Cost-Effectiveness of Aging Research

(Cross­posted from Hour­glass Magaz­ine: https://​​hour­​​blog/​​2018/​​9/​​27/​​cost-effec­tive­ness-of-ag­ing-re­search-why-solve-ag­ing-part-3)

Is ag­ing re­search a cost-effec­tive way of pre­vent­ing death and ill­ness? How does it com­pare to med­i­cal re­search more gen­er­ally, or to med­i­cal treat­ment, or to treat­ment of in­fec­tious dis­eases in poor coun­tries?

This post is go­ing to try to an­swer that ques­tion, in a quan­ti­ta­tive but very ap­prox­i­mate fash­ion.

If you want to play with the num­bers for your­self, you can check out the spread­sheet or Guessti­mate model.

Dr. Owen Cot­ton-Bar­ratt of the Fu­ture of Hu­man­ity In­sti­tute and the Global Pri­ori­ties Pro­ject has writ­ten a se­ries of es­says on the cost-benefit pri­ori­ti­za­tion of “prob­lems of un­cer­tain difficulty”—that is, prob­lems where the amount of re­sources needed to solve them might be any­where be­tween many or­ders of mag­ni­tude, with roughly equal prob­a­bil­ity. Should you spend re­sources on re­search­ing these hard prob­lems to­day?

Well, it de­pends on how tractable the prob­lem is (how far along we are to­wards solv­ing it), how much benefit solv­ing it would offer us, and how ne­glected the prob­lem is (how much has already been in­vested into it.)

Our model for the benefit of im­me­di­ate in­vest­ment said that the benefit was of size kB/​z. The three terms here line up pretty well with the com­po­nents of the three fac­tor model. The scale of the prob­lem is ex­pressed by B, the size of the benefit. The ne­glect­ed­ness gives us the term 1/​z, the re­cip­ro­cal of the amount of in­vest­ment so far. And the re­main­ing term, k, mea­sures the tractabil­ity of the prob­lem.

Cot­ton-Bar­rett has a quan­ti­ta­tive ar­gu­ment for why the tractabil­ity of the prob­lem shouldn’t mat­ter much to our will­ing­ness to in­vest in it:

Are there any les­sons to be drawn from this? One is that tractabil­ity may mat­ter less than the other two fac­tors. Un­der the box model dis­cussed, we have k*= p/​(ln(y/​z)), where y is some level of re­sources such that we be­lieve there is a prob­a­bil­ity p of suc­cess by the time y re­sources are in­vested. We might like to con­sider k= θk*, where θ is a fac­tor to ad­just for de­vi­a­tion from the box model. Then k it­self de­com­poses into three fac­tors: p, the like­li­hood of even­tual suc­cess; 1/​(ln(y/​z)), which tracks time we may have to wait un­til suc­cess; and θ, which ex­presses some­thing of whether we are cur­rently in a range where suc­cess is at all plau­si­ble.
If the prob­lem is some­thing that we be­lieve is likely to be out­right im­pos­si­ble, like con­struct­ing a per­pet­ual mo­tion ma­chine, then p will be very small and this will kill the tractabil­ity. If the prob­lem is nec­es­sar­ily hard and not solv­able soon, like send­ing peo­ple to other stars, then the box model will be badly wrong (or is best ap­plied with our cur­rent po­si­tion not even in the box), so θ can kill the tractabil­ity. But if it’s plau­si­ble that the prob­lem is sol­u­ble, and it might be easy — even if might also be ex­tremely hard — the re­main­ing com­po­nent of tractabil­ity is 1/​(ln(y/​z)). Be­cause of the log­a­r­ithm in this ex­pres­sion, it is hard for it to af­fect the fi­nal an­swer by more than an or­der of mag­ni­tude or so.

As an ap­pli­ca­tion of this model, the Global Pri­ori­ties Pro­ject es­ti­mates that re­search into the ne­glected trop­i­cal dis­eases with the high­est global DALY bur­den (di­ar­rheal dis­eases) could be 6x more cost-effec­tive, in terms of DALYs per dol­lar, than the 80,000 Hours recom­mended top char­i­ties.

They also es­ti­mate, us­ing the same model, that med­i­cal re­search as a whole is be­ing un­der­in­vested in. They es­ti­mate the cost-effec­tive­ness of med­i­cal re­search as a whole at $8000/​QALY—worse than the best in­ter­ven­tions for global poverty (at $50/​QALY) but sig­nifi­cantly more cost-effec­tive than most health in­ter­ven­tions funded by the NHS (go­ing up to about $50,000/​QALY).

Now let’s use that same model to look at ag­ing re­search.

Cost-Benefit Es­ti­mates of Aging Research

The model Cot­ton-Bar­ratt recom­mends for first-pass Fermi es­ti­mates is

Ex­pected benefit = p B/​(z log(y/​z))

Where B is the benefit in the case of suc­cess, z is the cur­rent spend­ing, and p/​log(y/​z) is the tractabil­ity, or the prob­a­bil­ity p of achiev­ing the goal once we’ve spent a given mul­ti­ple y of the cur­rent re­source spend z.

How does this ap­ply in the case of ag­ing?

Benefit Considerations

We imag­ine that a suc­cess­ful ag­ing in­ter­ven­tion will shift the DALY bur­den of age-re­lated dis­ease later, for con­crete­ness let’s say by ten years start­ing at age 50.

So a 60-year-old will have the dis­ease risk of a pre­sent-day 50-year-old, a 70-year-old will have the dis­ease risk of a 60-year-old, and so on. If we de­note by D_50 the ex­pected DALY bur­den of age-re­lated dis­ease on a 50- to 60-year-old and N_50 the num­ber of 50- to 60-year-olds in the world, the benefit of an ag­ing in­ter­ven­tion is:

B = N_60 (D_60-D_50) + N_70 (D_70-D_60) + N_80 (D_80-D_70) + N_90 (D_90- D_80).

The global DALY bur­dens for var­i­ous age-re­lated dis­eases at differ­ent ages, and the world pop­u­la­tions at those ages, are available from pub­lic statis­tics, so we can make an es­ti­mate of the benefit in terms of DALY gain per year from a suc­cess­ful anti-ag­ing in­ter­ven­tion. (This does not even ad­dress the is­sue that anti-ag­ing in­ter­ven­tions will also ex­tend life; so it’s a con­ser­va­tive es­ti­mate.)

Ne­glect­ed­ness Considerations

What is the cur­rent spend­ing level?

We can di­vide re­search on age-re­lated dis­ease into gen­eral ag­ing re­search and dis­ease-spe­cific re­search. If we con­sider only ag­ing re­search, then ag­ing re­search will ap­pear more ne­glected, and thus more cost-effec­tive; if we con­sider all age-re­lated dis­ease re­search (such as can­cer re­search, Alzheimer’s re­search, etc) ag­ing re­search will ap­pear less ne­glected and less cost-effec­tive. We’ll split the differ­ence by treat­ing the amount of ag­ing re­search as

A + theta O

Where A is ag­ing-spe­cific re­search, O is re­search into other age-re­lated dis­ease, and theta is a weight be­tween 0 and 1 to rep­re­sent how much we think dis­ease-spe­cific re­search “counts.”


For the DALY bur­den of the dis­eases of ag­ing we use the Global Bur­den of Disease 2016 statis­tics.

For es­ti­mates of the amount of ag­ing spend­ing we use the Na­tional In­sti­tute of Aging’s 2016 bud­get, the bud­gets of var­i­ous EU re­search or­ga­ni­za­tions[1], the R&D bud­get of Unity Biotech­nol­ogy, and spend­ing on “senes­cence” or “re­gen­er­a­tive medicine” from the Pharma Cog­ni­tive database.

For es­ti­mates of the amount of age-re­lated dis­ease spend­ing we use the Na­tional In­sti­tute of Health’s 2017 bud­get and the IFPMA’s es­ti­mates of 2017 pharma R&D spend­ing.

For ex­perts’ es­ti­mates of the tractabil­ity of ag­ing spend­ing, we use Aubrey De Grey’s pre­dic­tions[2] as an op­ti­mistic es­ti­mate, and the UK Longevity Panel’s pre­dic­tions[3] as a pes­simistic es­ti­mate.

We have high un­cer­tainty around all of these num­bers, but es­pe­cially of the amount of ag­ing spend­ing, since there’s no good way to es­ti­mate, to my knowl­edge, how much is be­ing spent in phar­ma­ceu­ti­cal R&D on ag­ing drugs, and no good data on the amount of ag­ing re­search dol­lars spent by pri­vate or­ga­ni­za­tions, some of which, like Cal­ico, may be quite well-funded. What counts as “ag­ing re­search” is also a some­what sub­jec­tive judg­ment; some ag­ing re­search may not la­bel it­self as such, and some re­search la­beled “ag­ing” may ac­tu­ally be dis­ease-spe­cific re­search not rele­vant to the un­der­ly­ing biol­ogy of ag­ing.


Our es­ti­mates of to­tal cur­rent ag­ing re­search spend­ing are $1.8 billion-4.5 billion, and we es­ti­mate a log-lin­ear dis­tri­bu­tion (the modal amount of spend­ing is likely on the low end, close to the NIH’s ag­ing bud­get, but there may be a long tail al­low­ing for much higher spend­ing, es­pe­cially if pri­vate drug com­pa­nies have more ag­ing-re­lated R&D than pub­lic databases es­ti­mate.)

Our rough es­ti­mate of to­tal age-re­lated dis­ease spend­ing is $104 billion, and we es­ti­mate a nor­mal dis­tri­bu­tion.

Our es­ti­mate of tractabil­ity is uniformly dis­tributed be­tween 0.1 and 1, with mean 0.56; this fol­lows the “un­cer­tain chance of suc­cess” model in Cot­ton-Bar­ratt’s calcu­la­tions.

Our es­ti­mate of to­tal benefit from de­lay­ing ag­ing by 10 years is 176,800,000 DALYs saved yearly wor­ld­wide, plus or minus 30M DALYs, and we as­sume a nor­mal dis­tri­bu­tion.

For theta=0 (only ag­ing-spe­cific re­search counts) the cost-effec­tive­ness is about $42/​DALY.

For theta=1 (all age-re­lated dis­ease R&D counts), the cost-effec­tive­ness is about $1050/​DALY, more effec­tive than GPP’s es­ti­mates of med­i­cal re­search as a whole.

Bot­tom Lines

With a very rough and pre­limi­nary anal­y­sis, it looks like ag­ing re­search could be com­pa­rable or su­pe­rior in cost-effec­tive­ness to the most cost-effec­tive global health in­ter­ven­tions.

GiveWell es­ti­mates a cost of $1965 for a gain of ~8 DALY-equiv­a­lents, or $437.50 per DALY, from giv­ing malaria-pre­vent­ing mosquito nets to chil­dren in de­vel­op­ing coun­tries. Th­ese es­ti­mates have changed quite a bit over time—some older num­bers from the re­search liter­a­ture es­ti­mate $14-110 per DALY from mosquito nets.

This means the es­ti­mated cost-effec­tive­ness of ag­ing re­search (even with a con­ser­va­tive value of theta) is solidly com­pet­i­tive with even the best cost-effec­tive­ness num­bers for de­vel­op­ing-world char­i­ties.

Of course, ag­ing re­search is much more spec­u­la­tive than di­rectly treat­ing or pre­vent­ing dis­ease by known meth­ods. If you want to buy a sure thing, re­search of any kind is not a great choice. Ad­di­tion­ally, these back-of-the-en­velope cost-effec­tive­ness es­ti­mates are much more spec­u­la­tive them­selves than the abun­dant em­piri­cal re­search on trop­i­cal dis­ease pre­ven­tion.

Another con­sid­er­a­tion is that ag­ing is a much big­ger prob­lem, in to­tal, than any spe­cific dis­ease. The to­tal DALY bur­den of the dis­eases of ag­ing is about 700 mil­lion DALYs, whereas the to­tal DALY bur­den of malaria is about 50 mil­lion, or more than 10x smaller.

If you like high-risk, high-re­turn, cost-effec­tive life­sav­ing pro­jects, then med­i­cal re­search in gen­eral may be a good buy, and ag­ing re­search es­pe­cially so, be­cause the level of ex­ist­ing fund­ing is so low, and the size of the im­pact of suc­cess is so high.


[1]Ge­hem, Maarten, and Paula Sánchez Díaz. Shades of gray­ing: re­search tack­ling the grand challenge of ag­ing for Europe. The Hague Cen­tre for Strate­gic Stud­ies, 2013.

[2]De Grey, Aubrey DNJ. “Life span ex­ten­sion re­search and pub­lic de­bate: so­cietal con­sid­er­a­tions.” Stud­ies in Ethics, Law, and Tech­nol­ogy 1.1 (2007).