This essay was jointly written by Peter Hurford and Marcus A. Davis. This is part of a series exploring the cost-effectiveness of vaccines.
Interventions related to vaccines seem to be highly cost-effective. The World Health Organization calls vaccines “one of the most powerful and cost-effective of all health interventions." (WHO, 2009, pXIV) and the Copenhagen Consensus says that "[v]accination may be the most effective public health intervention of all time” (Copenhagen Consensus Center Guide to Giving 2011, p37). GiveWell finds vaccine-related interventions to be highly cost-effective and have recommended one-off donations to vaccines on multiple occasions  through Incubation Grants. However, GiveWell has historically struggled to find room for more funding in this area.
How cost-effective might developing new vaccines be? GAVI, the leading funder for vaccine-related work, is cited by a few different sources as potentially saving a life for under $1000, though this estimate is not robust and this estimate is not related solely to work on developing new vaccines. I’d be curious to see more work on evaluating this for a few reasons:
It seems valuable as a benchmark to see how other interventions compare to vaccine-related work.
Comparing differences of cost-effectiveness within vaccine-related work (e.g., developing new vaccines versus further distributing vaccines that already exist) can aid our understanding of how interventions and implementations of interventions can differ.
Assessing the value of R&D for new vaccines could help us understand the value of funding R&D more generally.
Before looking in depth at the cost-effectiveness of vaccines, my first question was how long it takes to make a vaccine.
What does the literature say?
Turning to the literature, Light, Andrus, and Warburton (2009) outline the short pithy answer -- you need “a variable amount of time [...] plus about 5 years”. Here, the five years are the years necessary to get past clinical trials and registration, and the highly variable amount of time is the time you need to find a vaccine that will get past the clinical trials. After passing clinical trials, one would then need about 5-10 years to scale and distribute the vaccine (Hecht and Jameson, 2011).
It is reported anecdotally to take 12-15 years to discover new medicine (Light, Andrus, & Warburton, 2009), though it’s not clear what this estimate is based on or how this generalize to vaccines. Instead, maybe we could look more at the vaccine lifecycle. For a vaccine to be successful, it has to be successfully developed and then tested through pre-clincal stages, three stages of clinical trials with increasing sample sizes, and then be registered and licensed.
Conditional on a vaccine candidate being successful at every stage, it takes an average of ten years to develop from preclinical stages to launch and an average of 7.6 years to go from Phase I to launch (Struck, 1996). The total time to develop a vaccine would be longer, since vaccine candidates can fail and be restarted, multiple candidates are tried simultaneously, and there is additional unaccounted discovery time before preclinical trials.
Only 22% of developed vaccines are successful, start to finish (Struck, 1996), which would mean that 4.5 vaccine candidates are needed on average to produce a workable vaccine.
GSK themselves says it can take up to $1B and 20-50 years to create and fully distribute a vaccine at scale.
What does the historical “outside view” say?
Besides looking at literature, another potentially good way to learn how long it takes to make a vaccine is to use base rates and look at how long it took historically to make all of the past vaccines.
This method, however, has a number of limitations. There is a surprising amount of uncertainty about the correct start year of each vaccine, since it is difficult to know what the beginning of a research actually is and the transition from "not researched" to "researched" is very gradual. Also, there are often several distinct strands of research that contribute to the final discovery that can be somewhat overlapping (Light, Andrus, & Warburton, 2009).
When looking at the time between when the viral agent was first linked to the disease and the date that a vaccine has been licensed in the US for a few vaccines, it looks to be an average of 52 years (Gilmour, 2013, slide 3). However, this is misleading as many viral agents were identified before vaccine technology existed in earnest, creating very long lag times before vaccines could be created and inflating the estimate in a way that is not representative of current vaccine manufacturing capabilities.
It also unclear at what point the research can be said to have ended with a finished vaccine. The most intuitive approach is to use the date of licensure, but this date can vary wildly between countries (sometimes spanning multiple decades) and countries have inconsistent standards for what is needed to license a vaccine. Additionally, many of the earlier vaccines the modern clinical trial and licensing system did not yet exist and it’s not clear how much additional testing was needed from a prototype vaccine to mass rollout of the vaccine.
Regardless, looking back at history myself for 27 different vaccines, I find the following rough timelines:
Rabies - 4 years, 1881-1885 (Schwartz, 2001; Wikipedia)
Rubella - 7 years, 1962-1969 (Wikipedia)
Pertussis - 8 years, 1906-1914 (CDC, 2017)
Measles - 9 years, 1954-1963 (Rice, 2017a)
Influenza - 14 years, 1931-1945 (WHO, 2009 , p104; Wikipedia)
Japanese encephalitis - 20 years, 1934-1954 (Artenstein (ed.), 2010, p317; Barrett, 2014, p4)
Polio - 20 years, 1935-1955 (CDC, 2017; Wikipedia 1; Wikipedia 2; Wikipedia 3)
Tuberculosis - 21 years, 1900-1921 (Rice, 2017b)
Mumps - 22 years, 1945-1967 (ProCon, 2017)
Hepatitis A - 24 years, 1967-1991 (Melnick, 1995; Wikipedia)
Rotavirus - 26 years, 1980-2006 (Wikipedia 1; Wikipedia 2)
Smallpox - 26 years, 1770-1796 (Boylston, 2012; Wikipedia)
Yellow Fever - 27 years, 1912-1939 (Wikipedia; Frierson, 2010)
Hepatitis B - 38 years, 1943-1981 (CDC, 2017)
Tick-borne encephalitis - 39 years, 1937-1976 (Wikipedia; WHO, 2016, Slide 5; Baselli, et. al., 2011)
Diptheria - 40 years, 1883-1923 (WHO, 2009, p106; Rice, 2017c)
Tetanus - 40 years, 1884-1924 (CDC, 2017)
Hib disease - 44 years, 1933-1977 (Wikipedia 1; Wikipedia 2)
Ebola - ~46? years, 1976-2022(?) (Johnson, Lange, Webb, & Murphy, 1977; Pattyn, et. al., 1977; Lupton, et. al., 1980)
HIV - ~46? years, 1984-2030(?) (Esparza, 2013; Hecht & Jameson, 2011)
Typhoid - 58 years, 1838-1896 (Wikipedia 1; Wikipedia 2)
Malaria - ~58? years, 1967-2025(?) (GAVI, 2016; Wikipedia)
Pneumococcal disease - 66 years, 1911-1977 (CDC, 2017)
Meningitis - 68 years, 1906-1974 (Flexner & Jobling, 1907; CDC, 2015)
Taking these numbers literally, this gives a mean of 31.8 years of development, with a median of 27 years and a standard deviation of 17.7 years. If you exclude the vaccines still under development (HIV, malaria, and ebola), the mean is 29.5 years (median 26, SD 17.4).
It’s unclear what conclusions we should draw from either of these approaches. Firstly, it’s important to keep in mind that these statistics are unlikely to be representative of future vaccines, because the low-hanging fruit is likely already gone, early vaccines did not have to go through the modern clinical trial system, and because modern technology and investment could speed up R&D. On the other hand, modern vaccine development may be sped up by significant advances in technology. Lastly, the vaccines that take the longest are the least likely to have already been developed, simply because they stretch out over more time, which introduces a natural bias toward dealing with "harder" vaccines today above and beyond the "low hanging fruit" factor.
However, if you compare the eight vaccines that started development after 1940 and have completed with the 16 vaccines developed before 1940 and have completed, the difference in completion time is marginally significant at best (t-test p = 0.12).
Lastly, it’s interesting to see the difference between the timeline pointed to by the academic literature (“12-15 years” for medicine generally and around 20 years for vaccines), the timeline pointed to by GSK (“20-50 years”), and the timeline implied by the historical record (“mean of 31.2 years”). Taken together and weighing these three sources of evidence evenly, this suggests an average of 29 years for the typical vaccine, though with high uncertainty based on uncertainties in each approach and on many particular vaccines not being typical.
Next in the series: How much does it cost to research and develop a vaccine?
: For examples, reviewing GAVI in 2009 as a potential top charity, co-funding a drug-related intervention with the Gates Foundation in 2012, funding IDInsight to do an RCT on incentives for vaccines in 2014, funding JPAL to do an RCT on vaccines in 2015, funding New Incentives to work on immunization-related conditional cash transfers in 2016, and funding Charity Science Health to work on expanding demand for immunizations in India in 2016.
: See also notes on GAVI in particular and this comment on funding vaccine R&D. For more detail, see GiveWell’s overview of the vaccine funding landscape.
: Francis (2010) quotes GAVI as averting 5M deaths against $3.7B in funding, for a cost-effectiveness of $740 per life saved. Lob-Levyt (2011) quotes GAVI as averting 5.4M vaccine-related deaths against $4491M in vaccine-related spending or $831.67 per life saved. A GAVI press release quotes GAVI as saving 4M lives against $3.7B or $925 per life saved.
: Except the amount of time to get past clinical trials and registration is more like 6.3 years, on average (Struck, 1996; Keyhani, Diener-West, & Powe, 2006; Waye, Jacobs, & Schryvers, 2013).
: Struck (1996) (Table 4) specifies that vaccines take an average of 2.4 years to go from preclinical trials to Phase I, 2.0 years to go to Phase II from Phase I, 1.8 years to get to Phase III, 1.1 years to get to preregistration, and 1.3 years to get to registration. Each time is conditional on the prior step being successful.
: Struck (1996) (Table 3) specifies there is a 96% chance of a vaccine moving from registration to launch, a 68% chance of moving from Phase III to launch, a 54% chance of moving from Phase II to launch, a 39% chance of moving from Phase I to launch, and a 22% chance of moving from preclinical to launch. Inverting these probabilities using math, we can derive a 56% chance of moving from preclinical trials to Phase I, a 72% chance of moving from Phase I to Phase II, a 79% chance of moving from Phase II to Phase III, a 71% chance of moving from Phase III to registration, and a 96% chance of moving from registration to launch.
: Though it wasn’t until 1959 when a modern scientific vaccine was submitted for licensing after modern scientific scrutiny (WHO 2009, p105).
: This is not intended to be an exhaustive list of all vaccines, but is intended to be exhaustive of all vaccines that would be considered "important", such as the vaccines on the WHO list of essential medicines and notable vaccines under current development.
: Though the varicella vaccine was not actually licensed in the US until 1995.
: There was an earlier vaccine, approved in Russia in 1941, that was incubated in a mouse brain and only worked against a few strains. I am unsure how to count this.
: As a disclaimer, I paid $400 for the creation of this source via Vipul Naik. While the citation is to Issa Rice, given that it is on his domain, the actual author is Sebastian Sanchez.
: While a candidate ebola vaccine is currently in Phase III trials (see also WHO, 2015), it is not yet clear whether or not that candidate will succeed, so it is difficult to forecast how long it will take to develop an ebola vaccine. However, analysis in Struck (1996) (Table 4) would suggest there is only ~5 more years left until the ebola vaccine is registered (i.e., registration predicted for 2022) (see also Light, Andrus, & Warburton, 2009). This is a personal prediction with weak confidence and I can't find any public pronouncements from any official organizations or scientists about an estimated license date for an ebola vaccine.
: This is a good example of how there can be wide uncertainty about when vaccine research started. It's certainly not possible to start work on a vaccine before the cause of the disease is identified, which would be 1887 when the bacteria was first isolated. Also work would have to have started by 1917, when the first detailed scientific paper on the vaccine was published (Greenwood, 1917). However, this gives me a 30 year range within which I am uncertain as to whether research had started or not.
: I chose 1940 as a split because it felt "modern" enough (corresponding somewhat to the definition of a modern era as post-World War II) while still encompassing enough vaccines to have a reasonable amount on both sides. I did not look at how the numbers played out before committing to my split date.
: early mean 22.9 years, early median 23.5 years, early SD 10.7 years, late mean 32.8 years, late median 28.5 years, late SD 19.3 years.
: This is the rounded average of 20 years (academic literature), 35 years (midpoint of GSK's “20-50 years”), and 32 years (retrospective analysis).
This is extremely interesting, only now saw this article (I'm relatively new to the forum). Have you guys thought of publishing this (perhaps in combination with your other essay on the costliness of vaccine development) as a journal article? Beside being useful for science policy estimations, another domain of application for these results could be simulations of scientific inquiry (usually done in terms of agent-based models), where this data could serve as the basis of their empirical calibration. While this method has been increasingly popular in the domain of social epistemology, these models tend to be highly abstract, lacking the input of empirical data that would indicate which parameter choices and which results are relevant for the real world.
Thanks. We have not considered publishing as a journal article. I'm unsure of how that could be done, especially without formal academic credentials, and what the relevant costs and benefits would be. My initial guess is that it would be pretty time consuming without much benefit.
There are going to be a few more posts in this series on the path to creating some cost-effectiveness estimates, so stay tuned! :D
I don't think you necessarily need academic credentials: submissions to most relevant journals are fully blind, so nobody would actually know whether you have the credentials or not (and if the article is accepted, you can simply be independent scholars with no affiliation, that's really unimportant (as it should be)).
As for the costs: I think you wouldn't need too much time for this. Best would be to combine both essays into one article, make an intro into the topic, check again your sources and other relevant literature and send to a journal somewhere in the field of sociology of science/philosophy of science/science policy. Now, I am not a sociologist of science, so I am not familiar with other relevant literature on this topic (e.g. whether there already are similar estimations, which apply more rigorous standards, which suggest that you'd have to do the same - you could do some research and check this out, unless you've already done so). Just checking randomly online, I see there are studies such as this one, which employ a more rigorous methodology, but I'm not sure if there is something similar concerning time estimates.
Concerning your current sources, while Wikipedia is usually not an academic standard, if you have good reasons why it is for this kind of research (or at least in some of the cases), you could just explicitly state so in the text. Alternatively, if Wiki articles have their own (academic) sources, just cite those.
As for the benefits: I think there'd be a lot of benefits!
First, your results would be peer-reviewed, and even if the article is rejected you'd have a feedback from experts in the field, which would help you to revise your results and make them more accurate. In case someone in academia has already done a similar work, which you haven't been aware of, at least you'll learn this and integrate it with your results.
Second, your results could become a more reliable basis for discussions on science policy: a peer-reviewed source for other scholars and policy makers. (I'd also have a personal interest here: as a philosopher of science, I'd be extremely interested in using your results in my research, and they would be more reliable if they passed a peer-review procedure).
Third, your personal gain would be having a publication in an academic journal :)
Thanks, that's helpful, and I'm glad you find the research useful. I'll think about it and talk with Marcus (co-author).
Thanks for the post!