`success rates for investigational drugs
`
`Joseph A. DiMasi, PhD Boston, Mass
`
`The drug development process is known to be com(cid:173)
`plex, costly, and time-consuming.l-3 The process is also
`risky in that most compounds that undergo clinical test(cid:173)
`ing are abandoned without obtaining marketing
`approval. The rate at which pharmaceutical firms mar(cid:173)
`ket new therapies in the United States is an important
`measure of the viability of the drug development
`process.4 The cost of new drug development is also crit(cid:173)
`ically dependent on the proportion of drugs that fail in
`clinical testing.5-7 Estimates of industry success rates
`can be used in benchmarking exercises for project plan(cid:173)
`ning pur poses. Given the length and cost of the drug
`development process, careful consideration of all fac(cid:173)
`tors that have a significant impact on the process is
`needed to appropriately a llocate research and develop(cid:173)
`ment resources.
`In a series of studies of new drug development in the
`United States, the Tufts Center for the Study of Drug
`Development (CSDD) and others have provided
`descriptive data on how cumulative success rates for
`new chemical entities (NCEs) vary with time from
`investigational new drug application (IND) tiling.I,S-14
`Several studies have also examined c linical success
`rates for biotechnology-derived drugs. 15- 17 Statistical
`modeling can be helpful in analyzing success rates for
`recent periods because many of the compounds will still
`be in active testing at the time of the analysis. Tufts
`CSDD has also conducted a number of studies that use
`this approach to predict final success rates for groups
`
`From the Director of Economic Analysis, Tufts Center for the Study
`of Drug Development, Tufts University.
`This research was supported in part by a grant from the Drug Infor(cid:173)
`mation Association.
`Received for publication Nov 6, 2000; accepted Feb 26, 2001.
`Reprint requests: Joseph A. DiMasi, PhD, Tufts Cen.ter for the Study
`of Drug Development, Tufts University, 192 South St, Suite 550,
`Boston, MA 021 I I.
`Clin Pharmacal Ther 2001 ;69:297-307.
`Copyright © 2001 by Mosby, Inc.
`0009-9236n OOI/$35.00 + 0 13/1/115446
`doi: 10.1067/mcp.2001. 115446
`
`of compounds for which the ultimate fate of some of
`the compounds in the data set is not known.4-7,18-20
`T his study provides updated success rate analyses for
`NCEs. Success rate trends and variations in success
`rates by therapeutic class are presented. The hypothe(cid:173)
`sis that pharmaceutical firms have been moving com(cid:173)
`pounds through the process to either marketing
`approval or research abandonment more quickly is also
`examined. In addition, attrition rates for compounds
`entering clinical development phases are obtained.
`Finally, statistics on the reasons compounds fail 111
`development are given.
`
`METHODS
`Data used for this study were obta ined primarily
`from a Tufts CSDD database that contains information
`from ongoing surveys of pharmaceutical firms. The
`data provided for the most recent survey come from
`firms that have declined in number over the study
`period, as mergers have resulted in the combination of
`some of them. The data used for this study were
`obtained from the units and subsidiaries of what are
`now 24 parent firms. These firms provided data on
`NCEs first investigated in humans anywhere in the
`world or NCEs for which they were the first to file a
`US lND since 1963. T he data gathered include IND fil(cid:173)
`ing dates, the dates on which IND research was aban(cid:173)
`doned, reasons for termination of research, the latest
`phase compounds were in when research was aban(cid:173)
`doned, and the date of new drug application approval.
`A description of additional information included in this
`database is available elsewhere. ' Data were also
`obtained from public sources.2l ,22 Current success rates
`for these NCEs were examined (as of December 31,
`1999), and statistical analysis was applied to data on
`past rates of research abandonment and approval to pre(cid:173)
`dict future success rates. Analyses were conducted for
`NCEs with INDs first filed in 3- and 6-year periods
`from 198 1 to 1992. Data on more recent IN Ds were
`available but, given the length of theNCE development
`
`297
`
`Page 1 of 11
`
`Biogen Exhibit 2014
`Coalition v. Biogen
`IPR2015-01136
`
`
`
`298 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`,\1}\y 2001
`
`process, they are too recent to use for a comprehensive
`statistical analysis of success rates.
`Inclusion criteria. For purposes of this study, an
`NCE is defined as a new molecular compound not pre(cid:173)
`viously tested in humans. Excluded are new salts and
`esters of existing compounds, surgical and diagnostic
`material s, vaccines and other biologic agents, certain
`externally used compounds (such as disinfectants,
`antiperspirants, and sunscreens), and nutritional com(cid:173)
`pounds (such as natural forms of vitamins and sweet(cid:173)
`ening agents). Our definition of an NCE differs from
`the FDA's definition of a new molecular entity. The
`most notable difference is that the FDA's definition
`includes diagnostics, whereas our definition of an NCE
`does not.
`Statistical analysis of success rates. For the statisti(cid:173)
`cal analyses, residence time (the length of time from
`LND filing to either abandonment of research without
`marketing approval or to new drug application
`approval) was calculated for NCEs with INDs first filed
`in successive 3-year intervals from 1981 to 1992.
`Approval dates were available through December 31,
`1999, and were used in determining observed success
`rates. Residence times were also calculated as of the
`end of 1999. Observed and predicted cumulative
`approval success rates were calculated at each year
`from TND filing.
`NCEs were stratified according to source (self(cid:173)
`originated versus licensed-in or otherwise acquired) and
`therapeutic class. An NCE is defined as self-originated if
`it was developed entirely under the auspices of the
`responding firm. We define acquired NCEs to be com(cid:173)
`pounds that were obtained by the developing fmn through
`licensing, purchase, barter, or other means. To determine
`whether trends in success rates exist, we analyzed the data
`by the period during which the IND was filed.
`Predicted success rates for IND filing periods were
`determined from a 2-stage model of the approval
`process. NCEs with research still active as of Decem(cid:173)
`ber 31, 1999, constitute r:ight-censored observations for
`our data set. Survival analysis can make use of infor(cid:173)
`mation provided by censored data.23 NCEs were
`assumed to survive until either research was terminated
`without approval or marketing approval was achieved.
`Details of the selected models and the computational
`approach used to estimate final success rates are pro(cid:173)
`vided in the Appendix.
`The survey data also provided information on the lat(cid:173)
`est development or regulatory phase that abandoned
`NCEs were in at the Lime of termination. These data
`allow us to determine the distribution of research ter(cid:173)
`minations by phase. In combination with predicted
`
`approval rates for IND filing intervals, they also permit
`us to estimate the probability of approval once a com(cid:173)
`pound enters a given clinical phase and phase attrition
`rates (the percentage of compounds that enter a phase
`that are abandoned before the next phase is initiated).
`
`RESULTS
`Included in the CSDD database of investigational
`compounds are the development histories of 671 NCEs
`for which survey firms had filed a first IND from 1981
`to 1992. Of these, 508 were identified as self-originated
`and 163 were identified as acquired. Of the 508 self(cid:173)
`orignnated NCEs, 350 were initially investigated in
`humans in the United States. By the end of 1999, 20.9%
`of the NCEs with lNDs fi led from 1981 to 1992 had
`been approved for marketing in the United States. For
`this i(Jeriod, the current US approval success rates for
`NCEs that were acquired, self-originated, and self-orig(cid:173)
`inated and first tested in humans in the United States
`are 33.1 %, 16.9%, and 8.6%, respectively. These results
`illustrate the significance of previous testing on mea(cid:173)
`sured US success rates; success rates on IND filings are
`highe r for compounds that were licensed-in or first
`tested abroad.
`Time to research termination. Even though some of
`the drugs in our database are still active, survival analy(cid:173)
`sis can be used to establish the rates at which the NCEs
`with INDs filed during a given period will be dropped
`from active testing. The mean and median times to
`research termination for self-originated NCEs that were
`abandoned with INDs first filed during the periods from
`1981 to 1983, 1984 to 1986, 1987 to 1989, and 1990 to
`1992 are shown in Fig l. Because NCEs in the later
`intervals had less time for research to be terminated,
`the averages for the later periods may be somewhat
`understated relative to the earlier periods. However,
`previous research and our current data suggest that the
`likelihood of approval, as opposed to abandonment,
`increases with time from IND filing. If we could add
`termination times for NCEs that will eventually be ter(cid:173)
`minated, the impact should be much less on the median
`than on the mean.
`Even with these qualifications, the results at least
`suggest that, over time, pharmaceut ical firm s have
`made quicker decisions on research failures. Mean res(cid:173)
`idence time decreased 30% (1.5 years) from the
`1981-1983 to the 1990- 1992 IND filing intervals.
`Median time to research abandonment decreased 20%
`(0.8 years) for INDs fi led in the early 1990s relative to
`the early 1980s.
`Further evidence that the ultimate fate of investiga(cid:173)
`tional NCEs has tended to be resolved more rapidly
`
`Page 2 of 11
`
`
`
`CLINICAL D'HARMACO L.OGY & T H ERAPJ;trn CS
`VOLUME 69, N UMIIER 5
`
`DiMasi 299
`
`1981· 1983
`
`1987·1989
`1984·1986
`Period of INO Filing
`I ~:~ Mean D Median I
`
`1990·1992
`
`Fig 1. Mean and median time to research abandonment for self-originated new chemical entities
`(NCEs) with a first investigatio nal new drug application (IND) filed during a g iven period.
`
`1.0
`0.9
`0.8
`:>. 0.7
`~ 0.6
`:c 0.5
`ftl
`.0
`0 ... 0.4
`a.. 0.3
`0.2
`0.1
`0.0
`
`0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
`Years from IND Filing
`1-+-1981-1983 .... 1990-1992 1
`
`Fig 2 . Estimated survival cu~rves for self-orig inated NCEs with a first IND filed during a given
`period. The curves show the percentage of NCEs that had not been abandoned or approved for mar(cid:173)
`keting in the United States (ie, still active) a given number of years from the date of IND filing.
`The data were fitted to Weibull distributions.
`
`over time is shown in Fig 2. The curves in the figure
`are estimated survival c urves for the J 98 1-1983 to
`1990-1992 IND filing inte rvals. A point on the curve
`represems the probability that an investigational NCE
`will stm be active a given number of years from IND
`filing. An NCE is inactive at a given point in time if
`either research has been abandoned without marketing
`approvaJ or the compound has received FDA approval
`for marketing. It should be noted that the estimated sur(cid:173)
`vival curves account for censored data; that is, infor-
`
`mation regarding sti!R active NCEs is used to estimate
`final survival rates.
`Median survival time decreased from 4.9 years to 4.3
`years (12%) for the 1981- 1983 to 1990-1992 filing
`intervals, respectively. Faster action is also evident in
`the figure for different a mounts of time from JND fil (cid:173)
`ing. The percentages of NCEs for the 1990-1992 filing
`period that are still active are 6 to 7 percentage points
`lower than those for the 1981- 1983 filing period at 4
`to 10 years from IND filing.
`
`Page 3 of 11
`
`
`
`300 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`,\1}\y 2001
`
`'t:l
`411 >
`0 a
`:t ... c:
`
`411
`u
`Q;
`a..
`
`1981-1983
`
`1987-1989
`1984-1986
`Period of IND Filing
`
`I 0 Self-Originated 0 Acquired • All I
`
`1990-1992
`
`Fig 3. Current clinical approval success rates for NCEs by origin and period during which a first
`IND was fLied.
`
`Success rate trends. To estimate final success rates,
`results from the survival analyses must be combined
`with those from qualitative choice models of the con(cid:173)
`d itional probabi lity of approval at given residence
`times. The parameter est imates for both stages of the
`model are highly statistically significant, and good(cid:173)
`ness-of-fit measures ind icate strong agreement with
`the data. The parameter estimates used to determine
`the predicted final success rates reported here and the
`accompanying statistical results are available upon
`request.
`Current success rates (as of December 3 1, 1999) for
`self-originated, acquired, and all NCEs by IND fi ling
`interval are shown in Fig 3. Licensed compounds gen(cid:173)
`erally have undergone some testing before licensing
`and hav.e been shown to be promising candidates for
`marketing approval. The results support the hypothesis
`of such a screening effect for acquired NCEs; current
`success rates for acquired NCEs are notably higher than
`those for self-originated NCEs.
`A screening effect also appears to apply to self(cid:173)
`originated compounds that have undergone some clini(cid:173)
`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(cid:173)
`originated NCEs. Current success rates by IND fi ling
`interval for self-originated NCEs first tested in the
`United States are 33% to 65% lower than for self (cid:173)
`originated NCEs as a whole.
`Censoring has an impact on the results for all IND
`filing intervals, but the effect is much greater for the
`more recent intervals. The proportions of NCEs that are
`
`currently active are substantially higher for these later
`periods. Thus the lower current success rates for self(cid:173)
`originated NCEs in the 1987- 1989 and 1990-1992
`intervals may simply reflect the shorter amount of time
`available for the ultimate fare of those NCEs to have
`occurred. Trend analysis for these later periods must be
`aided by the application of statistical techniques to fore(cid:173)
`cast approval rates for the active NCEs.
`Current success rates, maximum possible success
`rates (assuming all active NCEs are approved), and
`predicted final success rates for self-originated NCEs
`by IND f iling interval are shown in Fig 4. The pre(cid:173)
`dicted fina l success rates fall between current and max(cid:173)
`imum possible success rates for all filing intervals.
`Although both predicted and maximum possible suc(cid:173)
`cess rates are lower for the 1987-1 989 interval rela(cid:173)
`tive t o the intervals in the earlier 1980s, the predicted
`success rate for the 1990- 1992 interval is 16% higher
`than for the interval with the next highest predicted
`success rate.
`Comparison of predicted and actual success rates for
`the early time periods can validate the performance of
`the statistical model. For NCEs with INDs first filed
`from 1981 to 1983, t he model predicts a cumulative
`success rate of 19.5% at 16 years from IND fi ling (the
`maximum amount of time available for all compounds
`in the group); the actual success rate for this group at
`16 years from IND filing is 19.8%. Similarly, NCEs
`with INDs first filed from 1984 to 1986 have a pre(cid:173)
`d icted success rate of 18.8% at 13 years from IND fi l(cid:173)
`ing and an actual success rate of 19.4%.
`Therapeutic classes. Previous research has indicated
`that success rates fo r NCEs vary by therapeutic
`
`Page 4 of 11
`
`
`
`CLINICAL D'HARMACOL.OGY & THERAPJ; trn CS
`VOLUME 69, N UMIIER 5
`
`DiMasi 301
`
`"0
`
`Q) > 0 a.
`t
`c Q)
`~
`Q) a..
`
`1981-1983
`
`1984-1986
`
`1987-1989
`
`1990-1992
`
`Period of INO Filing
`
`I 0 Current D Maximum • 1 Predicted J
`
`Fig 4. Current (as of December 31, 1999), maximum possible, and predicted final clinical approval
`success rates for self-originated NCEs by period during which a ftrst IND was filed. Maximum pos(cid:173)
`sible success rates were determined under the assumption that all active compounds are eventua.lly
`approved foir marketing. Predicted success rates were constructed with use of estimates for a sur(cid:173)
`vival analysis of residence time (time from IND fLiing to abandonment or US marketing approval)
`with a Wei bull distribution specification and estimates for the conditional probability of approval
`for a given residence time with a probit specification.
`
`Table I. Current and maximum possible success rates by therapeutic class for self-originated NCEs with INDs first
`filed from 1981 to 1992*
`
`Therapeutic class
`
`NCEs
`
`Approved NCEs
`
`Open NCEst
`
`Current
`success rate/
`
`Maxim.um.
`success rate;f:
`
`Analgesic/anesthetic
`Anti-infective
`Antineoplastic
`Cardiovascular
`Central nervous system
`Endocrine
`Gastrointestinal
`Immunologic
`Respiratory
`Miscellaneous
`
`49
`57
`38
`120
`110
`33
`15
`13
`25
`43
`
`NCE, New c hemical entily.
`*Therapeu1ic class informal ion is missing for live compounds.
`( As of December 3 1, 1999.
`+Assumes !hal all open NCEs will eventually be approved.
`
`10
`16
`6
`21
`16
`6
`3
`2
`3
`3
`
`4
`3
`6
`6
`14
`4
`2
`0
`0
`4
`
`20.4%
`28.1 %
`15.8%
`17.5%
`14.5%
`18.2%
`20.0%
`15.4%
`12.0 %
`7.0%
`
`28.6%
`33.3%
`31.6%
`22.5%
`27.3%
`30.3%
`33.3%
`15.4%
`12.0%
`16.3%
`
`class_6.20 The current and maximum possible success
`rates by IND filing interval for self-originated NCEs in
`9 specific therapeutic categories are shown in Table I.
`Because the number of compounds available for analy(cid:173)
`sis is greatly reduced when the data are stratified into
`the entire study period
`therapeutic categories,
`( 1981-1992) is used. For the immunologic and respi(cid:173)
`ratory categories the fate of all of the NCEs is known
`so that current, maximum, and final success rates are
`the same.
`
`For many of these therapeutic classes, the number of
`compounds with lND filings in an interval is too small
`for accurate statistical estimation. However, we had
`enough data and the fits with the statistical model
`described above were sufficiently good for us to estimate
`predicted final success rates for the analgesic/anesthetic,
`anti-infective, cardiovascular, and central nervous sys(cid:173)
`tem categories. The current, maximum possible, and
`predicted final success rates for these 4 classes are
`shown in Fig 5. Relative success rate results for these
`
`Page 5 of 11
`
`
`
`302 D iMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`,\1}\y 2001
`
`Cardiovascular
`
`CNS
`
`Analg/Anesth
`
`Ant II nfective
`
`I o Current [] Maximum • Predicted I
`
`F ig 5. Curre nt (as of December 31, 1999), maximum possible, and predicted final clinical approval
`success rates by therapeutic class for self-originated NCEs with a first IND filed from 198 1 to 1992.
`Maximum possible success rates were determined under the assumption that all active compounds
`are eventually approved for ma rketing. Predicted success rates were constructed with use of esti(cid:173)
`mates for a survival analysis of residence time (time from IND filing to abandonme nt or US mar(cid:173)
`keting approval) w ith a Wei bull distribution specification a nd estimates for the conditional proba(cid:173)
`bility of approval for a given residence time with a probit sp ecification.
`
`52.9
`
`352
`
`-c:
`1.) ... (I) a.
`
`(I)
`
`52.2
`
`~QR
`
`11.9
`
`1
`
`8.0
`
`I
`
`1981·1986
`
`1987·1992
`
`[ 0 Phase I
`
`Period of IND Filitng
`• Phase 111/RR I
`
`Q Phase II
`
`F ig 6. Distribution of research terminations for self-originated NCEs by clin ical phase and per:iod
`during whiclh a first lND was filed.
`
`classes are likely unaffected by time trends inasmuch
`as the number of filings for the last half of the study
`period as a percentage of total filings for the whole
`period for each of these 4 classes varied only from 47%
`to 55%. The predicted success rates range from approx(cid:173)
`imately I in 5 for cardiovascular NCEs to I in 3 for
`allli-infectives.
`Clinical phase attrition rates. Clinical approval suc(cid:173)
`cess rates yield patterns of success for the clinical
`
`development process as a whole, but they do not inform
`us of success and failure patterns during the clinical
`development process. Our data on the latest phase that
`an abandoned NCE was in at the time of termination
`give us a distribution of research terminations by phase.
`The distribution for self-originated NCEs is shown in
`Fig 6. Approximately half of clinical research fai lures
`occur in phase II. This is the case for both the first and
`second halves of the study period. For the late r IND til-
`
`Page 6 of 11
`
`
`
`C LINICAL D'HARMACO L.OGY & T H ERAPJ;trn CS
`VO LUME 69, N UMIIER 5
`
`DiMasi 303
`
`72.9
`
`78.5
`
`-c:
`u £
`
`Q)
`
`33.0
`
`2d.2
`
`32.7
`
`22.6
`
`1981 -1986
`
`1987-1992
`
`Period of IND Filling
`I 0 Phase I LJ Phase II 0 Phase ml
`Fig 7. Approval success rates for self-originated NCEs entering a given clinical phase.
`
`54.7
`
`58.8
`
`26.7
`
`27.1
`
`30.8
`
`21 .5
`
`1981-1986
`
`1987-1992
`
`Period of IND Filing
`• Phase Ill I
`Fig 8. Phase attrition rates (percentage of compounds entering a phase that faH in the phase) for
`self-originated NCEs by period during which a first TND was filed.
`
`[ 0 Phase I D Phase II
`
`ing period, however, proportionately more research fail(cid:173)
`ures occurred in phase I and proportionately fewer
`occurred in phase III or regulatory review.
`Statistical analysis yields predicted fina l success
`rates for self-originated NCEs for the 1981- 1986 and
`1987-1 992 fil ing intervals of24.2% and 22.6%, respec(cid:173)
`tively. Current approval and termination rates for these
`periods, along with the assumption that currently active
`NCEs that are predicted to eventually fai l will do so in
`phase HI or regulatory review, allow us to predict
`approval rates for NCEs that enter a clinical phase (Fig
`7). Although approval rates are similar for the early
`clinical phases in both periods, the likelihood of
`approval increased by 5.6 percentage points fo r phase
`
`III. This is consistent with the results displayed in Fig
`6, which showed relatively more terminations in phase
`I and relatively fewer in phase Til or later.
`The data on research terminations by phase and pre(cid:173)
`dicted success rates also allow us to determine phase
`attrition rates. Fig 8 s!hows that attrition rates are great(cid:173)
`est in phase II in which more than half of the investigated
`compounds fail. During the study period, failure rates
`increased for phases I and II but declined for phase III.
`Reasons for research abandonment. The database
`contained informatio n on the reasons research was
`abandoned for NCEs that had research tenninated with(cid:173)
`out marketing approval. We grouped the responses into
`3 major categories: safety (eg, "human toxicity" or "ani-
`
`Page 7 of 11
`
`
`
`304 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`,\1}\y 2001
`
`-c:
`
`Q.l
`
`...,
`~
`
`Economics
`
`Efficacy
`
`Safety
`
`Other
`
`lo 1981-1986 • 1987-1992 1
`
`F ig 9. Percentage of research termjnations for all NCEs by period of first lND filing and by pri(cid:173)
`mary reason for abandonment.
`
`mal toxi,city"), efficacy (eg, "activity too weak" or "lack
`of efficacy"), and economics (eg, "commercial market
`too limited" or "insufficient return on investment"). A
`relatively small number of the compounds that had been
`abandoned had reasons for termination that were not
`specific enough to be placed in I of these 3 categories.
`The shalfes of all reasons for abandonment for each of
`these categories by IND filing interval are shown in
`Fig 9.
`For the last half of the study period, economic and
`efficacy issues became relatively more prevalent, while
`safety issues became rei ati vely less prevalent, as rea(cid:173)
`sons for research termination. Because the time avail(cid:173)
`able for the fate of the compounds to have been deter(cid:173)
`mined is limited, the abandonment results for the inter(cid:173)
`val from 1987 to 1992 are biased toward causes that
`tend to be revealed relatively soon after filing. This
`censorirng effect also applies to the earlier interval but
`with much less impact. The economic share increased,
`even though research on NCEs terminated for eco(cid:173)
`nomic reasons tends to occur later in the development
`process than is the case for safety and efficacy (eg, for
`filings from 1981 to 1986, 45 % of the economic ter(cid:173)
`minations occurred at least 6 years from filing com(cid:173)
`pared with 35% of efficacy and 17% of safety termi(cid:173)
`nations).
`The censoring effect also applies when the data are
`analyzed by the phase that a compound was in when it
`was abandoned. This bias will tend to be lower if ear(cid:173)
`lier periods are examined. Considering the first half of
`the study period (NCEs that had an IND first fi led from
`1981 to 1986), compounds that had failed for economic
`or efficacy reasons were terminated much more fre-
`
`quently in late clinical testing phases. The percentage
`of failed compounds that were abandoned in phase III
`or during the regulatory review period was 26.6% for
`economic failures, 24.0% for efficacy failures, and
`8.3% for safety failures.
`Table II shows mean and median abandonment times
`for all NCEs by IND filing period and by the primary
`reason for termination. Average times to abandonment
`are lower for the later filing period, but this can result
`in part from the shorter period during which abandon(cid:173)
`ments can occur for this interval. For either period,
`however, both the mean and median time to research
`abandonment is longer for NCEs that were terminated
`primarily for economic than for other reasons. The data
`also show that economic considerations were the most
`frequent determinants underlying decisions to termi(cid:173)
`nate late-stage clinical research. During the entire study
`period, 39% of the terminations that occurred at least
`4 years from filing were for economic reasons, 32%
`were related to efficacy issues, and only 16% were fo r
`safety problems ( 13% were for other reasons).
`
`DISCUSSION
`A statistical model of the rate at which new drugs
`proceed through clinical testing to marketing approval
`was estimated for three 4-year and two 6-year IND fil(cid:173)
`ing intervals. Estimated approval success rates for self(cid:173)
`orignnated NCEs varied from 19% to 30% during the
`study period. The highest predicted success rate was
`for lhe most recent filing period ( 1990-1992). The
`resul ts suggest that approval rates have not declined
`over time and, quite possibly, have increased. A gen(cid:173)
`eral improvement in success rates can result from bet-
`
`Page 8 of 11
`
`
`
`CLINICAL D'HARMACOL.OGY & THERAPJ;trn CS
`VOLUME 69, N UMIIER 5
`
`DiMasi 305
`
`Table II. Time to research abandonment (in years) for NCEs by IND filing period
`
`Reason
`
`Economics
`Efficacy
`Safety
`Other
`
`1981-1986
`
`1987-1992
`
`n
`
`64
`71
`46
`34
`
`Mean (y)
`
`Median (y)
`
`4.4
`3.6
`2.6
`3.5
`
`4.0
`2.3
`2.5
`2.3
`
`ll
`
`45
`50
`26
`12
`
`Mean (y)
`
`Median (y)
`
`3.7
`2.7
`2.1
`2.7
`
`3.2
`2.6
`1.2
`2.2
`
`IND. Investigational new drug application.
`
`ter precl inical screening. The imp I ications fort he devel(cid:173)
`opment process are sign ificant because the clinical
`costs for some research failures will not be borne if suc(cid:173)
`cess rates increase. However, these savings would have
`to be balanced against al!ly additional costs associated
`with a better preclinical screening process.
`S uccess rates for self-originated NCEs differed sig(cid:173)
`nificantly by therapeutic class. Predicted or actual final
`success rates varied from 12% for respiratory drugs to
`33% for anti-infectives. Car<liovascular and central ner(cid:173)
`vous system drugs also had predicted success rates that
`were substantially below that for anti-infectives. Some
`of the differences in success rates by therapeutic class
`might be explained generally by differences in the uncer(cid:173)
`tainty with which regulatory standards would be satis(cid:173)
`fied. For example, efficacy end points for anti-infectives
`are usually clearly defined and relatively easy to assess.
`In contrast, the difficulties in establishing efficacy for
`psychotropic compounds have been well described.24,25
`The length of time that an NCE spent in clinical test(cid:173)
`ing or regulatory review before the fate of the drug
`(abandonment or approval) was determined decreased
`during t he study period. Estimated median survival
`times for self-originated NCEs decreased 0.6 years for
`IND filings in the early 1990s compared with those a
`decade earlier. These results are consistent with data on
`shorter US clinical development times for late 1990s
`approvals.2,3 In addition, our data on the time to research
`termination for compounds that have been abandoned
`suggest that pharmaceutical firms have been abandon(cid:173)
`ing unsuccessful compounds more quickly. Faster fail(cid:173)
`ures and shorter development times for drugs that do get
`approved imply, other things being equal, lower research
`and development costs per approved new drug. How(cid:173)
`ever, these gains can easily be offset if the out-of-pocket
`costs of conducting clinical trials have increased.
`Our data on clinical phase attrition rates not only sup(cid:173)
`port the hypothesis that pharmaceutical firms have
`acted more quickly in terminating developmelll on
`unsuccessful compounds but also allow us to better pin(cid:173)
`point when in the process these gains were made.
`Development costs are reduced more if a compound
`
`that ultimately fails is abandoned sooner. Our results
`indicate that firms have indeed tended to abandon their
`failed compounds earlier in the process. Reductions in
`failure rates for phase UI and regulatory review appear
`to be associated with corresponding increases in fail(cid:173)
`ure rates for phase I. It should be noted, however, that
`quicker decisions to abandon projects may also increase
`the likelihood of making a type II error (accepting the
`hypothesis that an investigational drug will not meet
`efficacy and safety standards and earn a reasonable
`return when in fact it would have done so if pursued).
`Furthermore, failure rates for phase II testing remained
`essentially constant. Some expensive phase III trials
`may be avoided if phase II testing can be made more
`informative so as to weed out more of those compounds
`that w ill fail to achieve regulatory approval.
`Om results indicate that conunercial factors became
`relatively more important over time as the primary rea(cid:173)
`son for abandoning development of investigational
`NCEs. Censoring may affect the results for the more
`recent time periods. NCEs that fai led for economic rea(cid:173)
`sons, however, tended to last longer in testing than
`NCEs that fai led for efficacy or safety reasons. Thus
`the censoring in the data suggests that the final results
`will show that the tre nd for economics is even steeper
`than currently observed. Given that economic factors
`increased in importance as a reason for research termi(cid:173)
`nation and that these commercial considerations have
`tended to be a deciding factor relatively late in the
`development process, the improvement in attrition rates
`that we have observed is all the more impressive.
`Cl inical success rates and phase attrition rates for
`new drugs are important indicators of how effectively
`pharmaceutical firm s are using the resources that they
`devote to research amd development. The proficiency
`with which this is done is a consequence of a complex
`set of regulatory, economic, and firm -specific factors.
`Reliable success rate and phase attrition rate estimates
`are an important tool for evaluation of the efficiency
`with which industry conducts clinical drug develop(cid:173)
`ment. Our results on the risks in drug development
`should aid in this process.
`
`Page 9 of 11
`
`
`
`306 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`,\1}\y 2001
`
`References
`I. DiMasi JA, Seibring MA, Lasagna L. New drug develop(cid:173)
`ment in the United States from 1963 to 1992. Clin Phar(cid:173)
`macol Ther 1994;55:609-22.
`2. Kaitin KI, Healy EM. The new drug approvals of 1996,
`1997 , and 1998: drug development trends in the user fee
`era. Drug Inf J 2000;34: 1-14.
`3. Kaitin KI, DiMasi JA. Measuring the pace of new drug
`development in the user fee era. Drug lnf J 2000;34:673-80.
`4. DiMasi JA. New drug innovation and pharmaceutical
`industry structur