Risks in new drug development: Approval
`success rates for investigational drugs
`
`Joseph A. DiMasi, PhD Boston, Mass
`
`The drug developmentprocess is knownto be com-
`plex, costly, and time-consuming.!-3 The processis also
`risky in that most compoundsthat undergoclinicaltest-
`ing are abandoned without obtaining marketing
`approval. The rate at which pharmaceutical firms mar-
`ket new therapies in the United States is an important
`measure of the viability of the drug development
`process.‘ The cost of new drug developmentis also crit-
`ically dependent on the proportion of drugs that fail in
`clinical testing.>-7 Estimates of industry successrates
`can be used in benchmarkingexercises for project plan-
`ning purposes. Given the length and cost of the drug
`developmentprocess, careful consideration ofall fac-
`tors that have a significant impact on the processis
`neededto appropriately allocate research and develop-
`ment resources.
`
`In a series of studies of new drug developmentin the
`United States, the Tufts Center for the Study of Drug
`Development
`(CSDD) and others have provided
`descriptive data on how cumulative successrates for
`new chemical entities (NCEs) vary with time from
`investigational new drug application (IND)filing.!.8-!4
`Several studies have also examined clinical success
`rates for biotechnology-derived drugs.!5-17 Statistical
`modeling can be helpful in analyzing successrates for
`recent periods because many of the compoundswillstill
`be in active testing at the time of the analysis. Tufts
`CSDDhasalso conducted a numberofstudies that use
`this approachto 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-
`mation Association.
`
`Received for publication Nov 6, 2000; accepted Feb 26, 2001.
`Reprint requests: Joseph A. DiMasi, PhD, Tufts Center for the Study
`of Drug Development, Tufts University, 192 South St, Suite 550,
`Boston, MA 02111.
`Clin Pharmacol Ther 2001;69:297-307.
`Copyright © 2001 by Mosby, Inc.
`0009-9236/2001/$35.00 + 0
`13/1/115446
`doi:10.1067/mep.2001.115446
`
`of compoundsfor which the ultimate fate of some of
`the compoundsin the dataset is not known.47.18-20
`This study provides updated successrate analyses for
`NCEs. Success rate trends and variations in success
`
`rates by therapeutic class are presented. The hypothe-
`sis that pharmaceutical firms have been moving com-
`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 in
`developmentare given.
`
`METHODS
`
`Data used for this study were obtained primarily
`from a Tufts CSDD databasethat 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
`NCEsfirst investigated in humans anywherein the
`world or NCEsfor which they werethefirst to file a
`US IND since 1963. The data gathered include INDfil-
`ing dates, the dates on which IND research was aban-
`doned, reasons for termination of research, the latest
`phase compounds were in when research was aban-
`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.?!.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 abandonmentand approval to pre-
`dict future success rates. Analyses were conducted for
`NCEswith INDsfirst filed in 3- and 6-year periods
`from 1981 to 1992. Data on more recent INDs were
`available but, given the length of the NCE development
`
`297
`
`ALL 2029
`MYLAN PHARMACEUTICALS V. ALLERGAN
`IPR2016-01131
`
`1
`
`ALL 2029
`MYLAN PHARMACEUTICALS V. ALLERGAN
`IPR2016-01131
`
`

`

`298 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`MAY2001
`
`process, they are too recent to use for a comprehensive
`statistical analysis of successrates.
`Inclusion criteria. For purposes of this study, an
`NCEis defined as a new molecular compound notpre-
`viously tested in humans. Excludedare new salts and
`esters of existing compounds, surgical and diagnostic
`materials, vaccines and other biologic agents, certain
`externally used compounds (such as disinfectants,
`antiperspirants, and sunscreens), and nutritional com-
`pounds(such as natural forms of vitamins and sweet-
`ening agents). Our definition of an NCEdiffers 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
`doesnot.
`Statistical analysis of success rates. For the statisti-
`cal analyses, residence time (the length of time from
`INDfiling to either abandonmentof research without
`marketing approval or
`to new drug application
`approval) was calculated for NCEs with INDsfirstfiled
`in successive 3-year intervals from 1981 to 1992.
`Approval dates were available through December31,
`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 INDfiling.
`NCEs werestratified according to source (self-
`originated versus licensed-in or otherwise acquired) and
`therapeutic class. An NCEis defined asself-originated if
`it was developed entirely under the auspices of the
`responding firm. We define acquired NCEs to be com-
`poundsthat were obtained by the developing firm through
`licensing, purchase, barter, or other means. To determine
`whethertrends in successrates exist, we analyzed the data
`by the period during which the IND wasfiled.
`Predicted success rates for IND filing periods were
`determined from a 2-stage model of the approval
`process. NCEswith researchstill active as of Decem-
`ber 31, 1999, constitute right-censored observations for
`our data set. Survival analysis can make useofinfor-
`mation provided by censored data.23 NCEs were
`assumedto survive until either research was terminated
`
`without approval or marketing approval wasachieved.
`Details of the selected models and the computational
`approachusedto estimate final success rates are pro-
`vided in the Appendix.
`The survey data also provided information on thelat-
`est development or regulatory phase that abandoned
`NCEswere in at the time of termination. These data
`allow us to determine the distribution of researchter-
`
`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-
`poundenters a given clinical phase and phaseattrition
`rates (the percentage of compoundsthat enter a phase
`that are abandonedbefore the next phase is initiated).
`
`RESULTS
`
`Included in the CSDD database ofinvestigational
`compoundsare the developmenthistories of 671 NCEs
`for which survey firmshadfiled a first IND from 1981
`to 1992. Of these, 508 wereidentified as self-originated
`and 163 wereidentified as acquired. Of the 508self-
`originated NCEs, 350 were initially investigated in
`humansin the United States. By the end of 1999, 20.9%
`of the NCEs with INDs filed from 1981 to 1992 had
`been approved for marketing in the United States. For
`this period, the current US approval successrates for
`NCEsthat were acquired,self-originated, and self-orig-
`inated andfirst tested in humansin the United States
`are 33.1%, 16.9%, and 8.6%, respectively. These results
`illustrate the significance of previous testing on mea-
`sured USsuccessrates; success rates on IND filings are
`higher for compounds that were licensed-in or first
`tested abroad.
`Time to research termination. Even though someof
`the drugsin our databasearestill active, survival analy-
`sis can be used to establish the rates at which the NCEs
`with INDsfiled 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 INDsfirst filed during the periods from
`1981 to 1983, 1984 to 1986, 1987 to 1989, and 1990 to
`1992 are shown in Fig 1. Because NCEsin 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 NCEsthat will eventually be ter-
`minated, the impact should be muchless on the median
`than on the mean.
`Even with these qualifications, the results at least
`suggest that, over time, pharmaceutical firms have
`made quicker decisions on research failures. Meanres-
`idence time decreased 30% (1.5 years) from the
`1981-1983 to the 1990-1992 INDfiling intervals.
`Median time to research abandonmentdecreased 20%
`(0.8 years) for INDsfiled in the early 1990srelative to
`the early 1980s.
`Further evidence that the ultimate fate of investiga-
`tional NCEs has tended to be resolved more rapidly
`
`2
`
`

`

`CLINICAL PHARMACOLOGY & THERAPEUTICS
`VOLUME69, NUMBER 5
`
`DiMasi 299
`
`
`
`
`
`YearsfromINDFiling
`
`
`ihn
`
`1981-1983
`
`1984-1986
`
`1987-1989
`
`1990-1992
`
`Period of IND Filing
`@ Mean & Median
`
`Fig 1. Mean and median time to research abandonmentforself-originated new chemicalentities
`(NCEs) withafirst investigational new drug application (IND)filed during a given period.
`
`0.7
`
`0.6
`0.5
`
`
`
`
`
`
`
`
`
`
`
`1.0
`
`0.9
`
`0.8
`
`0.4
`
`0.3
`
`0.2
`
`0.1
`
`0.0
`
`Probability
`
`
`
`
`
`0123 4
`5
`6
`7
`8
`9 10 11 12 13 14 15
`Years from IND Filing
` —* 1981-1983 —m 1990-1992
`
`Fig 2. Estimated survival curves for self-originated NCEswithafirst IND filed during a given
`period. The curves show the percentage of NCEsthat had not been abandonedor approved for mar-
`keting in the United States (ie, still active) a given numberof years from the date of IND filing.
`The data werefitted to Weibull distributions.
`
`over time is shown in Fig 2. The curves in the figure
`are estimated survival curves for the 1981-1983 to
`1990-1992 IND filing intervals. A point on the curve
`represents the probability that an investigational NCE
`will still be active a given numberof years from IND
`filing. An NCEis inactive at a given point in time if
`either research has been abandoned without marketing
`approval or the compound has received FDA approval
`for marketing. It should be noted that the estimated sur-
`vival curves account for censored data; that is, infor-
`
`mation regardingstill active NCEsis used to estimate
`final survival rates.
`Median survival time decreased from 4.9 years to 4.3
`years (12%) for the 1981-1983 to 1990-1992filing
`intervals, respectively. Faster action is also evident in
`the figure for different amounts of time from IND fil-
`ing. The percentages of NCEsfor the 1990-1992filing
`periodthatarestill active are 6 to 7 percentage points
`lower than those for the 1981-1983 filing period at 4
`to 10 years from IND filing.
`
`3
`
`

`

`300 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`MAY2001
`
`
`
`
`
`
`
`
`
`PercentApproved
`
`1984-1986
`1987-1989
`Period of IND Filing
`
` |OSelf-Originated Acquired BAll|
`
`
`
`
`1981-1983
`
`1990-1992
`
`Fig 3. Currentclinical approval success rates for NCEsby origin and period during whicha first
`IND wasfiled.
`
`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-
`ditional probability of approval at given residence
`times. The parameter estimates for both stages of the
`modelare highly statistically significant, and good-
`ness-of-fit measures indicate strong agreement with
`the data. The parameter estimates used to determine
`the predicted final success rates reported here and the
`accompanyingstatistical results are available upon
`request.
`Current success rates (as of December 31, 1999) for
`self-originated, acquired, and all NCEs by IND filing
`interval are shown in Fig 3. Licensed compounds gen-
`erally have undergone sometesting before licensing
`and have been shownto be promising candidates for
`marketing approval. The results support the hypothesis
`of such a screening effect for acquired NCEs; current
`successrates for acquired NCEsare notably higher than
`those for self-originated NCEs.
`A screening effect also appears to apply to self-
`originated compoundsthat have undergone someclini-
`cal testing abroad before an IND hasbeenfiled in the
`United States. The success rates for self-originated
`NCEsthat were first tested in humansin the United
`States are much lowerthan the successrates forall self-
`originated NCEs. Current success rates by IND filing
`interval for self-originated NCEsfirst tested in the
`United States are 33% to 65% lowerthan forself-
`originated NCEsas a whole.
`Censoring has an impacton the results for all IND
`filing intervals, but the effect is much greater for the
`morerecent intervals. The proportions of NCEsthat are
`
`currently active are substantially higher for these later
`periods. Thus the lower current successrates for self-
`originated NCEs in the 1987-1989 and 1990-1992
`intervals may simply reflect the shorter amount of time
`available for the ultimate fate of those NCEs to have
`occurred. Trend analysis for these later periods must be
`aided by the applicationofstatistical techniquesto fore-
`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 INDfiling interval are shown in Fig 4. The pre-
`dicted final success rates fall between current and max-
`imum possible success rates for all filing intervals.
`Although both predicted and maximum possible suc-
`cess rates are lower for the 1987-1989 interval rela-
`tive to 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
`successrate.
`
`Comparison of predicted and actual successrates for
`the early time periods can validate the performance of
`the statistical model. For NCEs with INDs first filed
`from 1981 to 1983, the model predicts a cumulative
`successrate of 19.5% at 16 years from IND filing (the
`maximum amountof time available for all compounds
`in the group); the actual success rate for this group at
`16 years from INDfiling is 19.8%. Similarly, NCEs
`with INDsfirst filed from 1984 to 1986 have a pre-
`dicted successrate of 18.8% at 13 years from INDfil-
`ing and an actual successrate of 19.4%.
`Therapeutic classes. Previous research hasindicated
`that success rates for NCEs vary by therapeutic
`
`4
`
`

`

`CLINICAL PHARMACOLOGY & THERAPEUTICS
`VOLUME69, NUMBER 5
`
`DiMasi
`
`301
`
`
`
`
`
`PercentApproved
`
`
`
` 12.3
`
`
`31.1 997
`
`225 19.1
`
`15.9
`
`1981-1983
`
`1984-1986
`
`1987-1989
`
`1990-1992
`
`Period of IND Filing
` OCurrent @ Maximum @ Predicted Cee
`
`Fig 4. Current (as of December 31, 1999), maximum possible,and predictedfinal clinical approval
`successrates for self-originated NCEs by period during whicha first IND wasfiled. Maximum pos-
`sible success rates were determined under the assumptionthat all active compoundsare eventually
`approved for marketing. Predicted success rates were constructed with use of estimates for a sur-
`vival analysis of residence time (time from IND filing to abandonmentor US marketing approval)
`with a Weibull 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 successrates by therapeutic class for self-originated NCEs with INDsfirst
`filed from 1981 to 1992*
`
`Therapeutic class
`
`NCEs
`
`Approved NCEs
`
`Open NCEs7
`
`Current
`success ratey
`
`Maximum
`success rate¢
`
`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,Newchemical entity.
`*Therapeutic class information is missing for five compounds.
`+As of December 31, 1999.
`Assumesthat all open NCEswill 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.®.20 The current and maximum possible success
`rates by IND filing interval for self-originated NCEs in
`9 specific therapeutic categories are shownin TableI.
`Because the number of compoundsavailable for analy-
`sis is greatly reduced whenthedataarestratified into
`therapeutic
`categories,
`the
`entire
`study period
`(1981-1992) is used. For the immunologic and respi-
`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
`compoundswith IND 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 weresufficiently good forus to estimate
`predicted final success rates for the analgesic/anesthetic,
`anti-infective, cardiovascular, and central nervous sys-
`tem categories. The current, maximum possible, and
`predicted final success rates for these 4 classes are
`shownin Fig 5. Relative success rate results for these
`
`5
`
`

`

`302 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`MAY2001
`
`
`
`
`
`
`
`PercentApproved
`
`Cardiovascular
`
`CNS
`
`Analg/Anesth
`
`
`
`
`| Current & Maximum @ Predicted|
`
`Antiinfective
`
`Fig 5. Current (as of December 31, 1999), maximum possible, and predicted final clinical approval
`successrates by therapeutic class for self-originated NCEs with a first IND filed from 1981 to 1992.
`Maximum possible success rates were determined under the assumption that all active compounds
`are eventually approved for marketing. Predicted success rates were constructed with use ofesti-
`mates for a survival analysis of residence time (time from IND filing to abandonment or US mar-
`keting approval) with a Weibull distribution specification and estimates for the conditional proba-
`bility of approval for a given residence time with a probit specification.
`
`
`
`
`
`Percent
`
`52.9
`
`11.9
`
`
`
`
`
`1981-1986
`
`1987-1992
`
`Period of IND Filing
`
`| OPhase!| GPhasell PhaseIIV/RR|
`
`Fig 6. Distribution of research terminationsfor self-originated NCEsbyclinical phase and period
`during whicha first IND wasfiled.
`
`classes are likely unaffected by time trends inasmuch
`as the numberoffilings for the last half of the study
`period as a percentage oftotal filings for the whole
`period for each of these 4 classes varied only from 47%
`to 55%. The predicted success rates range from approx-
`imately | in 5 for cardiovascular NCEsto | in 3 for
`anti-infectives.
`Clinical phaseattrition rates. Clinical approval suc-
`cess rates yield patterns of success for the clinical
`
`developmentprocess as a whole, but they do not inform
`us of success and failure patterns during the clinical
`developmentprocess. Our data on the latest phase that
`an abandoned NCEwasin 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 failures
`occurin phaseII. This is the case for both the first and
`second halves of the study period. Forthe later IND fil-
`
`6
`
`

`

`CLINICAL PHARMACOLOGY & THERAPEUTICS
`VOLUME69, NUMBER 5
`
`DiMasi 303
`
`
`
`
`
`
`
`Percent
`
`1981-1986
`
`1987-1992
`
`Period of IND Filing
`OPhase! Phasell_ @ Phase Ill|
`
`Fig 7. Approval success rates for self-originated NCEsentering a given clinical phase.
`
`
`
`
`Percent
`
`1981-1986
`
`1987-1992
`
`Period of IND Filing
`|CPhase! Phase ll @ PhaseIll
`
`Fig 8. Phaseattrition rates (percentage of compoundsentering a phasethatfail in the phase) for
`self-originated NCEs by period during whichafirst IND wasfiled.
`
`ing period, however, proportionately moreresearchfail-
`ures occurred in phase I and proportionately fewer
`occurred in phaseIII or regulatory review.
`Statistical analysis yields predicted final success
`rates for self-originated NCEsfor the 1981-1986 and
`1987-1992filing intervals of 24.2% and 22.6%, respec-
`tively. Current approval and termination rates for these
`periods, along with the assumption that currently active
`NCEsthat are predicted to eventually fail will do so in
`phase II or regulatory review, allow us to predict
`approval rates for NCEsthatenter a clinical phase (Fig
`7). Although approval rates are similar for the early
`clinical phases in both periods,
`the likelihood of
`approvalincreased by 5.6 percentage points for phase
`
`III. This is consistent with the results displayed in Fig
`6, which showedrelatively more terminations in phase
`I and relatively fewer in phaseIII orlater.
`The data on research terminations by phase and pre-
`dicted success rates also allow us to determine phase
`attrition rates. Fig 8 showsthatattrition rates are great-
`est in phase II in which morethan half of the investigated
`compoundsfail. During the study period, failure rates
`increased for phases I and II but declined for phase III.
`Reasons for research abandonment. The database
`contained information on the reasons research was
`abandoned for NCEsthat had research terminated with-
`out marketing approval. We grouped the responsesinto
`3 major categories: safety (eg, “human toxicity”or “ani-
`
`7
`
`

`

`304 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`MAY2001
`
`
`
`Percent
`
`
`
`
`
`
`Economics
`
`Efficacy
`
`Safety
`
`Other
`
`@ 1981-1986 M 1987-1992
`
`
`
`Fig 9. Percentage of research terminations for all NCEs byperiod offirst IND filing and by pri-
`mary reason for abandonment.
`
`mal toxicity’), efficacy (eg,“activity too weak”or “lack
`of efficacy”), and economics (eg, “commercial market
`too limited” or “insufficient return on investment’). A
`relatively small numberof the compoundsthat had been
`abandonedhadreasons for termination that were not
`specific enoughto be placed in 1 of these 3 categories.
`The sharesof all reasons for abandonmentfor each of
`these categories by IND filing interval are shown in
`Fig 9.
`Forthe last half of the study period, economic and
`efficacy issues becamerelatively more prevalent, while
`safety issues becamerelatively less prevalent, as rea-
`sons for research termination. Becausethe time avail-
`able for the fate of the compoundsto have been deter-
`minedis limited, the abandonmentresults for the inter-
`val from 1987 to 1992 are biased toward causes that
`tend to be revealed relatively soon after filing. This
`censoring effect also applies to the earlier interval but
`with much less impact. The economic share increased,
`even though research on NCEsterminated for eco-
`nomic reasonstendsto occurlater in the development
`processthan is the case for safety and efficacy (eg, for
`filings from 1981 to 1986, 45% of the economicter-
`minations occurred at least 6 years from filing com-
`pared with 35% of efficacy and 17% of safety termi-
`nations).
`The censoring effect also applies when the data are
`analyzed by the phase that a compoundwasin whenit
`was abandoned. This bias will tend to be lowerif ear-
`lier periods are examined. Consideringthefirst half of
`the study period (NCEsthat had an IND first filed from
`1981 to 1986), compoundsthat had failed for economic
`or efficacy reasons were terminated much morefre-
`
`quently in late clinical testing phases. The percentage
`of failed compoundsthat were abandonedin 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 abandonmenttimes
`for all NCEs by INDfiling period and by the primary
`reason for termination. Average times to abandonment
`are lowerfor the later filing period, but this can result
`in part from the shorter period during which abandon-
`ments can occur for this interval. For either period,
`however, both the mean and median time to research
`abandonmentis 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-
`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%
`wererelated to efficacy issues, and only 16% were for
`safety problems (13% were for other reasons).
`
`DISCUSSION
`
`A statistical model of the rate at which new drugs
`proceed throughclinical testing to marketing approval
`wasestimated for three 4-year and two 6-year IND fil-
`ing intervals. Estimated approval successrates for self-
`originated NCEs varied from 19% to 30% during the
`study period. The highest predicted success rate was
`for the most recent filing period (1990-1992). The
`results suggest that approval rates have not declined
`over time and, quite possibly, have increased. A gen-
`eral improvementin successrates can result from bet-
`
`8
`
`

`

`CLINICAL PHARMACOLOGY & THERAPEUTICS
`VOLUME69, NUMBER 5
`
`DiMasi
`
`305
`
`Table II. Time to research abandonment(in years) for NCEs by INDfiling period
`1981-1986
`
`1987-1992
`
`Reason
`
`Economics
`Efficacy
`Safety
`Other
`
`n
`
`64
`71
`46
`34
`
`Mean (y)
`
`Median(y)
`
`44
`3.6
`2.6
`3.5
`
`4.0
`2.3
`2.5
`2.3
`
`n
`
`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 preclinical screening. The implications for the devel-
`opmentprocess are significant because the clinical
`costs for someresearchfailures will not be borne if suc-
`
`cess rates increase. However, these savings would have
`to be balanced against any additional costs associated
`with a better preclinical screening process.
`Success rates for self-originated NCEs differed sig-
`nificantly by therapeutic class. Predicted or actual final
`successrates varied from 12% for respiratory drugs to
`33% for anti-infectives. Cardiovascular and central ner-
`vous system drugs also had predicted success rates that
`weresubstantially below that for anti-infectives. Some
`of the differences in success rates by therapeutic class
`might be explained generally by differences in the uncer-
`tainty with which regulatory standards would besatis-
`fied. For example, efficacy end points for anti-infectives
`are usually clearly defined andrelatively easy to assess.
`In contrast, the difficulties in establishing efficacy for
`psychotropic compoundshave been well described.24.25
`The length of time that an NCEspentin clinical test-
`ing or regulatory review before the fate of the drug
`(abandonmentor approval) was determined decreased
`during the study period. Estimated median survival
`times for self-originated NCEs decreased 0.6 years for
`INDfilings in the early 1990s compared with those a
`decadeearlier. 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 compoundsthat have been abandoned
`suggest that pharmaceutical firms have been abandon-
`ing unsuccessful compounds more quickly. Faster fail-
`ures and shorter developmenttimes for drugs that do get
`approved imply, other things being equal, lower research
`and development costs per approved new drug. How-
`ever, these gains can easily be offset if the out-of-pocket
`costs of conducting clinical trials have increased.
`Ourdata on clinical phaseattrition rates not only sup-
`port the hypothesis that pharmaceutical firms have
`acted more quickly in terminating development on
`unsuccessful compoundsbutalso allow usto better pin-
`point when in the process these gains were made.
`Development costs are reduced more if a compound
`
`that ultimately fails is abandoned sooner. Ourresults
`indicate that firms have indeed tended to abandontheir
`failed compoundsearlier in the process. Reductionsin
`failure rates for phase III and regulatory review appear
`to be associated with corresponding increasesin fail-
`ure rates for phaseI. It should be noted, however, that
`quicker decisions to abandonprojects 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 whenin fact it would have doneso if pursued).
`Furthermore, failure rates for phase II testing remained
`essentially constant. Some expensive phaseIII trials
`may be avoided if phase II testing can be made more
`informative so as to weed out more of those compounds
`that will fail to achieve regulatory approval.
`Ourresults indicate that commercial factors became
`relatively more important over time as the primary rea-
`son for abandoning development of investigational
`NCEs. Censoring mayaffect the results for the more
`recent time periods. NCEsthat failed for economicrea-
`sons, however, tended to last longer in testing than
`NCEsthat failed for efficacy or safety reasons. Thus
`the censoring in the data suggests that the final results
`will show that the trend for economics is even steeper
`than currently observed. Given that economicfactors
`increased in importance as a reason for research termi-
`nation and that these commercial considerations have
`
`tended to be a deciding factor relatively late in the
`developmentprocess, the improvementinattrition rates
`that we have observedis all the more impressive.
`Clinical success rates and phaseattrition rates for
`new drugs are important indicators of how effectively
`pharmaceutical firms are using the resourcesthat they
`devote to research and 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 phaseattrition rate estimates
`are an importanttool for evaluation of the efficiency
`with which industry conducts clinical drug develop-
`ment. Our results on the risks in drug development
`should aid in this process.
`
`9
`
`

`

`306 DiMasi
`
`CLINICAL PHARMACOLOGY & THERAPEUTICS
`MAY2001
`
`References
`1.
`
`10.
`
`DiMasiJA, Seibring MA, Lasagna L. New drug develop-
`mentin the United States from 1963 to 1992. Clin Phar-
`macol Ther 1994;55:609-22.
`. Kaitin KI, Healy EM. The new drug approvals of 1996,
`1997, and 1998: drug developmenttrendsin the user fee
`era. Drug Inf J 2000;34: 1-14.
`. Kaitin KI, DiMasi JA. Measuring the pace of new drug
`developmentin the user fee era. Drug Inf J 2000;34:673-80.
`. DiMasi JA. New drug innovation and pharmaceutical
`industry structure: trends in the output of pharmaceutical
`firms. Drug Inf J 2000;34:1169-94.
`. DiMasi JA, Hansen RW, Grabowski HG, Lasagna L. Cost
`of innovation in the pharmaceutical industry. J Health
`Econ 1991;10:107-42.
`. DiMasi JA, Hansen RW, Grabowski HG, Lasagna L.
`Research and developmentcosts for new drugs by thera-
`peutic category: a study of the US pharmaceutical indus-
`try. Pharmacoecononics 1995;7:152-69.
`. DiMasi J, Grabowski HG, Vernon J. R&D costs, innova-
`tive output and firm size in the pharmaceutical industry.
`Int J Econ Bus 1995;2:201-19.
`. Wardell WM,Hassar M, Anavekar SN, Lasagna L. The
`rate of development of new drugs in the United States,
`1963 through 1975. Clin Pharmacol Ther 1978;24:133-45.
`. Wardell WM,DiRaddoJ, Trimble AG. Developmentof
`new drugs
`originated
`and
`acquired by United
`States-owned pharmaceutical firms, 1963-1976. Clin
`Pharmacol Ther 1980;28:270-7.
`Wardell WM, May MS, Trimble AG. New drug develop-
`ment by US pharmaceutical firms with analyses of trends
`in the acquisition and origin of drug candidates, 1963-
`1979. Clin Pharmacol Ther 1982;32:407-17.
`Mattison N, Trimble AG, Lasagna L. New drug develop-
`mentin the United States, 1963 through 1984. Clin Phar-
`macol Ther 1988;43:290-301.
`DiMasi JA, Bryant NR, Lasagna L. New drug develop-
`mentin the United States from 1963 to 1990. Clin Phar-
`macol Ther 1991;50:471-86.
`US Congress, Office of Technology Assessment. Phar-
`maceutical R&D: costs, risks, and rewards. Washington:
`US GovernmentPrinting Office, 1993.
`Tucker SA, Blozan C, Coppinger P. The outcome of
`research on new molecular entities commencingclinical
`research in the years 1976-79; OPE Study 77. Rockville
`(MD): US Food and Drug Administration, Office of Plan-
`ning and Evaluation; 1988.
`Bienz-Tadmor B, DiCerbo PA, Tadmor G, Lasagna L.
`Biopharmaceuticals and conventional drugs: clinical suc-
`cess rates. Biotechnology (NY) 1992;10:521-5.
`Struck MM.Biopharmaceutical R&D success rates and
`developmenttimes. Biotechnology (NY) 1994;12:674-7.
`Gos