RESEARCH SYNTHESIS: TIME AND SUCCESS RATE OF PHARMACEUTICAL R&D
v1.0 researched and written by Ryan Kimmitt and Marcela Vieira, reviewed by Suerie Moon, copyedited by Anna Bezruki. Last updated July 2020
The literature on research and development (R&D) timeframes and success rates in the pharmaceutical sector is considerable.* Most of the literature focuses on the development of drugs and on the clinical development stage of the process. The topic is a key component in pharmaceutical pricing models and current debates on the productivity of R&D.
Pharmaceutical research and development, Pharmaceutical R&D, drug development, medical product development, clinical success rates, transition rates, attrition rates, success rates, likelihood of success, LOA, probability of success, POS, clinical phase length, phase length, phase timelines, clinical timelines.
Search was conducted using a combination of search mechanisms, mainly in English, with no specific time period of publication.
Synthesis of the literature
Research and development for medical products is a longstanding topic of research. Estimating costs and time to produce new health technologies has wide implications for the pharmaceutical industry, governments, and global health policies as a whole. Accurate estimation of time spent in each development stage as well as success rates could allow actors to be more efficient in their allocation of resources to product development.
This research synthesis examines success rates and timeframes of pharmaceutical R&D . While there is public data available on clinical trial and regulatory approval length, there is a dearth of data on time spent in preclinical development, limiting some of the estimated timeframes. Success rates and timeframes are combined to lay the foundation for predictive models that estimate the cumulative probability of approval for a product – a critical component of product development planning. The main reasons for development termination are also explored.
Though the research is aimed at all health technologies and all organizations conducting product development, the literature is dominated by projects related to the development of new drugs, with some studies providing information disaggregated by therapeutic class, and by pharmaceutical industry data. In addition, the conspicuous absence of research relating to different kinds of organizations, development processes outside the United States, and different products outside of strictly pharmaceuticals are noticeable lacunae in the literature.
Table 1. Summary of recent estimates of success rates and timeframes of new drug development
Summary of the contents
This research synthesis is organized into the following topics:
1) Estimation of historical success rates
- Overall – new drugs
- New drugs by therapeutic class
- Other technology types
2) Estimation of historical timeframes
- Overall - new drugs
- New drugs by therapeutic class
- Other technology types
1) Estimation of historical success rates
Overall - new drugs
The literature on success rates in pharmaceutical R&D varies in both its use of terminology and the way in which data are presented. In general, attrition rate is defined as the percentage of projects that entered a particular phase and passed to the subsequent phase, with different papers using different metrics, namely: phase transition rate, approval rate, likelihood of approval, probability of success, and failure rate. While the terminology varies, the methods used to calculate common success metrics have remained similar over time, allowing for historical comparisons. By pooling the literature, we are able to form a relatively comprehensive picture of the success rate trends from the 1960s through the mid-2010s.
Sheck et al. (1984) wrote a foundational piece detailing the clinical success rates for new chemical entities (NCEs) in the United States for different 3-year cohorts (from January 1st, 1963 through December 31st, 1979) and comparing them. The authors calculated success rates based on the number of drugs that obtained regulatory approval by the FDA in the 8-year period following receipt of permission to start human clinical trials (investigational new drugs - INDs). The results show a cumulative success rate of 9.4% for the period from 1963-65, 8.5% for 1966-1968, 13.7% for 1969-1971, and 9.5% for 1972-1974. The authors note that success rates could, in reality, be higher, as some projects were still under development at the time calculations were made. No cumulative success rates are provided for 1974-1979 as not enough time had elapsed since the IND had been obtained.
Using a similar methodology, DiMasi (2001) looked at roughly the next decade, from 1981 to 1992, using data obtained from a Tufts CSDD database with self-reported data from 24 US pharmaceutical companies. The author divided this period into 4 cohorts, with overall clinical success rates fluctuating between 20.5% and 23.2%, until the final period of 1990-1992, which saw a drop to 17.2%. This drop can mostly be accounted for by the fact that the time between the last IND filed during that period and the paper’s publication was significantly less time than it would typically take for an entire cohort of INDs to reach completion (either approval or termination). Because of this, DiMasi writes that “the results suggest that approval rates have not declined over time, and, quite possibly, have increased.” Clinical success rates are broken down for NCEs that were acquired, self-originated, and self-originated and first tested in humans in the United States, respectively at 33.1%, 16.9% and 8.6%. The results show that licensed-in compounds have higher success rates, and that self-originated compounds first tested outside of the United States are more likely to be successful.
Kola and Landis (2004) investigated the following decade, with data from 1991-2000 for ten big pharmaceutical companies in the United States and Europe. The authors estimated a cumulative success rate from Phase I through registration at 11%, a rate markedly lower than had been estimated for the previous decade by DiMasi (2001), and closer to the estimations by Sheck et al. (1984). The authors also provided figures by stage of development, showing a success rate of 60% for Phase I, 40% for Phase II, 59% for Phase III and 77% for registration. The analysis has a few caveats, namely that the small 4-year difference between the last IND date in the dataset and the publishing of the paper does not seem to be addressed.
Abrantes-Metz et al. (2004) also looked at data from the 1990s, covering a period from 1989 through 2002 using data from Pharmaprojects (a large industry drug development database). The study focuses on “new drugs”, including those composed of chemicals, biologicals, and natural products. They used a sophisticated model to calculate the success rates of each phase, finding significantly different rates than Kola and Landis (2004), despite covering a similar timeframe. The success rates were estimated at 80.7%, 57.7%, and 56.7% for Phases I, II, and III respectively. It should be noted that the registration is included in Phase III. The cumulative success rate was 26.4%. Figures are also separated by company size, showing higher success rates in Phase I and II for “non-big pharma”, while Phase III has higher success rate for “big pharma”. The study’s estimate does not have the characteristic dip in phase II success rates that appears in other studies, and the reason for the high phase II rate is not explicitly addressed in the paper.
Paul et al. (2010) developed a model of R&D productivity based on assumptions of attrition rates, timeframes, and costs for each phase of discovery and development. The model was constructed using data from 13 large pharmaceutical companies provided by the Pharmaceutical Benchmarking Forum, as well as the authors’ own internal data from Eli Lilly and Company. The probability of successful transition is presented for eight development phases: Target-to-hit: 80%, Hit-to-lead: 75%, Lead optimization: 85%, Preclinical: 69%, Phase I: 54%, Phase II: 34%, Phase III: 70% and Submission to launch: 91%.
DiMasi et al. (2010) also looked at the decade from 1993 to 2009 to estimate clinical approval probabilities using data from the 50 largest pharmaceutical firms (by sales). The authors found that clinical success rates began to drop in the mid-1990s and estimated an overall success rate of 19% for the entire period for all compounds and 16% if considering only self-originated drugs (compared to 11% for Kola and Landis, 2004). The paper details several interesting features of the data. For example, despite the two half-decade periods in the study having similar success rates from phase I to approval, the later cohort had lower success rates in earlier phases (I and II). The authors argued that clinical testing has become more complex, potentially explaining higher failure rates for earlier phases due to a higher standard of success in general during these phases. Another finding is that licensed-in compounds have higher success rates, since typically licensed-in compounds have already undergone some screening or testing prior to licensing.
Arrowsmith (2011a, 2011b and 2013) investigated success/failure rates in phase II, phase III and submission in the period from 2008 to 2012 in a series of three papers. The first (Arrowsmith 2011a) focused on phase II success rates from 2008 to 2010 and found that phase II has lower success rates than any other phase of development, as also found in previous studies. The author provided figures from previous periods analysed by the Centre for Medical Research (CMR) that show success rates for phase II at 28% for 2006-2007 and 18% for 2008-2009. The second paper (Arrowsmith 2011b) looked at Phase III success rates from 2007 to 2010. The success rates were around 50% (actual figure is not provided), a decline from previous years. The third article (Arrowsmith and Miller, 2013) looked at phase II and phase III failures together from 2011-2012. The paper states that Phase II success rates for new projects remained below 20%, but Phase III rates have improved 7% from 2009-2011 compared to 2007-2009 (actual figures are not provided). The authors concluded that the figures “may be an indication that the industry, as a whole, is designing Phase II programmes that are able to support early termination decisions and thereby avoiding a number of costly Phase III failures”.
Hay et al. (2014) similarly conducted a study from 2003 through 2011, continuing chronologically from where Abrantes-Metz et al. (2004) and DiMasi et al. (2010) ended. The study uses significantly more companies (850 compared with 50 for DiMasi, 10 for Kola and Landis, and an unknown number for Abrantes-Metz). The results show a clinical likeliness of approval (LOA) rate of 10.4%, comparable to Kola and Landis (2004). Hay et al. also found a dip in the probability of Phase II success (32.4% compared to 64.5% for Phase I and 60.1% for Phase III). Hay also calculated rates for lead indications and finds that a lead indication has a LOA rate of 15.3%, and compared this to lead indication rates of 19% for DiMasi, 11% for Kola and Landis, and 26.4% for Abrantes-Metz. Importantly, a lead indication may be changed, meaning that if a lead indication were to fail in Phase I testing, if another indication transitioned from Phase I to Phase II, the lead indication would then change to the second indication. This means that lead indication rates will always be greater than or equal to overall rates, causing upward bias.
Smietana et al. (2016), also using data from the Pharmaprojects database for a total of more than 9,200 compounds, calculated the cumulative success rates from Phase I to launch for five cohorts between 1996 and 2014. The LOA percentage dropped each period until 2011, and rebounded for the 2012-2014 cohort to reach 11.6% (1996-99: 16.4%; 2000-03: 10.8%; 2004-07: 10.0%; 2008-2011: 7.5%). The paper also finds Phase II success rates to be significantly lower than either Phase I or III, never crossing the 50% threshold (Phase I hovers around 65%; Phase III fluctuates between 55% and 70% across the five cohorts). For the last cohort (2012-2014), Phase I success rate is estimated at 58%, Phase II at 39% and Phase III at 67%. The study also separated probability of success for in-licensed and “non-partnered” compounds for the five cohorts, showing higher likelihood of approval for licensed-in compounds across the entire period (for the last period from 2012-2014, licensed-in compounds had 20% cumulative success rates from Phase I to launch versus 12% for non-partnered compounds. In comparison, for the first period from 1997-1999, it was 54% for in-licensed versus 13% for non-partnered).
Following their 2010 study, DiMasi et al. (2016) estimated phase transition probabilities (success rate) based on 1,442 compounds (first tested in humans from 1996 to 2007) self-originated from the top 50 pharmaceutical companies. Phase I to Phase II transition was estimated at 59.52%, Phase II to III at 35.52%, Phase III to submission at 61.95%, and submission to approval at 90.35%. The overall probability of success was estimated to be 11.83%. The authors concluded that “This success rate is substantially lower than the rate of 21.50% estimated for the previous study, but consistent with several recent studies of clinical success rates.”
Thomas et al. (2016), on a BIO, Biomedtracker and Amplion report, tracked pharmaceutical clinical success rates from 2006 through 2015 using data from 7,455 development programs, across 1,103 companies in the Biomedtracker database. The key findings of the report are a total cumulative success rate of 9.6% for the entire period (11.9% for non-oncology drugs). Consistent with other studies, the study found that Phase II success rate (at 30.7 percent) was lower than those of the other phases (63.2% for Phase I; 58.1% for Phase III).
Wong, Siah, and Lo (2019) conducted a study to calculate rates of success using a larger dataset and a new way to calculate success rates by phase. This study is a significant contribution to the literature because it incorporates big data, dwarfing datasets used by Kola and Landis, DiMasi, and others, and because the authors propose a new measure of phase success to account for missing information that should be counted as a success. For example, consider a drug that has a completed Phase I trial recorded on clinicaltrials.gov, and is listed in ongoing Phase III trials. If there is no Phase II trial available, under normal circumstances, the drug will not count as having passed Phase II. Because the typical “phase-to-phase” success rate is calculated as the cumulative number of drugs which advance from phase x to phase x+1 divided by the total number of drugs which entered phase x, missing data are not counted. However, it is a fairly common occurrence that drugs will have missing data or special regulatory pathways; in these instances, using standard approaches, successes will not be counted towards the aggregate total of success from these periods. In contrast, Wong, Siah, and Lo would interpolate a Phase II success in this example, recognizing that there is no way that a drug would have made it to Phase III while failing testing in Phase II (i.e., given the binary nature of trial outcomes, this can justifiably be recorded as a success). The authors also move away from a simple “phase-to-phase” calculation of probability of success (POS), which is common in the literature. The standard calculation defines POS as the product of each individual phase success (i.e., the probability of passing Phase I multiplied by the probability of passing Phase II, and so forth for each phase). The authors instead propose a “phase-by-phase” calculation, wherein all possible drug paths (each unique indication being tested for a particular drug constitutes one “path”) are measured and the proportion of paths that pass from Phase I to approval are counted as one success. Given the methodological differences between the authors and the established literature, success rates would be expected to be higher, and this is, in fact, the case. Wong et al.’s estimates are notably higher than the generally cited 10% figure for Phase I-to-approval success rates. The authors found rates of 66.4%, 48.6%, and 59.0% for Phase I, II, and III trials, respectively, and a rate of 13.8% for all drugs entering Phase I testing.
Dowden and Munro (2019) presented and analysed data for new active substances (including chemicals and biologicals) from CMR International (which operates a consortium of about 30 innovative biopharmaceutical, including large, mid-sized and small companies) for the period of 2010 to 2017. The study shows probability of success from each phase to launch (Phase I to launch, Phase II to launch, and Phase III to launch) and Phase II to Phase III for every two years of the period of the study. For the first period (2010-2012), Phase I to launch is estimated at 6% and for the last period (2015-2017) at 7%. Phase II to Phase III varies from 23% (2010-2012) to 25% (2015-2017).
In a recently published study, Pammolli et al. (2020) investigated drug development projects using and extensive data set of more than 50,000 projects between 1990 and 2017 from R&D Focus, a proprietary database. Only projects started in the US, Europe or Japan were included. Similar to Wong et al. (2019), if information was missing for one phase of the development process but there was information on a more advanced phase, it was counted as success. The authors found that attrition rates have been decreasing at all stages of clinical research in recent years, though they are still higher than in the decades spanning 1990-1999. Pammolli et al. (2020) also found a reduction in attrition rates over time for the preclinical stage. For the final period of 2010-2013, the estimated attrition rates (failures) for each development phase were: Preclinical: 89.5%, Phase I: 55.5%, Phase II: 80.4%, Phase III: 68.8% and Registration 28.7%. For comparison with other studies in this synthesis, we present these as success rates instead: Preclinical: 10.5%, Phase I: 44.5%, Phase II: 19.6%, Phase III: 31.2% and Registration: 71.3%. The success rates estimated by Pammolli et al. are significantly lower than other estimations available in the literature.
Pammolli et al. (2020) also investigated attrition rates across different organization types classified as “pharmaceutical”, “biotech” and “non-industrial” institutions. The calculations showed that “non-industrial” have higher attrition rates (lower success rates) across all development phases (except at registration), followed by “biotech” and “pharmaceutical.” The authors concluded that “biotechnology companies have reached levels of productivity in project development that are equivalent to those of large pharmaceutical companies.”
New drugs by therapeutic class
The literature shows that success rates can vary substantially according to therapeutic class. Several different papers disaggregate success rates for different therapeutic class and compare them to a baseline, showing that certain areas have higher success rates than others. All studies mention therapeutic class and disease type as important factors in the efficacy of the research and development process.
DiMasi (2001) found varying success rates among 10 different therapeutic classes. Analgesic/anaesthetic, anti-infective, and gastrointestinal candidates each had success rates above 20%, while candidates with respiratory, central nervous system, and immunologic indications had success rates around 15%. Anti-infectives had the highest success rate (28.1%); respiratory (12%) and miscellaneous (7%) had the lowest. According to the author, some of the variance in success rates can be explained by differences in the way regulatory standards are satisfied by different disease types. For example, efficacy endpoints may be less concrete for certain diseases, or surrogate endpoints may be used.
Kola and Landis (2004) disaggregated success rates by nine therapeutic classes and found considerable variation between them. Cardiovascular candidates had the highest rate of success (20%), followed by those for arthritis and pain (17%) and infectious diseases (16%), whereas central nervous system (8%), oncology (5%) and women’s health (4%) had the lowest. The study provided information by therapeutic class for each development stage (Phase I, Phase II, Phase III and registration), as shown in the figure below:
Source: Kola and Landis (2004)
Abrantes-Metz et al. (2004) provide success rates disaggregated by therapeutic area and type of technology. The authors found that biologicals had a success rate of 90% in Phase I, 67% in Phase II and 70% in Phase III; that chemicals had a success rate of 84% in Phase I, 66% in Phase II and 66% in Phase III; and that natural products had a success rate 90% in Phase I, 77% in Phase II and 61% in Phase III. The study also provides success rates for 14 therapeutic areas, varying between 70% (transdermal) and 97% (parenteral – subcutaneous) in Phase I; 43% (respiratory) and 81% (parenteral – subcutaneous) in Phase II, and 33% (anti-Alzheimer’s Disease) and 94% (anti-HIV/AIDS) in Phase III. As noted above, the authors included registration as part of Phase III.
DiMasi et al. (2010) ran a similar analysis, finding variance in phase transition rates and LOA rates between eight therapeutic classes. The rates were extremely low for therapies with small N, given that the time period for entering product development (1993-2004) was only 6 years prior to publishing. Nevertheless, the phase transition probabilities offer some insight into which types of therapies tend to perform more effectively than others in clinical testing. Immunologic drugs, for example, had a maximum success rate of 36%; the next highest class was about 20%. Musculoskeletal and miscellaneous drugs had estimated LOA rates of about 20%, while cardiovascular, CNS, GI/metabolism, and respiratory classes had estimated rates around 9%. The study also estimated clinical success rates for self‐originated compounds classified by small and large molecules and found that, over the entire study period, 13% of small molecules and 32% of large molecules succeeded.
Hay et al. (2014) addressed the differences in phase transition and LOA rates based on indication and disease type. They too found large differences in rates depending on the class of drugs, with infectious disease, autoimmune, and endocrine classes with the highest LOA rates (12-17%). On the other hand, neurology, cardiovascular, and oncology drugs had significantly lower success rates, from 7-9%. Oncology drugs in particular were shown to have a very low probability of success.
Additionally, the report by Thomas et al. (2016) found several differences in therapeutic class, specifically that oncology consistently had lower transition rates, going so far as to calculate the cumulative rates with and without oncology included (the LOA for all candidates was 9.6%, and 11.9% for all candidates except those with oncology indications). Of 14 major disease areas, the authors found that haematology (26.1%), infectious diseases, and ophthalmology had the highest cumulative success rates, while psychiatry, cardiovascular disease and oncology (5.1%) had the lowest.
The above-mentioned study by Smietana et al. (2016) provides disaggregated data for small molecules and biologic drugs. The authors found that cumulative success rates remain consistently higher for biologics across the entire period of the study (1996-2014), fluctuating from 5% to 16% for small molecules and 12% to 18% for biologics.
Wong, Siah, and Lo (2019) also disaggregate by class, finding a 3.4% probability of success for oncology candidates. Autoimmune and genitourinary candidates had the next lowest rates at around 15% each. Ophthalmology candidates and vaccines performed extremely well with a roughly 33% probability of success for each.
Dowden and Munro (2019) disaggregated their late-stage development success rate for new active substances targeting rare versus non-rare indications. There are some fluctuations across the study period (2010-2017), but in the initial and final periods, success rates for both indications were found to be very similar (2010-2012: 50% for rare and 49% for non-rare; 2015-2017: 61% for rare and 63% for non-rare). Data are also provided for seven different therapeutic areas, with anti-infectives showing the highest overall probability of success from Phase I to launch (16%) and nervous system candidates the lowest (3%).
Pammolli et al. (2020) disaggregated their data into 13 therapeutic classes and calculated average phase-by-phase attrition rates for 2000-2009 and 2010-2013. Taking Phase II as an example, in the second period, the therapeutic class with the highest attrition rate (i.e., the lowest success rate) was “genito-urinary system and sex hormones” and the one with lowest attrition rate (highest success rate) was “antiparasitic products, insecticides and repellents”.
Looking at product development for anti-malarial medicines at the Medicines for Malaria Venture (MMV), a non-profit organization, Burrows et al. (2017) estimated success rates in product development by phase from 2009–2014 for MMV and compared to benchmark data for anti-infectives. The authors conclude that “malaria drug discovery has an attrition rate that is no better and no worse than that in the pharmaceutical industry for anti-infectives overall, and significantly better than for other therapeutic areas, such as neurology and oncology”. Success rates for MMV were estimated at: Preclinical: 50%, Phase I: 70%, Phase IIa: 78%, Phase IIb: 75%, Phase III: 67% and Registration: 100%.
Other technology types
Davis et al. (2011) investigated the differences in success rate between prophylactic vaccines and pharmaceuticals overall. They found that, for the 1995 cohort of IND application submissions, the ratio of failures to successes was 8.3 in prophylactics vaccines and 7.7 in other drugs.
Terry et al. (2018) developed a modelling tool to estimate the costs of launching new health products called the Portfolio-To-Impact (P2I) Model. The model is based on assumptions for costs, timeframes and attrition rates for each phase of development from late preclinical stage to Phase III clinical trials. The assumptions were based on Parexel’s R&D cost sourcebook and refined by interviews “with a wide variety of stakeholders from Product Development Partnerships, biopharmaceutical and diagnostic companies, and major funders of global health R&D.” The model has different assumptions for different types of products, called “archetypes,” including vaccines, new chemical entities, repurposed drugs, biologics, and diagnostics. Young et al. (2018) further refined the P2I Model. A summary of the success rates per archetype using version 2 of the P2I Model is provided in the figure below.
Source: Young et al., 2018, p. 8.
2) Estimation of historical timeframes
The literature on phase lengths is more recent than that on success rates, becoming a recognized aspect of pharmaceutical development in the early 2000s. Earlier papers used survival analyses to understand development time, typically noting the cumulative proportion of products that reached an event (FDA approval) over some period of time. These calculations are still commonplace and are an important aspect of predicting overall success rates for drugs, but individual phase lengths have helped elucidate specific areas for improvement in the R&D process.
One of the first papers to look at phase lengths was Reichert (2003), which divided the total time spent in development into clinical and approval time. Historically, regulatory time averaged about 2 years in the 1970s, increasing to a high of 37 months in 1984-1985. The regulatory time then decreased steadily during the 1990s, to a low of 12.7 months in 1998-1999. Clinical time followed a similar pattern, taking about 50 months during the 1970s, increasing to about 75 months in the 1980s, and remaining high throughout the 1990s, reaching an apex of 92.5 months in 1994-1995, before decreasing steadily to a low of 63 months in 2000-2001. Consistent with literature on success rate, times also varied widely depending on therapeutic categories, but over the 4-year intervals there was no clear pattern for certain therapeutics taking more or less time compared to the other categories. The author also analysed the trends for approval times in view of three major pieces of legislature implemented in the U.S. in the 1980s: the Bayh-Dole Act, Orphan Drug Act, and Hatch-Waxman act. The patterns suggest that the legislative pieces succeeded at shortening FDA approval time.
Keyhani et al. (2006) investigated development times and the claim that rising drug prices in the US were justified because of longer development times. They investigated 168 drugs approved between 1992 and 2002 using data from publicly available sources (the authors highlight that most of the previous studies had been based on proprietary data). They found that the median clinical trial period was 5.1 years (61.2 months) and that the medial regulatory review period was 1.2 years (14.4 months). The authors concluded that “clinical trial periods have not increased during this time frame, and regulatory review periods have decreased. Therefore, it is unlikely that longer clinical trial times are contributing to rising prescription drug prices”.
The above-mentioned Abrantes-Metz (2004) paper also estimated phase duration in addition to success rates. The authors found mean duration times of 22.1, 34, and 44.9 months for Phase I, II, and III and, an expected duration of 96.6 months for successful drugs. They provided information disaggregated by successful and failed drugs, showing that Phase I and Phase II duration are significantly lower for successful drugs and Phase III duration is lower for failed drugs. They also presented data separated by size of the developer, showing that “big pharma” has longer development rates in Phase I, but lower rates in Phases II and III in comparison to “non-big pharma.”
The R&D model developed by Paul et al. (2010) includes assumptions of cycle times for eight different phases of the development process. Original cycle times were provided in years, but are presented here in months to allow for comparison with other estimates. Target-to-hit: 12, Hit-to-lead: 18, Lead optimization: 24, Preclinical: 12, Phase I: 18, Phase II: 30, Phase III: 30, and Submission to launch: 18.
Pregelj, Verreynne, and Hine (2015) conducted a linear regression analysis of clinical trial lengths, including disaggregation by phase. The study looked at 14,319 Phase I, II, and III trials between 2005 and 2009. Ultimately the regression suggested that the estimated marginal means of trial lengths are reduced from about 25 months (all phases clustered between 23-26 months) to around 20 months, with Phase I tests dropping precipitously to about 15 months. [AB1] The study found notable differences between originating firms and development times. Industry-originated products were consistently faster to advance through clinical trials, while U.S. Federal backed products were slower and “other” firms were in the middle.
DiMasi et al. (2016) detailed the cost of producing a new drug, in which capitalized costs play a key role. Timeframes are a major component of calculating capitalized costs, so a key focus of the paper is the length of each phase of clinical testing and regulatory approval time. The paper found that the mean phase lengths are 33.1, 37.9, and 45.1 months for Phase I, II, and III, respectively. The authors also found an average gap of 19.8, 30.3, and 30.7 months after each phase of testing was completed. They found an average time of 31.2 months from synthesis to first human testing, down considerably from 52.0 months in a previous study. The regulatory phase was estimated to take 16.0 months. Overall, the expected time from the beginning of clinical trials to regulatory submission was 80.8 months (96.8 months for time from testing to approval). Note that these figures are not a sum of individual phase times and phase gaps, as there are overlaps across phases and phase gaps.
Martin (2017) investigated phase lengths across clinical trials looking at the decade from 2006-2015 across 3-year periods (overlapping 1 year between each two cohorts) and found increasing phase lengths across the decade, especially for Phase II and Phase III trials. Phase I trials had a median length of 31 months in 2006 before dipping to a low of 27 months for the 2010-2012 period. The final period in the study showed clinical trial lengths of 32 months, a modest increase over the 10-year period. For Phase II trials, trial times have increased gradually from 2006 – 2009, and have increased by 5 months over the second half of the decade to a median of 39 months from 33 months in 2006. For Phase III trials, lengths went from 33 months in 2006 to a high of 42 for the 2011-2013 period, and have since dipped to a median of 40 months, remaining substantially higher than in 2006. The author proposed several potential reasons for these increases, including that companies seem to be changing trial design strategies and have been creating more complex late-stage trials. In addition, trial size has increased in Phase II studies, moving from a median of 88 participants from 2005-2007 to 108 from 2013-2015. Phase III trials, however, saw a reduction in participants over the same period, from 408 to 347. Treatment cycles also increased, and were 23% longer in 2013-2015 than 2010-2012.
Wong, Siah, and Lo (2019) estimated phase lengths in their analysis. They found that during the period from 2005-2015, trial lengths were 1.6, 2.9, and 3.8 years (19.2, 34.8, and 45.6 months) for Phase I, II, and III testing, respectively. The authors also analysed the difference between terminated and successful projects and found that Phase II trials tend to conclude 8.1 months earlier for candidates that do not advance to Phase III testing. Phase III trials concluded 3.2 months later in successful candidates. Differences in Phase I times were insignificant.
Pammolli et al. (2020) also investigated timeframes in their above-mentioned study. The authors calculated median phase duration per each phase of development in 3-years intervals (1990-1999, 2000-2009 and 2010-2013). Phase III had longer duration across the entire period and registration the shortest. For the last interval, phase lengths were estimated at: Preclinical: 19 months, Phase I: 5 months, Phase II: 23 months, Phase III: 34 months and Registration: 3 months. It should be noted that these estimates are much lower than other estimates available in the literature.
New drugs by therapeutic class
Abrantes-Metz et al. (2004) provided estimates for development times by phase, disaggregated by type of compound and therapeutic area. For successful drugs, they show that biologicals have an average duration of 17.87 months in Phase I, 31.87 months in Phase II and 45.63 months in Phase III; chemicals have an average duration of 19.63 months in Phase I, 29.41 months in Phase II and 47.74 months in Phase III, and natural products have an average duration of 21.5 months Phase I, 19.44 months in Phase II and 46.14 months in Phase III. The study also disaggregates phases by 14 therapeutic areas, demonstrating variations between 10.73 months (anti-hypertension) and 22.43 months (transdermal) in Phase I; 21.57 months (anti-HIV/AIDS) and 46.11 months (anti-Alzheimer’s Disease) in Phase II, and 24.31 months (anti-HIV/AIDS) and 63.4 months (anti-Parkinson’s Disease) in Phase III. As noted above, the authors included registration as part of Phase III. There are also estimations provided for failed projects.
In the study by Martin (2017), the author conducted a regression analysis and found that trials of large molecules took more time than those of small molecules, even when accounting for variations in study size and disease complexity. Conversely, Beall et al. (2019) investigated the difference in development times between biologic (large molecules) and chemical drugs (small molecules) and found no significantly difference. Looking at US Patent and Trademark Office (USPTO) and Merck Index data, the study found that median total development times were about 12 years from first filing to FDA approval for each method (12.1 USPTO and 12.4 Merck Index). The development times between small molecules and biologics were not significantly different in either regression (despite being slightly shorter for biologics in the Merck Index).
Other technology types
Davis et al. (2011), in the above-mentioned study investigating the differences between prophylactic vaccines and pharmaceuticals overall, found that clinical development times were not significantly different between vaccines and other drugs. There was, however, a significant difference in pre-clinical development time, with prophylactic vaccines taking 3.7 years and other drugs taking 2.8 years. Aside from this, there were no notable differences in phase lengths.
The above-mentioned Portfolio-To-Impact (P2I) Model (Terry et al. 2018, Young et al. 2018) also contains assumptions for time length for each development phase, summarized in the table below for different types of technologies included in the model (“archetypes”).
Source: Young et al., 2018, p.8.
Using the P2I Model to analyse the vaccine portfolio of the European Vaccine Initiative (EVI), a not-for-profit organization, Gunn et al. (2019) provide information on timeframes for the historical development of vaccines within the organization. The portfolio includes candidates for various diseases of poverty and emerging infectious diseases at different stages of development. The preclinical phase was estimated at 36 months, Phase I at 17.4 and Phase II at 22.5.
3) Main reasons for development termination
The above-mentioned paper by DiMasi (2001) investigated the reasons for termination of projects under clinical development, in two cohorts of 5-year periods (from 1981-1986 and 1987-1992). Results for each period, respectively, are 29.8%/33.8% for economic reasons, 33.0%/37.6% for efficacy, 21.4%/19.6% for safety, and 15.8%/9.0% for other reasons.
Kola and Landis (2004) also investigated the reasons for failure and point out that over the course of the decade, the reasons for failure shifted from primarily pharmacokinetic/bioavailability reasons (~40% in 1991) to commercial (20%), toxicology (20%), and cost of goods (9%) reasons (which notably was not mentioned as a reason for failure in 1991) in 2000. Clinical safety and efficacy as causes of failure stay roughly constant between 1991 and 2000 (10-12% and 30-28%, respectively). Overall, the 1990s are seen as a time of sea change, in which clinical success rates plummeted and never fully recovered. The authors suggest that this could be the result of several factors, including the targeting of more complex diseases in clinical trials, competing with enhanced standards of care, and more demanding regulatory authorities.
The trio of Arrowsmith papers published by Thomson Reuters (Arrowsmith 2011a, 2011b and 2013) also outlined major causes for failure in Phase II and Phase III tests. For Phase III tests, the major reasons for failure from 2007-2010 were efficacy (66%), safety (21%), and commercial (7%). For Phase II tests from 2008-2010, the four major factors in failure were efficacy (51%), strategic (29%), safety (19%), and pharmacokinetics/bioavailability (1%). Therapeutic area also played a role, with cancer and alimentary drugs failing more often than other those in areas. The final study for Phase II and III studies from 2011-2012 found that the primary factors for failure were efficacy (56%), safety (28%), and strategic (7%). The 2013 study showed that, of the 83 failures in Phase III trials, 28% were cancer drugs, 18% were nervous system and 13% were alimentary/metabolism (including obesity and diabetes). One conclusion was that large numbers of failures occurred in drugs with novel mechanisms of action in areas of unmet need.
Harrison (2016) found similar results for the time period between 2013-2015. Looking at Phase II and Phase III failures, the primary factors in failure were efficacy (52%), safety (24%), strategy (15%), commercial (6%), and operational (3%). Strategic reasons were a significantly higher cause of failure in Phase II (21%) than Phase III (14%). The highest percentage of failures were in oncology (32%) and CNS (17%), confirming that certain therapeutic areas are more difficult than others.
Hwang et al. (2016) explored failures of investigational drugs in Phase 3 trials across several countries from January 1st, 1998 – December 31st, 2008, with follow up in 2015. The study included the United States, Canada, Australia, Switzerland, and the EU. Of 640 novel drugs and biologics, 53.8 were unapproved and 46.2% were approved (35.9% in the United States). Approximately 30% were biologics and 70% were pharmacologic. Of drugs that were not approved, 56.7 were not approved due to lack of efficacy, 17.2% due to lack of safety, and 21.5% for commercial reasons. In only 4.7% were the reasons for a lack of approval unknown. Commercial reasons were significantly more likely to be the cause of failure for small and medium sized enterprises (P < .001). A major drawback to the study is that it relied on publicly available information. This may cause bias since the FDA’s response letters denying a drug approval are not required to explain precise reasons for failure.
Schumacher et al. (2016) examined changing R&D models in the pharmaceutical industry and broad organizational archetypes for future development. Laying out the argument for a changing R&D model, the authors included a list of issues impacting pharmaceutical R&D. They identified various reasons for high attrition rates, including a lack of reliability in published data; biopharmaceutical issues including PK; poor predictive models in discovery and preclinical research; target-based drug discovery and advanced complexity of target selection; competition for proprietary target; complex process for target validation; complexity of clinical trials in treating chronic diseases; increasing demands from regulatory authorities and funders; and a lack of knowledge in small organizations resulting in lower PTRS from phase I to submission than for large organizations. Whereas many of the issues identified in the literature are specific to clinical trials, Schumacher includes many issues that affect pre-clinical development and drug discovery. While the broader focus of the paper is innovative R&D strategies, these issues continue to plague health product development in this potential transition period.
Dahlin et al. (2016) examined the types of pharmaceutical product development strategies that are most likely to reach market from a sample of 2,562 clinical trials conducted in 406 US pharmaceutical companies between 1993 and 2004. Product development strategies were categorized into 4 distinct paths: (i) a novel strategy – indicating that the firm was undertaking product development for a drug and indication that they had no prior experience conducting clinical trials for; (ii) a drug experience strategy – combining previous drug development expertise with a new indication; (iii) an indication experience – showing previous clinical trial experience with an indication but for a new drug; and, (iv) a combined experience strategy – denoting prior expertise conducting clinical trials for the drug and indication. The results, amongst other things, show that combined experience strategies were the most successful (and least likely to be utilized), followed by drug experience strategies, with novel strategies trailing closely, and, finally, indication experience strategies being the least successful means to get a product to market.
Lauer et al. (2017) investigated the effect that costs and attrition rates have on the development of cardiovascular medicines. Many of the factors that have been found to be increasing prices and lowering success rates overall in pharmaceuticals are also potential drivers of costs for cardiovascular medicines specifically. The study cites several issues, including complex trial design, restrictive inclusion and exclusion criteria, strict regulations, excessive source-data verification, and the effect of clinical conduct on workflow. There is a lengthy explanation of barriers to cost-efficient clinical trials, which traverses several responsible parties for increased prices. Sponsor-induced risk aversion may lead to extraneous steps and longer protocols. A disconnect between research and care may cause inefficiencies because healthcare professionals are unable to interpret clinical research methods. Regulatory boards are decentralized and less efficient than they could be. Trial regulations are antiquated and are designed for smaller trials. There are a plethora of other causes and actors, but the takeaway is that there are many complex and diverse barriers to improving clinical R&D processes.
Insufficient number of estimates of early discovery and preclinical phase lengths and/or success rates
Insufficient data on R&D of health technologies, other than new molecular entities (NMEs)
Insufficient data on R&D conducted outside of the United States, Europe, and Japan
Insufficient information on product development by organizations other than pharmaceutical and biotech companies
Insufficient number of studies using big data techniques; most of the literature is based on limited data sets
 A prior research synthesis focused on costs. See: Knowledge Portal on Innovation and Access to Medicines, Research Synthesis: Costs of Pharmaceutical R&D, available at: https://www.knowledgeportalia.org/cost-of-r-d
* For the purposes of this review, we have established three categories to describe the state of the literature: thin, considerable, and rich.
- Thin: There are relatively few papers and/or there are not many recent papers and/or there are clear gaps
- Considerable: There are several papers and/or there are a handful of recent papers and/or there are some clear gaps
- Rich: There is a wealth of papers on the topic and/or papers continue to be published that address this issue area and/or there are less obvious gaps
Scope: While many of these issues can touch a variety of sectors, this review focuses on medicines. The term medicines is used to cover the category of health technologies, including drugs, biologics (including vaccines), and diagnostic devices.
Disclaimer: The research syntheses aim to provide a concise, comprehensive overview of the current state of research on a specific topic. They seek to cover the main studies in the academic and grey literature, but are not systematic reviews capturing all published studies on a topic. As with any research synthesis, they also reflect the judgments of the researchers. The length and detail vary by topic. Each synthesis will undergo open peer review, and be updated periodically based on feedback received on important missing studies and/or new research. Selected topics focus on national and international-level policies, while recognizing that other determinants of access operate at sub-national level. Work is ongoing on additional topics. We welcome suggestions on the current syntheses and/or on new topics to cover.
Open Access: This research synthesis is published Open Access, and distributed in accordance with the Creative Commons Attribution Non Commercial International (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. Third party material are not included. See: https://creativecommons.org/licenses/by-nc/4.0/.