`
`VOLUME 58 NUMBER 1
`
`JULY 1995
`
`COMMENTARY
`success
`rates for new drugs entering
`clinical testing in the United States
`
`Joseph A. DiMasi, PhD Boston, Mars.
`
`risky, and time-
`is a complex,
`Drug development
`consuming process. Research on new chemical enti-
`ties (NCEs) that undergo clinical testing is abandoned
`without marketing approval in a substantial majority
`of cases. The extent and speed at which the develop-
`ment process makes new therapies available
`to the
`public are important measures of its viability. The cost
`of new drug development
`is also critically dependent
`on the proportion of drugs that fail in clinical test-
`ing.‘,’
`In addition,
`reliable data on industry success
`rates can serve as useful benchmarks for project plan-
`ning purposes. The investments required to move new
`drugs through development
`to marketing approval are
`substantial, and efficient use of resources requires
`careful consideration of expected costs and benefits.
`Several studies have examined clinical success rates
`(i.e.,
`the percentage of drugs tested in humans that
`
`for various periods and
`obtain marketing approval)
`with varying degrees of completeness. In a series of
`studies of new drug development
`in the United States,
`the Tufts Center for the Study of Drug Development
`(CSDD) has provided descriptive data on how cumu-
`lative success rates vary with
`time
`from
`investiga-
`tional new drug application (IND) filing.3-8 The Office
`of Technology Assessment has presented similar mea-
`sures on relatively recent data, but with less time for
`approvals to have occurred.’ The Office of Technol-
`ogy Assessment used data supplied by the Food and
`Drug Administration
`(FDA)
`(extending out
`to 54
`months after IND
`filing) to examine success rates for
`new drugs and biologics with INDs filed from 1976 to
`1978 and from 1984 to 1986. Tucker et al. lo examined
`FDA data on the development histories of new drugs
`with IND applications filed from 1976 to 1978 and es-
`timated final success rates for this group of drugs, us-
`ing
`the success rates for drugs whose
`fates were
`known
`to infer success rates for compounds whose
`fates were unknown at the time of the study.
`Statistical modeling can be helpful in analyzing suc-
`cess rates for recent periods because many of the com-
`pounds will still be in active testing at the time of the
`analysis. Cox” and Sheck et al. l2 were the first to ap-
`ply a statistical methodology
`to the problem of esti-
`mating success rates for new drugs. As part of their
`1
`
`the Center
`
`for
`
`the Study
`
`of Drug Development,
`
`Tufts Univer-
`
`23. 1995.
`Feb.
`26, 1994; accepted
`Aug.
`of
`the Study
`for
`A. DiMasi,
`PhD, Center
`Tufts University,
`192 South St., Suite 550,
`
`Book,
`
`Inc.
`
`From
`sity.
`for publication
`Received
`requests:
`Joseph
`Reprint
`Drug Development,
`Boston, MA
`02111.
`Cm
`PFMMACOL THER 1995;58:1-14.
`Copyright
`0
`1995 by Mosby-Year
`0009-9236/95/$3.00
`+ 0
`13/l/64425
`
`Page 1 of 14
`
`Biogen Exhibit 2013
`Coalition v. Biogen
`IPR2015-01136
`
`
`
`2 DiM~i
`
`CLINICAL. PHARMA CQLQGY &THERAPEUTICS
`JULY 1995
`
`the research and development
`for estimating
`method
`cost of new drug development, DiMasi et al.‘,’ used a
`statistical approach similar to that of Sheck et all2
`to
`determine final U.S. approval success rates for NCEs
`first tested in humans anywhere
`in the world during
`1970 to 1982. The studies by DiMasi et al.‘,2 also
`provided estimates of phase attrition
`rates (i.e.,
`the
`rates at which compounds drop out of active testing in
`the clinical development phases) for these compounds.
`Two studies have examined clinical success rates
`for biotechnology-derived drugs. 13,14 Bienz-Tadmor et
`all3 defined success as the submission of a product li-
`cense application
`for biotechnology
`drugs with an
`IND application
`filed from 1980 to 1988. Submission
`success rates were determined
`through a nonstatistical
`mathematical
`technique. The limited amount of time
`that biotechnology drugs could have been in testing at
`the time of analysis and the small number of biotech-
`nology drugs that have been approved precluded anal-
`ysis of approval rates. Approval success rates for bio-
`technology drugs reported as under development
`in
`Pharma projects l5 from 1983 to 1991 have been ex-
`amined by Struck.14 The approval success rates were
`built on estimates of the transitional probabilities of
`proceeding
`from one phase of development
`to the
`next. However, given that 90% of the drugs were still
`active at the time of the study, the transitional proba-
`bilities for the later development phases are much less
`reliable than those for the earlier phases. In addition,
`implicit
`in the method is the assumption that success
`rates for biotechnology drugs that entered develop-
`ment in the late 1980s and early 1990s will be the
`same as success rates for biotechnology drugs that en-
`tered development
`in the mid 1980s.
`Nevertheless, both studies predicted success rates
`that are substantially higher than those that have been
`reported
`in
`the past for
`traditional chemical com-
`pounds. The results described by Bienz-Tadmor et
`a1.13 also suggest that biotechnology drugs differed
`from chemical drugs in the pattern by which approval
`rates change with time from the start of clinical test-
`ing. Thus success rate analyses for chemical and bio-
`technology compounds should be conducted sepa-
`rately .
`The methodologic approach used in this study to
`predict success rates is similar to that used by Sheck et
`all2 However, an improved
`technique
`for predicting
`final success rates for groups of NCEs with INDs filed
`in a given period is developed here. I am also able to
`use more recent information on the fate of compounds
`from the periods studied by Sheck et al. ,12 and I am
`able
`to examine
`later periods.
`In addition, unlike
`
`Sheck et a1.,12 success rates for therapeutic classes
`and for licensed products are examined here.
`Earlier success rate studies have not considered how
`the size of a pharmaceutical
`firm may affect attrition
`rates or the ultimate success rate. Firms of different
`sizes may differ in the amount of risk they wish to as-
`sume, in their capacities to discover promising com-
`pounds, or in how effectively
`they manage their de-
`velopment efforts. A substantial number of economic
`studies have examined, with mixed results, various
`hypotheses about how firm performance
`is related to
`firm size for the pharmaceutical and other industries. l6
`Studies of this type have implications
`for
`the effi-
`ciency of various market structures, a topic of particu-
`lar relevance given the ramifications
`that a changing
`pharmaceutical marketplace and proposed health-care
`reforms may have on the structure of the U.S. phar-
`maceutical
`industry. Although a clinical success rate
`offers information on only one dimension of the per-
`formance of a pharmaceutical
`firm,
`it is undoubtedly
`an extremely important dimension.
`
`METHODS
`from a
`this study were obtained
`Data used for
`CSDD database. The CSDD database was derived
`from a survey of 36 U.S. pharmaceutical firms. These
`firms provided data on NCEs first investigated in hu-
`mans anywhere in the world or for which they were
`the first to file a U.S. IND
`from 1963 to 1989. The
`data gathered include IND
`filing dates, the dates on
`which IND research was abandoned (as of December
`31, 1989), and reasons for termination of research. A
`description of additional
`information
`included
`in this
`database is available elsewhere.* U.S. approval dates
`were obtained from public sources.“,i8 Current suc-
`cess rates for these NCEs were examined as of De-
`cember 3 1, 1993, and statistical analysis was applied
`to data on past rates of research abandonment and ap-
`proval to predict future success rates. Data on INDs
`filed in the last half of the 1980s were available but,
`given the length of the NCE development process,
`these data are too recent to use for a comprehensive
`statistical analysis of success rates. However, ob-
`served success rates through 1993 were determined for
`INDs filed during this and earlier periods.
`Inclusion criteria. For purposes of this study, an
`NCE is defined as a new molecular compound not pre-
`viously tested in humans. Excluded are new salts and
`esters of existing compounds, surgical and diagnostic
`materials, vaccines and other biologics, certain exter-
`nally used compounds (such as disinfectants, antiper-
`spirants, and sunscreens), and nutritional compounds
`
`Page 2 of 14
`
`
`
`CLINICAL
`VOLUME
`
`PHARMA COLOGY & THERAPEUTICS
`58, NUMBER
`1
`
`DiMasi
`
`3
`
`forms of vitamins and sweetening
`(such as natural
`agents). This definition of an NCE differs
`from the
`FDA’s definition of a new molecular entity, most no-
`tably in that the FDA’s definition
`includes diagnostics,
`whereas this definition of an NCE does not.
`Statistical analysis of success rates. For the statis-
`tical analyses, residence time (the length of time from
`IND
`filing to either abandonment of research without
`marketing approval or to U.S. marketing approval)
`was calculated for NCEs with INDs filed from 1964 to
`1984. Approval dates were available through Decem-
`ber 3 I, 1993, and were used in determining observed
`success rates. Residence times were calculated as of
`the end of 1989 (research termination dates were
`available comprehensively only through December 3 I,
`1989). Observed and predicted cumulative approval
`success rates were calculated at each year from IND
`filing. The study period begins at 1964 because some
`of the 1963 INDs were for NCEs on which clinical
`testing had been done in the United States before 1963
`(the INDs on these drugs were filed to meet the re-
`quirements of the 1962 Amendments of the Federal
`Food, Drug, and Cosmetic Act of 1938).
`NCEs were stratified according
`to source (self-
`originated versus licensed or otherwise acquired) and
`therapeutic class. An NCE
`is self-originated
`if the
`same firm that discovered the compound also develops
`it. Licensed NCEs are defined as compounds that were
`obtained by the developing
`firm
`through
`licensing,
`purchase, barter, or other means. The data are ana-
`lyzed by the period during which the IND was filed to
`determine whether trends in success rates exist.
`Predicted success rates for IND filing periods were
`determined
`from a two-stage model of the approval
`process. NCEs with research still active as of Decem-
`ber 3 1, 1989, constitute right-censored observations
`for our dataset. Survival analysis can make use of in-
`formation provided by censored data.” NCEs were
`assumed to survive until either research was termi-
`nated without approval or marketing approval was
`achieved. Details of the selected models and the com-
`putational approach used to estimate
`final success
`rates are provided in the Appendix.
`risky
`size. The willingness
`to undertake
`Firm
`projects or the proficiency in bringing drug candidates
`to market may vary by firm size. A commonly used
`measure of company size is the firm’s sales revenues.
`For some firms that develop new drugs, the revenues
`for pharmaceutical operations are only a small portion
`of total company sales. In the context of our analyses,
`pharmaceutical sales is a more relevant measure of
`firm size than is total company sales.
`
`Company pharmaceutical sales data are available in
`several public sources. However,
`the sources are not
`complete for a given year and are not available for all
`years. Pharmaceutical sales in 1986 for each of the
`firms in the CSDD database were obtained
`from two
`public sources.20,21 These data were used to group
`firms in the database into size classes. Approval suc-
`cess rate estimates were calculated for these classes
`for NCEs with INDs filed from 1980 to 1984.
`The early to mid 1980s was a period during which
`the structure of the pharmaceutical
`industry was rela-
`tively stable; there were few mergers or acquisitions
`that involved separate pharmaceutical
`firms. By con-
`trast, significant merger and acquisition activity oc-
`curred in the pharmaceutical
`industry during the early
`1970s and, especially, the late 1980s. The firms in the
`CSDD database that filed INDs
`from 1980 to 1984
`had not merged with or acquired other pharmaceutical
`firms by 1986. In addition,
`little, if any, of the 1986
`sales revenues would have come from
`the sale of
`NCEs with INDs filed from 1980 to 1984. Thus the
`1986 sales levels provide a reasonable basis for classi-
`fying firms by size for a success rate analysis of NCEs
`with INDs filed from 1980 to 1984.*
`The 32 firms in the CSDD database (the number
`that existed and had filed INDs
`from 1980 to 1984)
`were divided into three groups of roughly equal num-
`ber. For the analyses shown below,
`the 10 firms that
`had the highest pharmaceutical sales in 1986 were
`placed in the large-firm class, the 10 firms with the
`next highest sales levels were placed in the medium-
`sized-firm class, and the remaining 12 firms were
`placed in the small-firm class. Small, medium, and
`large firms earned less than $1.3 billion, between $1.3
`billion and $2.1 billion, and more than $2.1 billion in
`pharmaceutical sales in 1986, respectively.
`
`*Measures of the extent of a firm’s research and development op-
`erations have also been examined for relationships with measures
`of firm performance. Data on company pharmaceutical research
`and development expenditures are publicly available
`for some
`firms, but I was able to obtain such data for only about half of the
`firms in the database. Company total research and development
`figures are available for all of the firms. However, using only
`corporate research and development expenditures or mixing phar-
`maceutical research and development expenditures with corporate
`research and development expenditures prorated, for example, on
`the basis of the sales distribution, would yield unreliable mea-
`sures of the size of the research and development effort. Some
`firms have substantial nonpharmaceutical research and develop-
`ment efforts, and the proportion of corporate research and devel-
`opment devoted to drugs is probably different (larger) than the
`proportion of corporate sales that is obtained from drugs. The gap
`between these proportions is also likely to differ by firm,
`
`Page 3 of 14
`
`
`
`4 DtMasi
`
`CLINICAL
`
`PHARhfACOLOGY
`
`&THERAPEUTICS
`JULY 1995
`
`,oo ,~
`
`1964 to 1969 /ND Filings
`
`1970 to 1974 IND Filings
`
`so
`90
`70
`60
`50
`
`40
`
`'0
`0
`
`0
`
`;
`
`~
`
`"'*"'
`
`i5
`Md.=S.S
`
`,I.\.
`10
`Years
`
`20
`15
`IND Filing
`
`from
`
`25
`
`30
`
`Mdk2.6
`
`Years from IND Filing
`
`100 h
`
`1975 to 1979 IND Filings
`
`loo h
`
`1980 to 1984 IND Filings
`
`g
`a s
`u)
`E
`i&o
`
`t
`
`70
`60
`
`50
`40.
`
`20
`10
`
`o-
`0
`
`-
`
`I\
`
`(5
`Md.11.4
`
`10
`Years
`
`20
`15
`IND Flllng
`
`from
`
`25
`
`30
`
`Md.42
`
`Years
`
`from
`
`IND Flllng
`
`Fig. 1. Estimated survival curves for self-originated new chemical entities (NCEs) with a first
`investigational new drug application (IND) filed during a given period. The curves show the per-
`centage of NCEs that had not been abandoned or approved for marketing in the United States
`(i.e., still active) a given number of years from the date of IND filing. Also shown are median
`survival times. The data were fitted to Weibull distributions.
`
`RESULTS
`in the CSDD data-
`firms
`The 36 pharmaceutical
`base of investigational NCEs filed 1943 INDs
`from
`1964 to 1989. Of
`these, 1501 were
`identified as
`self-originated and 389 were identified as licensed; 53
`are of unknown source. Of the 1501 self-originated
`NCEs, 1127 were
`initially
`investigated
`in humans
`in the United States. By the end of 1993, 17.2% of
`the NCEs with
`INDs
`filed from 1964 to 1989 had
`been approved for marketing in the United States. For
`this period,
`the current U.S. approval success rates
`for NCEs
`that were
`licensed, self-originated,
`and
`self-originated
`but
`first
`tested
`in humans
`in
`the
`United States are 29.8%, 14.6%, and 10.4%, respec-
`tively.
`These results illustrate the significance of prior test-
`ing on U.S. success rates. Success rates on IND
`fil-
`ings are higher for compounds
`that were licensed or
`first tested abroad. The impact of screening on overall
`success rates varied over time. For example, the pro-
`portion of self-originated NCEs that were first tested
`
`trend.
`in the United States shows a marked downward
`Whereas 79% of self-originated NCEs with IND
`fil-
`ings from 1964 to 1984 were first tested in humans in
`the United States, only 61% of self-originated NCEs
`with INDs filed from 1985 to 1989 were first tested in
`the United States.
`Time to research termination. Even though some
`of the drugs in our database are still active, survival
`analysis can be used to establish the rates at which the
`population of NCEs with
`INDs
`filed during a given
`period will drop out of active testing. The residence
`time distributions
`for INDs
`filed from 1964 to 1969,
`1970 to 1974, 1975 to 1979, and 1980 to 1984 are
`shown in Fig. 1. The figures indicate that the median
`time to either research abandonment or marketing ap-
`proval increased by 1.8 years from the 1960s to the
`late 1970s. Median residence time decreased slightly
`for the early 1980s to 4.2 years. Although
`the data are
`too recent to develop success rate estimates for the
`late 1980s IND filings, enough time has elapsed to de-
`velop a survival curve for
`this period. The median
`
`Page 4 of 14
`
`
`
`CLINICAL
`VOLUME
`
`P HAKMACOLOGY
`58, NUMBER
`1
`
`& THERAPEUTICS
`
`DiMasi
`
`5
`
`7964-7969
`
`IND Filings
`
`1970-1974
`
`IND Filings
`
`24 i
`
`0
`
`5
`
`10
`Years
`
`20
`15
`from
`IND Filing
`
`25
`
`30
`
`24,
`
`20
`
`16
`
`12
`a
`
`4
`
`0
`
`0
`
`5
`
`10
`Years
`
`20
`15
`from
`IND Filing
`
`25
`
`30
`
`1975-1979
`
`IND Filings
`
`1980-1984
`
`/ND Filings
`
`-Predtcted
`
`*Observed
`
`24
`
`24Y
`‘0 2o
`$
`2
`
`16
`
`::
`a
`
`12
`
`0
`
`5
`
`10
`Years
`
`15
`from
`
`20
`IND Filing
`
`25
`
`30
`
`0
`
`5
`
`10
`Years
`
`20
`15
`from
`IND Filing
`
`25
`
`30
`
`Fig. 2. Predicted and observed cumulative approval success rates for self-originated NCEs with
`INDs filed during various periods. For the observed success rates, note was taken of U.S. approv-
`als through December 3 1, 1993. Predicted success rates were constructed by combining estimates
`from a survival analysis of residence time (time from IND filing to abandonment or U.S. market-
`ing approval) with a Weibull distribution specification and estimates of the conditional probability
`of approval for a given residence time from a probit specification.
`
`is also 4.2
`
`for 1985 to 1989 filings
`
`residence time
`years.
`Success rate trends. To estimate final success rates,
`results from the survival analyses are combined with
`those from qualitative choice models of
`the condi-
`tional probability of approval at given residence times.
`The parameter estimates for both stages of the model
`are highly statistically significant, and goodness-of-fit
`measures indicate strong agreement with the data. The
`parameter estimates used to determine the predicted fi-
`nal success rates reported here and the accompanying
`statistical results are available on request.
`The modeling process involves two separate statisti-
`cal procedures. Combining
`them, therefore, produces
`two primary sources of estimation error. Nonetheless,
`as shown by Fig. 2, the fits of the predicted cumula-
`tive success rate curves to curves representing the ac-
`tual experience of self-originated NCEs through 1993
`
`are tight, especially for the first three intervals. The
`predicted curve for the early 1980s slightly underesti-
`mates the observed success rates beyond 5 years from
`IND
`filing. However,
`the predicted curve
`for
`the
`1960s (a period for which there is now almost com-
`plete information) also underestimated observed suc-
`cess rates after a number of years from
`IND
`filing.
`The final predicted success rate, though, differs from
`both the observed and the maximum possible success
`rate for this filing interval by less than one percentage
`point.
`Current success rates (as of December 3 1, 1993),
`maximum possible success rates (assuming all open
`INDs are approved), and predicted final success rates
`for IND filing intervals are shown in Table I. Except
`for the 1964 to 1969 interval,
`the predicted final suc-
`cess rates fall between current and maximum possible
`success rates for all groups and filing intervals. How-
`
`Page 5 of 14
`
`
`
`6 DiMa&
`
`CLINICAL
`
`PHAFtMACOLCGY
`
`&THERAPEUTICS
`JULY 1995
`
`Table I. Current cumulative,* maximum possible,7 and predicted final* approval success rates for NCEs by IND
`filing interval
`
`No. of IND
`applications
`
`Current success
`rate (%)
`
`No. of active
`INDsJ
`
`Maximum success
`rate (%)
`
`final
`Predicted
`success rate (%)
`
`Self-originated
`13.7
`438
`1964-69
`18.4
`315
`1970-74
`17.8
`208
`1975-79
`16.0
`288
`1980-84
`Self-originated (first tested in United States)
`196469
`399
`10.8
`1970-74
`249
`13.7
`150
`13.3
`1975-79
`1980-84
`184
`8.2
`Licensed
`1964-69
`1970-74
`1975-79
`1980-84
`All NCEs
`1964-69
`1970-74
`1975-79
`1980-84
`
`95
`67
`55
`91
`
`579
`385
`264
`379
`
`32.6
`41.8
`38.2
`25.3
`
`15.7
`22.3
`22.0
`18.2
`
`2
`5
`14
`40
`
`1
`4
`8
`25
`
`0
`2
`3
`13
`
`3
`7
`17
`53
`
`14.2
`20.0
`24.5
`29.9
`
`11.0
`15.3
`18.7
`21.7
`
`32.6
`44.8
`43.6
`39.6
`
`16.2
`24.2
`28.4
`32.2
`
`13.3
`19.2
`22.9
`20.5
`
`10.4
`14.3
`17.3
`9.5
`
`31.4
`44.5
`41.6
`30.3
`
`15.6
`23.4
`26.8
`23.5
`
`INDs,
`
`investigational
`
`new drug applications.
`
`NCEs, New chemical entities;
`*As of December
`31, 1993.
`by assuming
`rate- is determined
`tThe maximum
`possible
`success
`approved as of December 3 1, 1993, will eventually
`be approved.
`of research
`to termination
`$Final success
`rates are estimated using a two-stage parametric model. Times
`a Weibull distribution
`to the data. The conditional
`probability
`of approval at a given
`time
`to termination
`$As of December 31, 1989, and not yet approve> as of D&ember
`31, 1993..
`
`that all NCEs with
`
`IND applications
`
`that were active as of December
`
`31, 1989, and not yet
`
`are estimated by fitting
`(research abandonment or approval)
`of research
`is predicted
`from a pmbit specification.
`
`ever, each of the predicted success rates for the 1964
`to 1969 interval are within 1.2 percentage points of
`the maximum success rate.
`Licensed compounds generally have undergone
`some testing before licensing and have been shown to
`be promising candidates for marketing approval. Thus
`one would expect licensed drugs to have higher than
`average success rates as a result of a screening pro-
`cess. A comparison of the predicted success rates in
`Table I for self-originated and licensed NCEs suggests
`a strong screening effect. For the first two intervals,
`predicted success rates for
`licensed drugs are more
`than twice as high as for self-originated drugs. The
`screening effect appears to have diminished over time,
`with predicted success rates for licensed drugs only
`50% higher for the filings of the early 1980s.
`A screening effect also appears to apply to self-
`originated compounds that have undergone some clini-
`cal testing abroad before an IND has been filed in the
`United States. The success rates for self-originated
`NCEs that were first tested in humans in the United
`States are much lower
`than the success rates for all
`self-originated NCEs.
`the
`from
`Predicted
`final success rates increased
`1960s to the late 1970s for self-originated and all
`
`NCEs. Success rates for filings of the early 1980s on
`self-originated and all NCEs are modestly lower than
`success rates for the late 1970s. However,
`they are
`very close to the success rates for the early 1970s.
`A relatively short time frame for analysis of NCEs
`with INDs filed in the late 1980s precludes statistical
`estimation of success rates for this group of NCEs.
`However, observed success rates through 1993 for the
`late 1980s can be compared with success rates for ear-
`lier periods at the same amount of time from IND fil-
`ing. Cumulative success rates at 4 years from IND fil-
`ing for 1980 to 1984 filings, 1985 to 1986 filings, and
`1987 to 1989 filings are 2.5%, 2.7%, and 1.4%, re-
`spectively. Each of these groups have had at least 4
`years from
`IND
`filing
`to obtain approval. The time
`from
`IND
`filing,
`though,
`is relatively short, and the
`trend conceivably could change in the future. How-
`ever,
`if the trend persists, then either success rates
`will have declined during
`the 1980s or development
`times will have increased, or both.
`filings and
`Fig. 3 shows how the number of IND
`the number of NCE approvals (3-year moving aver-
`ages) have changed over time for the firms in the da-
`tabase. Although
`the number of IND
`filings has de-
`clined
`from
`the mid 1960s to the mid 197Os, the
`
`Page 6 of 14
`
`
`
`CLINICAL
`VOLUME
`
`P HABMACOLOGY
`58, NUMBER
`1
`
`& THERAPEFTICS
`
`DiMasi
`
`7
`
`110
`
`100
`
`so
`
`80
`
`70
`
`60
`
`50
`
`40
`
`I
`
`30
`
`20
`
`10
`0 1
`1963
`
`1965
`
`1967
`
`1969
`
`1971
`
`1973
`1975
`1977
`1983
`1961
`1979
`Year of IND FilinglNCE Approval
`
`1965
`
`1907
`
`1989
`
`1991
`
`1993
`
`Fig. 3. Three-year moving averages of IND filings (1963 to 1989) and NCE approvals (1993 to
`1993). NCE approvals are for the 36 firms that responded to the survey that yielded the IND fil-
`ings shown. (Data [IND filings] from DiMasi JA, et al. CLIN PHARMACOL THER 1994;55:609-22.)
`
`increased some-
`number of NCE approvals generally
`what during
`the 1970s through
`the mid 1980s-the
`period during which most of the approvals from the
`IND
`filings of the 1960s and early 1970s would have
`occurred. This reflects the increase in success
`rates
`over time noted previously and an increasing percent-
`age of filings on licensed drugs and on self-originated
`drugs first tested abroad.
`The decline in IND filings during the last half of the
`1980s (11.3% relative to the early 1980s) portends a de-
`clining number of NCE approvals in the mid to late
`1990s if success rates for late 1980s IND filings remain
`the same or decline relative to rates for the early 1980s
`(as suggested previously),
`if the distributions of IND
`filings by source and class remain basically the same,
`and if average development
`times do not change ap-
`preciably. The distributions of NCEs by therapeutic
`class and source (licensed/self-originated)
`are essen-
`tially the same for the early and the late 1980s. How-
`ever, the proportion of self-originated NCEs that were
`first tested abroad increased slightly (35% to 39%).
`Because the distribution of NCEs by class and source
`remained
`largely
`the same for
`the early and
`late
`1980s an assumption of constant success rates during
`the 1980s would imply a decline of approximately 10
`approvals for the 1985 to 1989 filings relative to the
`1980 to 1984 filings (79 versus 89, with use of the
`predicted final success rates in Table I).
`
`the focus here is
`NDA submission rates. Although
`on approval success rates, submission of a new drug
`application
`(NDA) with
`the FDA
`for marketing ap-
`proval is a significant milestone
`in the development
`and regulatory history of a new drug. Firms submit
`NDAs when they believe that the evidence from clini-
`cal studies is sufficiently
`favorable
`to convince the
`regulatory authority
`that the compound’s benefit/risk
`ratio is adequate for approval. The same statistical
`methods that were used to estimate approval success
`rates were applied to data on NDA submissions. Fig.
`4 shows predicted final NDA submission rates by IND
`filing interval (i.e.,
`the probability
`that an NCE with
`an IND filed during an interval will have an NDA sub-
`mitted).
`the
`The results in Fig. 4 can be combined with
`results on predicted
`final approval success rates to
`estimate NDA success rates (i.e.,
`the probability of
`obtaining marketing approval for NCEs that have had
`an NDA submitted). The estimates of approval and
`submission success rates indicate little variation over
`time in the proportion of NDAs that will be approved.
`The results show a slight downward
`trend in NDA
`success rates, with the models predicting
`that 86.7%
`of the NDAs submitted on NCEs with
`INDs
`filed
`from 1964 to 1969 and 82.2% of the NDAs submitted
`on NCEs with INDs filed from 1980 to 1984 will be
`approved.
`
`Page 7 of 14
`
`
`
`8 DiMmi
`
`CLINICAL
`
`PHARMACOLOGY
`
`& THERAPEUTICS
`JULY 1595
`
`Table II. Reasons for abandonment of research on NCEs by IND filing interval
`safety
`time to
`Median
`abandonment
`(yr)
`
`No. of NCEs
`abandoned*
`
`% With research
`terminated
`
`Eficacy
`% With research
`time to
`Median
`terminated
`abandonment
`(yr)
`
`IND filing
`interval
`19641969
`1970-1974
`19751979
`1980-1984
`19851989
`
`1964-1989
`
`406
`247
`152
`194
`100
`1099
`
`23
`28
`26
`33
`26
`
`27
`
`1.6
`2.9
`2.7
`2.1
`1.4
`
`2.0
`
`53
`46
`43
`36
`41
`
`46
`
`1.7
`2.7
`2.9
`2.5
`1.2
`
`2.1
`
`for
`identified a reason
`firms
`for which surveyed
`only NCEs
`*Includes
`research abandoned by December 31, 1989, 17% had no reason specified.
`or economics
`iIncludes
`reasons
`that could not be unambiguously
`placed
`in one of the safety, efficacy,
`isolation problems,
`a new analog or metabolite was pursued, and not as promising
`as another NCE.
`
`termination
`
`of research. Of
`
`the 1322 NCEs with
`
`lNDs
`
`filed during 1964 to 1989 and with
`
`categories. Examples
`
`include such reasons as formulation
`
`or
`
`Table III. Current cumulative* and maximum possible? approval success rates for self-originated NCEs by
`therapeutic category and IND
`filing interval
`
`IND filing
`
`interval
`
`1964-1969
`Current success
`rate (%)
`
`Maximum success
`rate (Yo)
`
`No.
`INDs
`
`1970-1974
`Current success Maximum success
`rate
`rate (%)
`
`(70)
`
`9.8
`24.4
`25.0
`12.8
`11.3
`7.7
`20.0
`13.3
`
`11.8
`24.4
`25.0
`12.8
`11.3
`7.7
`20.0
`13.3
`
`55
`49
`9
`49
`67
`44
`3
`23
`
`23.6
`22.4
`44.4
`12.2
`13.4
`15.9
`33.3
`30.4
`
`25.5
`22.4
`55.5
`12.2
`17.9
`15.9
`33.3
`30.4
`
`No.
`INDs
`5 1
`90
`8
`86
`106
`39
`10
`15
`
`is determined by assuming
`rate
`be approved.
`
`that all NCEs with
`
`INDs
`
`that were active as of December 31, 1989, and not yet approved as of
`
`Therapeutic
`category
`Analgesic/anesthetic
`Antiinfective
`Antineoplastic
`Cardiovascular
`CNS
`Endocrine
`Gastrointestinal
`Respiratory
`system.
`CNS , Central nervous
`*As of December
`31, 1993.
`Whe maximum possible
`success
`December 3 1, 1993, will eventually
`
`The database
`research abandonment.
`Reasons for
`contained
`information on the reasons research was
`abandoned
`for NCEs
`that had research terminated
`without marketing approval. We grouped
`the re-
`sponses into three major categories: safety (e.g., “hu-
`man
`toxicity” or “animal
`toxicity”),
`efficacy (e.g.,
`“activity
`too weak” or “lack of efficacy”), and eco-
`nomics (e.g., “commercial market too limited” or “in-
`sufficient
`return on
`investment”).
`For a relatively
`small number of the compounds
`for which research
`was abandoned, reasons for termination were not spe-
`cific enough to be placed in one of these three catego-
`ries. The share of all reasons for abandonment and the
`median time to abandonment
`for each of these catego-
`ries by IND
`filing
`interval are shown
`in Table II.
`Some of the results are similar to those of Tucker et
`al.,” who found,
`for example,
`that 33% of the re-
`search discontinuations
`for new drugs with IND appli-
`cations filed from 1976 to 1978 were for economic
`reasons.
`
`issues became
`the early 198Os, economic
`Through
`relatively more prevalent and issues related to lack of
`efficacy became relatively less prevalent as reasons for
`research termination. Because the time available for
`the fate of the compounds to have been determined
`is
`limited,
`the abandonment results for the 1985 to 1989
`interval are biased toward causes that tend to be re-
`vealed relatively soon after filing. This censoring ef-
`fect may also apply to earlier intervals, but with much
`less impact. The economics share increased through
`the late 198Os, even though research on NCEs termi-
`nated for economic reasons tends to occur later in the
`development process than is the case for safety and ef-
`ficacy (e.g., for 1980 to 1984 filings, 55% of the eco-
`nomics terminations occurred at least 4 years from fil-
`ing, compared with 30% of efficacy terminations and
`2 1% of safety terminations).
`As can be seen from the median abandonment times
`in Table II, since the early 1970s termination of re-
`search for safety issues tended to occur earlier in the
`
`Page 8 of 14
`
`
`
`CLINICAL
`VOLUME
`
`PHARMACOLOGY
`58, NUMBER
`1
`
`& THERAl’EUTlCS
`
`DiMasi
`
`9
`
`Economics
`
`U