
`
`Indications
`Number
`Percentage
`
`Drugs
`Percentage
`
`Number
`
`Company size
`Large pharma/biotech
`(>$5 billion sales)
`Small to mid-sized
`pharma/biotech
`($0.1 billion–
`$5 billion sales)
`Emerging biotech
`(<$0.1 billion sales)
`Total
`Company type
`Private
`Public
`Total
`
`33
`
`90
`
`712
`
`835
`
`412
`423
`835
`
`4%
`
`11%
`
`3,573
`
`1,099
`
`48%
`
`15%
`
`2,075
`
`724
`
`47%
`
`16%
`
`85%
`
`2,700
`
`37%
`
`1,652
`
`37%
`
`–
`
`7,372
`
`–
`
`4,451
`
`–
`
`49%
`51%
`–
`
`1,269
`6,103
`7,372
`
`17%
`83%
`–
`
`841
`3,601
`4,451
`
`19%
`81%
`–
`
`nature biotechnology volume 32 NumBeR 1 JANuARY 2014
`
`41
`
`© 2014 Nature America, Inc. All rights reserved.
`
`npg
`
`Abraxis EX2079
`Apotex Inc. and Apotex Corp. v. Abraxis Bioscience, LLC
`IPR2018-00151; IPR2018-00152; IPR2018-00153
`
`

`

`F E AT U R E
`
` Box 2 Metrics of success: ‘Phase Success’ and ‘Likelihood of
`Approval’
`
`There are two different types of success rates reported in this study: ‘Phase Success’ and
`‘Likelihood of Approval’ (LOA). ‘Phase Success’ is calculated as the number of drugs that
`moved from one phase to the next phase divided by the sum of the number of drugs that
`progressed to the next phase and the number of drugs that were suspended. The n value
`associated with the Phase Success represents the number of drugs that have advanced
`plus the number of drugs that have been suspended, which we label as phase transitions.
`For example, if there were 100 drugs in phase 2 development and 50 transitioned to
`phase 3, 20 were suspended and 30 remained in phase 2 development, the phase 2
`Phase Success would be 71.4% (50/70; n = 70).
`Our second metric, LOA, denotes the probability of reaching FDA approval from the
`current phase, and is also expressed as a percentage. LOA is calculated as the product
`of each Phase Success probability leading to FDA approval. The n value associated with
`LOA is the sum of the n values for each Phase Success included in the LOA calculation.
`For example, if a drug is currently in phase 2, and the Phase Success for phase 2 is 30%
`(n = 20), phase 3 is 50% (n = 10), and FDA approval is 80% (n = 5), then the LOA for
`the phase 2 drug would be 12% (30% × 50% × 80% = 12%, n = 35). This calculation is
`illustrated in Supplementary Figure 2.
`
`data to annotate drugs by their biochemi-
`cal composition (e.g., peptide, nucleic acid,
`monoclonal antibody (mAb)) and molecu-
`lar size (i.e., large and small molecules).
`For example, FDA often designates large-
`molecule biologics, such as proteins and pep-
`tides, as NMEs. Indeed, large molecules, as
`defined by the BioMedTracker biochemical
`categories, comprise 13% of the NME data set,
`making direct FDA NME to biologic classifica-
`tion comparisons somewhat imprecise. FDA’s
`biologic classification comprises a wider group
`that includes the Center for Drug Evaluation
`and Research (CDER) regulated products,
`such as antibodies, cytokines, growth fac-
`tors and enzymes, as well as the Center for
`
`a
`
`Lead indications
`
`All indications
`
`86%
`
`67% 64%
`
`68%
`
`60%
`
`39%
`
`32%
`
`b
`
`83%
`
`15.3%
`
`10.4%
`
`Biologics Evaluation and Research (CBER)
`regulated products including blood isolates,
`gene therapies and cell therapy.
`FDA’s non-NME classification often
`includes drugs with the same molecular
`properties as NMEs, but which are frequently
`reformulations or combinations of approved
`products. The majority of non-NMEs also use
`the 505(b)(2) pathway to gain FDA approval.
`Vaccines were also treated as a separate class
`in this analysis, and generic and over-the-
`counter drugs were not included. A com-
`parative analysis of FDA classifications and
`BioMedTracker categories can be found in
`Supplementary Table 1. The metrics for the
`different therapeutic modality types is pro-
`vided in Table 4.
`NMEs were found
`to have the
`low-
`est success rates in
`every phase of devel-
`opment; biologics
`had nearly twice the
`LOA from phase 1
`(14.6%, n = 1,173)
`as NMEs (7.5%, n =
`3,496) for all indi-
`cations (Table 4).
`Similar results are
`seen when the data
`are reclassified into
`l a r g e - m o l e c u l e
`(excluding
`low
`molecular weight
`chemicals and ste-
`roids) and small-
`molecule NMEs:
`13.2% (n = 1,834) and
`
`7.6% (n = 3,029), respectively. In addition, the
`LOA from phase 1 for mAbs (14.1%, n = 639),
`a good proxy for CDER-regulated biologics,
`was also consistent with these broader defini-
`tions of biologics.
`Non-NMEs had the highest LOA from
`phase 1 of 20.0% (n = 855), with success rates
`well above those of the NME and biologic
`classifications in every phase. However, many
`non-NMEs begin development in phase 2 or
`phase 3, so the actual approval rate is likely
`higher (assuming that successful phase 1 out-
`comes would contribute positively to the LOA
`from phase 1).
`When analyzing lead indications only (i.e.,
`on a per drug basis), we find similar rankings
`for NME, biologic and non-NME, but at much
`higher success rates. The LOA from phase 1
`for biologics and non-NMEs are near one in
`four and NMEs approach one in eight (12.0%,
`n = 2,124), almost twice what was found when
`all indications were considered.
`
`Success rates by disease
`We found substantial variation in success rates
`among disease, as listed in Table 5 from high-
`est to lowest LOA from phase 1. Oncology
`drugs had the lowest LOA from phase 1 at
`6.7% (n = 1,803). Drugs for the ‘other’ disease
`group, which combined allergy, gastroenterol-
`ogy, ophthalmology, dermatology, obstetrics-
`gynecology and urology indications due to
`small sample size, had the highest LOA from
`phase 1, at 18.2% (n = 720). Drugs for infec-
`tious disease and autoimmune-immunol-
`ogy groups had the next two highest LOAs
`from phase 1, at 16.7% (n = 537) and 12.7%
`(n = 549), respectively.
`On a lead indication basis, also in Table 5,
`we found that cardiovascular drugs had the
`lowest LOA from phase 1 at 8.7% (n = 318)
`and the ‘other’ disease category again had
`the highest success rate at 24.5% (n = 499).
`The largest difference between lead and all-
`indication for LOA from phase 1 was observed
`in oncology: 6.7% (n = 1,803) for lead indica-
`tion and 13.2% (n = 796) for all indications.
`Oncology drugs also had the most nonlead
`indications (56% of all development paths
`compared with 28% of non-oncology indi-
`cations) as a result of the large number of
`cancers investigated using the same drug.
`Unfortunately, in oncology, when all indi-
`cations are considered, only around 1 in
`15 drugs entering clinical development in
`phase 1 achieves FDA approval compared
`with close to 1 in 8 using the lead indication
`methodology. As noted above, the result for
`lead indications represents the most success-
`ful development path for a particular com-
`pound, thereby addressing LOA on a per drug
`
`Phase 1 to
`phase 2
`
`Phase 3 to
`Phase 2 to
`NDA/BLA
`phase 3
`Phase success
`
`NDA/BLA to
`approval
`
`LOA from
`phase 1
`
`Figure 1 Phase success and LOA rates. (a) Phase success rates for lead
`and all indications. The rates represent the probability that a drug will
`successfully advance to the next phase. (b) LOA from phase 1 for lead and
`all indications. Rates denote the probability of FDA approval for drugs in
`phase 1 development.
`
`42
`
`volume 32 NumBeR 1 JANuARY 2014 n ature biotechnology
`
`© 2014 Nature America, Inc. All rights reserved.
`
`npg
`
`Abraxis EX2079
`Apotex Inc. and Apotex Corp. v. Abraxis Bioscience, LLC
`IPR2018-00151; IPR2018-00152; IPR2018-00153
`
`

`

`Box 3 Methods used in this study
`
`F E AT U R E
`
`Table 2 Definitions of terms used in this study
`BioMedTracker term
`Description for purposes of this study
`I
`Drug is currently in phase 1
`
`I/II, II, IIb
`II/III, III
`NDA/BLA
`
`Approved, withdrawn from market,
`approved (Generic competition)
`Suspended
`Approved in Europe, Approved in
`other than US/EU, Development,
`Development outside US
`
`Data used for this study were extracted
`from BioMedTracker using a probability
`of technical success (PTS) tool, which
`identified all ‘Advanced’ and ‘Suspended’
`drugs by development phase from
`January 1, 2003, to December 31,
`2011. BioMedTracker tracks the clinical
`development and regulatory history
`of investigational drugs to assess its
`Likelihood of Approval (LOA) from phase
`1 by the FDA. The database is populated
`in near real-time with updated information
`from press releases, corporate earnings
`calls, investor and medical meetings, and
`numerous other sources. These data are
`recorded in BioMedTracker and tagged with a date.
`Phase is defined as the stage of clinical development in the
`United States (Table 2). Although it is rare, drugs that were
`removed from development in the United States, but approved
`in Europe (e.g., vildagliptin for type II diabetes) were considered
`‘suspended’ for the sake of our analysis. In this time period,
`7,372 development paths were analyzed, encompassing 4,451
`unique compounds. 5,820 unique phase transitions were used
`to determine the reported success rates. Table 4 includes the
`number of observed transitions by phase (a description of the
`success rate analysis is described). Phase 2 transitions accounted
`for the highest percentage of the data set with 39% (n = 2,268),
`compared with 33% in phase 1 (n = 1,918), 17% in phase 3 (n =
`975) and 11% in NDA/BLA
`(n = 659). Nonlead indications comprise 38% (n = 2,132) of the
`5,820 total transitions and success rates by phase can be found in
`Supplementary Table 2.
`Development paths track a specific indication for each drug. For
`example, Rituxan (rituximab) in non-Hodgkin’s lymphoma qualifies
`as a development path different from Rituxan in multiple sclerosis
`(MS). BioMedTracker assigns a unique internal identifier that can
`be used to isolate all development paths. In addition to tracking
`the phase of development, BioMedTracker assigns ‘lead’ status
`to certain development paths. This is used to denote the most
`advanced indication in clinical development for a specific drug.
`Drugs can only have one lead development path, except in specific
`circumstances where two development paths are being developed
`simultaneously (e.g., type I and type II diabetes). For example,
`the Avastin (bevacizumab) colorectal cancer development path
`was marked as a ‘lead’ indication, and other Avastin development
`paths were labeled ‘nonlead’. Using this metric, Avastin clinical
`development can more accurately be viewed as a series of
`successes and failures, as opposed to simply one success and no
`failures. However, a drug’s lead indication may also change if it
`fails in development in the lead indication. The lead indication
`success rate will therefore be higher due to selection bias than the
`nonlead success rate. This bias does not affect the LOA from
`phase 1 rate for all indication development paths.
`
`Drug is currently in phase 2
`Drug is currently in phase 3
`Application for approval has been submitted to the FDA and is
`currently under review
`Drug has been approved for marketing in the United States
`
`Drug is no longer in development
`The company developing this drug does not plan to market it in the
`United States
`
`• Disease area (e.g., autoimmune, cardiovascular, oncology)
`• Indication (e.g., diabetes, acute coronary syndrome)
`In contrast with many earlier studies, which included only a
`limited sample of drugs from large companies, the current study
`included BioMedTracker data from small biotech companies as
`well as specialty and large pharmaceutical firms.
`
`Phase success and LOA rates calculation. A common method of
`determining drug development success rates detailed in DiMasi
`et al.6 and Abrantes-Metz et al.9 was used in this study. Phase
`Success, defined as the probability of a drug moving from phase
`X to phase X + 1, was used as the basis for all analyses. To arrive
`at this value, the following questions are used to categorize each
`drug development path: first, was the drug development path
`ever in phase X? Second, if so, did it advance to phase X + 1?
`And third, was it ‘Suspended’? After categorizing all drug
`development paths, Phase Success is calculated by dividing
`the number of development paths that advanced from phase X
`to phase X + 1 by the sum of the number of development paths
`that advanced from phase X to phase X + 1 and the number
`of development paths that were suspended from phase X –
`Advanced/(Advanced + Suspended) = Phase Success.
`Using this method, we arrived at the probabilities of an
`‘average’ drug advancing from phase 1 to phase 2, from phase 2
`to phase 3, from phase 3 to filing the NDA/BLA and from filing
`the NDA/BLA to FDA approval. We then compounded these
`probabilities to determine the probability (LOA) that a drug in
`phase X is approved. For example, the LOA for a drug which
`has entered phase 2 is the product of the phase success rates
`from phase 2, phase 3 and NDA/BLA. An example calculation is
`illustrated in Supplementary Figure 2.
`For purposes of this analysis, all indications that were
`advanced or suspended in any phase during our collection
`time frame were included. In practice, this means a drug that
`‘entered’ the analysis in 2003 in phase 2, and later advanced to
`phase 3, was included in the study. This method was selected
`because there are relatively few drugs that entered development
`in phase 1 in the range of years analyzed and have subsequently
`progressed through final FDA review, and there is less disclosure
`of drugs in phase 1 development. Abrantes-Metz et al.9 also
`used a similar method and stated, “We did it this way because
`the data set has very few drugs with complete information for
`all… phases.” Drugs that remained in the same phase were
`censored, as were those that moved back a phase but were not
`suspended9.
`
`BioMedTracker also records a number of other variables including
`the following:
`• FDA classification (e.g., NME, non-NME, biologic or vaccine)
`• Biochemical profile (e.g., small molecule, monoclonal
`antibody, antisense)
`
`nature biotechnology volume 32 NumBeR 1 JANuARY 2014
`
`43
`
`© 2014 Nature America, Inc. All rights reserved.
`
`npg
`
`Abraxis EX2079
`Apotex Inc. and Apotex Corp. v. Abraxis Bioscience, LLC
`IPR2018-00151; IPR2018-00152; IPR2018-00153
`
`

`

`F E AT U R E
`
`Table 3 Comparison of our study with previous drug development success rate studies
`DiMasi et al.6 lead
`This study (2013) all
`This study (2013)
`indications
`lead indications
`indications
`Phase
`Phase
`Phase
`success
`success
`success
`64.5%
`66.5%
`71%
`32.4%
`39.5%
`45%
`60.1%
`67.6%
`64%
`83.2%
`86.4%
`93%
`
`Phase LOA
`10.4%
`16.2%
`50.0%
`83.2%
`10.4%
`
`Phase LOA
`15.3%
`23.1%
`58.4%
`86.4%
`15.3%
`
`Kola et al.8 lead
`indications
`Phase
`success
`68%
`38%
`55%
`77%
`
`Phase LOA
`11%
`16%
`42%
`77%
`11%
`
`Abrantes-Metz et al.9
`lead indications
`Phase
`success
`80.7%
`57.7%
`56.7%
`NA
`26.4%c
`
`Phase LOA
`NA
`NA
`NA
`NA
`NA
`
`Phase LOA
`19%
`27%
`60%
`93%
`19%
`
`Phase 1 to phase 2
`Phase 2 to phase 3
`Phase 3 to NDA/BLA
`NDA/BLA to approval
`LOA from phase 1a
`Number of drugs in
`sample advanced or
`suspendedb
`1989–2002
`1991–2000
`1993–2009
`Dates of source data
`(14 years)
`(10 years)
`(17 years)
`(duration)
`NA
`10
`50
`835
`Number of companies
`aProbability of FDA approval for drugs in phase 1 development. bTotal number of transitions used to calculate the success rate (the n value noted in the text). cAbrantes-Metz, et al.9 reported 26.4% from phase 1 to phase 3.
`If we were to conservatively apply the 83.2% NDA/BLA success rate found in this study, Abrantes-Metz would yield the highest LOA from phase 1 (21%). NA, data not available.
`
`5,820
`
`4,736
`
`1,316
`
`NA
`
`2,328
`
`2003–2011 (9 years)
`
`The development paths with the two low-
`est rates of phase 3 success were oncol-
`ogy and cardiovascular disease, with 45.2%
`(n = 221) and 52.8% (n = 89), respectively.
`Figure 2 also highlights the large step-up in
`success rates from phase 2 to phase 3 for auto-
`immune, endocrine and respiratory diseases,
`increasing from 34% to 68%, 34% to 67%, and
`28% to 63%, respectively. The low LOA from
`phase 1 in oncology rate results primarily from
`the lack of such a step-up, with a low phase 2
`rate of 28.3% (n = 827), followed by a phase 3
`success rate of only 45.2% (n = 221).
`
`basis. Using the lead indication methodology
`to determine success rates, the scope of the
`challenge in oncology drug development
`would be dramatically underestimated.
`The largest variation in success rates across dis-
`ease groups was observed in phase 2. In Table 5
`all-indication phase 2 success rates ranged
`from 26.3% (for cardiovascular) to 45.9% (for
`infectious disease). In phase 3, all indication
`success rates ranged from 45.2% (for oncol-
`ogy) to 71.1% (for other). In contrast, phase 1
`and NDA/BLA (As only one application, NDA
`or BLA, will be filed for any single indication,
`rates are given below for NDA/BLA.) filing
`success rates were more consistent across dis-
`ease groups. All indication data from Table 5
`are charted in Figure 2 to illustrate the large
`differences in phases 2 and 3 and LOA from
`phase 1 success rates across disease areas.
`
`versus 6.7% (n = 1,803), respectively, reducing
`the probability of FDA approval in the full data
`set from nearly one in eight to over one in ten.
`Interestingly, the LOA from phase 1 for small-
`molecule NMEs was similar for oncology (6.6%,
`n = 1,163) and non-oncology (7.9% n = 2,333)
`indications, and biologics and non-NMEs
`accounted for much of the difference. For
`example, oncology biologics had a 7.3%
`(n = 429) LOA from phase 1 compared
`with 19.4% (n = 744) for non-oncology
`biologics.
`Table 7 shows phase success and LOA rates
`in subcategories of cancer type for oncology
`drugs. Although a high number of transitions
`in all phases were seen for the solid tumor
`(n = 1,358) and hematological (n = 409) sub-
`groups, further classification of oncology indica-
`tions results in low numbers of transition from
`phase 3 to NDA/BLA. As is true of the full data
`set, drugs in phase 2 for oncology subgroups
`display more transitions and represent the
`strongest data for specific-indication success
`rate analysis. Oncology phase 2 success rates
`ranged from 50.0% (n = 12) in head and neck
`cancer to 20.9% (n = 24) in prostate cancer;
`however, the phase 2 rank order by tumor type
`was uncorrelated with LOA from phase 1 (linear
`regression, R2 = 0.26). On average, phase 2 suc-
`cess rates were higher in hematological tumors
`(34.6%, n = 179) than in solid tumors (26.3%,
`n = 636). Only two phase 3 oncology indica-
`tions had more than 20 transitions: breast cancer
`(n = 25) and non–small cell lung cancer (n = 23),
`which together accounted for ~28% of the solid
`tumor phase 3 transitions (n = 172). Because of
`even smaller sample sizes, cancer type success
`rates were not analyzed by lead indication.
`
`Success rates for neurology, autoimmune
`and endocrine disease drugs. Neurology and
`autoimmune/immunology disease groups are
`
`Success rates for oncology and non-oncology
`drugs. As oncology drugs made up the larg-
`est portion of the total data set (31.0% of all
`transitions) and had the lowest LOA from
`phase 1 (6.7%, n = 1,803), we investigated
`their contribution to
`success rates for the
`entire data set. To
`accomplish this, we
`removed all oncology
`drug development
`paths and compared
`these results to the
`full data set and
`oncology develop-
`ment paths alone.
`Table 6 shows phase
`success and LOA
`rates for drugs for all
`disease groups, oncol-
`ogy and non-oncol-
`ogy
`development
`paths. The LOA from
`phase 1 across non-
`oncology indications
`is nearly twice that
`for oncology alone,
`12.1% (n = 4,017)
`
`LOA from phase 1
`
`20
`
`18
`
`16
`
`14
`
`12
`
`10
`
`8 6 4 2 0
`
`Phase 2
`
`Phase 3
`
`
`
`LOA
`
`
`
`18%
`
`71%
`
`
`
`17%
`
`68%
`
`67%
`
`65%
`
`
`
`13%
`
`44%
`
`46%
`
`63%
`
`12%
`
`
`
`11%
`
`
`
`34%
`
`34%
`
`32%
`
`28%
`
`Other
`
`Autoim m une
`Infectious disease
`
`Endocrine
`
`Respiratory
`All diseases
`
`60%
`
`
`
`10%
`
`61%
`
`
`
`9%
`
`53%
`
`45%
`
`30%
`
`7%
`
`
`26%
`
`
`28%
`
`7%
`
`Cardiovascular
`Neurology
`
`Oncology
`
`100
`
`90
`
`80
`
`70
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`0
`
`Phase success
`
`Figure 2 Phase success and LOA from phase 1 by disease for all indications.
`The bars represent phase 2 and phase 3 success rates and the line
`represents LOA from phase 1.
`
`44
`
`volume 32 NumBeR 1 JANuARY 2014 n ature biotechnology
`
`© 2014 Nature America, Inc. All rights reserved.
`
`npg
`
`Abraxis EX2079
`Apotex Inc. and Apotex Corp. v. Abraxis Bioscience, LLC
`IPR2018-00151; IPR2018-00152; IPR2018-00153
`
`

`

`F E AT U R E
`
`Cumulative
`
`First review
`
`90%
`
`88%
`
`88%
`
`84%
`
`81%
`
`78%
`
`78%
`
`71%
`
`68%
`
`61%
`
`57%
`
`68%
`
`71%
`
`64%
`
`48%
`
`36%
`
`b
`
`100
`
`90
`
`80
`
`70
`
`60
`
`50
`
`40
`
`30
`
`20
`
`10
`
`Approvals
`
`
`•
`
`
`•
`
`
`•
`
`•
`
`
`
`•
`
`•
`
`
`•
`
`
`
`•
`
`Third
`FDA review
`86.2%
`87.0%
`78.0%
`77.6%
`
`Fourth
`
`Fifth
`
`87.3%
`88.5%
`79.1%
`79.5%
`
`87.5%
`88.8%
`79.1%
`79.5%
`
`0
`
`Infectious disease
`
`All diseases
`Respiratory
`Autoim m une
`
`O ncology
`
`Neurology
`Cardiovascular
`
`Endocrine
`
`a
`
`100
`
`90
`
`80
`
`70
`
`60
`
`50
`
`40
`
`Cumulative approval success (%)
`
`• 
` •
`
`First
`
`Second
`
`All
`Lead
`NMEs - all
`NMEs - lead
`
`56.9%
`53.3%
`55.5%
`52.2%
`
`79.0%
`78.7%
`72.4%
`72.0%
`
`Figure 3 NDA/BLA su