Expected cost per life saved of the TAME trial

Septem­ber 2019 up­date: the TAME trial has been funded. See com­ment by av­turchin. This post is now out­dated.

Edit: Upon sug­ges­tion (see com­ments) I set up a Guessti­mate model. There, you can find wider con­fi­dence in­ter­vals and prob­a­bil­ity dis­tri­bu­tions for my fi­nal es­ti­mates. You may want to com­bine my es­ti­mates with your own pri­ors on gen­eral in­ter­ven­tion’s effec­tive­ness and thereby po­ten­tially cor­rect for the high lev­els of un­cer­tainty in my model. Other­wise do not take this anal­y­sis too se­ri­ously, since it is not very ro­bust and it has am­ple mar­gin of im­prove­ment. I en­courage any­one to make their own ad­just­ments or even make a new anal­y­sis based off on this one if in­ter­ested in im­prov­ing this work. Read the com­ments to get an idea of what fur­ther mod­ifi­ca­tions could be im­ple­mented.

I en­courage ev­ery­one in­ter­ested in try­ing to eval­u­ate in­ter­ven­tions in the cause area of ag­ing to read at least the para­graphs “The Idea” and “Fu­ture di­rec­tions in eval­u­at­ing the cause area of ag­ing”, since they con­tain ideas broadly ap­pli­ca­ble for eval­u­a­tions in this area, re­main­ing valid even if the calcu­la­tion of the cost-effec­tive­ness of this spe­cific in­ter­ven­tion could be flawed.



In this post I will try to calcu­late the ex­pected cost per life saved of the Tar­get­ing Aging With Met­formin (TAME) trial, in an at­tempt to im­prove Turchin’s es­ti­mate. I found some of Turchin’s as­sump­tions were un­nec­es­sary or un­jus­tified. He did not provide an ex­pected value calcu­la­tion and he did not ap­ply the nec­es­sary dis­counts. His figure is true only if some of his many as­sump­tions are, and this led to a re­sult that I thought to be many or­ders of mag­ni­tudes off: $0.24 per life saved. In re­al­ity, if you scroll down to the Calcu­la­tion sec­tion, you can see that, ac­count­ing for the ice­break­ing effect of the TAME trial on the FDA, I came to a re­sult that is not dis­tant to Turchin’s. This hap­pened be­cause Turchin did not ac­count for the ice­break­ing effect, com­pen­sat­ing the omis­sion of the dis­counts. Other­wise, If the ice­break­ing effect is dis­re­garded, the re­sults are differ­ent, yet still show high cost-effec­tive­ness. Again, you can find those re­sults in the Calcu­la­tion sec­tion. In the last sec­tion I out­lined fu­ture di­rec­tions and strate­gies that could be helpful to eval­u­ate po­ten­tially much greater op­por­tu­ni­ties in the field of ag­ing re­search.

Writ­ing this anal­y­sis was challeng­ing and some pa­ram­e­ters re­main some­what sub­jec­tive. It’s pos­si­ble I made some mis­takes, so I will pre­sent the calcu­la­tion in a way that makes it eas­ier to change val­ues and redo it. If you find mis­takes or make other im­por­tant ob­ser­va­tions, please com­ment and I’ll mod­ify the post as soon as pos­si­ble. In gen­eral I tried to pro­ceed in a similar way to how GiveWell’s CEAs [1] are made.

The TAME (Tar­get­ing Aging With Met­formin) trial is mo­ti­vated by two rea­sons:

  1. Ac­cord­ing to AFAR [2], the TAME trial could have an ice­break­ing effect on the FDA. It will:
    [...] pave the way for the Food and Drug Ad­minis­tra­tion (FDA) to con­sider ag­ing a mod­ifi­able con­di­tion and an offi­cial “in­di­ca­tion” for which treat­ments can be de­vel­oped and ap­proved. This could lead to in­creased fund­ing and re­search, and ex­panded and bet­ter drug de­vel­op­ment to tar­get ag­ing fac­tors, not just spe­cific age-re­lated dis­eases and con­di­tions. TAME can open the door to on­go­ing re­search and other drug de­vel­op­ments that will help us all live healthier, longer.

  2. Test­ing met­formin on a healthy pop­u­la­tion could prove its benefi­cial effects (com­pres­sion of late life mor­bidity and life ex­ten­sion) on healthy peo­ple in a definite way. After that, the offi­cial “bless­ing” of met­formin by the FDA would cause physi­ci­ans to start sug­gest­ing it to their pa­tients.

The idea

The core of Turchin’s idea is right: Let’s say Longevity Es­cape Ve­loc­ity [3] will hap­pen at date x. If we ex­tend the lives of peo­ple who would have died be­fore date x, mak­ing them reach date x, we are then ”sav­ing their lives”. This could be done by rais­ing the life ex­pec­tancy of a por­tion of the world us­ing sim­ple in­ter­ven­tions such as met­formin.

The given in this ra­tio­nale is that Longevity Es­cape Ve­loc­ity will hap­pen at some point, and I think this is very safe to as­sume. There is no phys­i­cal law that makes Longevity Es­cape Ve­loc­ity im­pos­si­ble [4] and there are ex­am­ples in na­ture of neg­ligibly senes­cent an­i­mals. Early ba­sic re­search about strate­gies to bring ag­ing un­der med­i­cal con­trol [5] by tar­get­ing its core hal­l­marks is already be­gin­ning to trans­late into com­mer­cial ap­pli­ca­tion [6].

Vari­ables needed in the calculation

A) Ex­pected num­ber of years added by met­formin: 1 year

Many stud­ies sug­gest that met­formin could post­pone age-re­lated patholo­gies [7]; Gw­ern picked some rele­vant ones in this de­tailed ar­ti­cle. The largest study is by Ban­nister et al. It’s par­tic­u­larly rele­vant be­cause of the num­ber of sub­jects and be­cause it in­tro­duced a pop­u­la­tion with­out di­a­betes to be com­pared with a di­a­betic one:

We iden­ti­fied 78 241 sub­jects treated with met­formin, 12 222 treated with sulpho­ny­lurea, and 90 463 matched sub­jects with­out di­a­betes. This re­sulted in a to­tal, cen­sored fol­low-up pe­riod of 503 384 years.

It found an all cause mor­tal­ity re­duc­tion in di­a­bet­ics pa­tients treated with met­formin com­pared with a pop­u­la­tion on non di­a­bet­ics:

With refer­ence to ob­served sur­vival in di­a­betic pa­tients ini­ti­ated with met­formin monother­apy [sur­vival time ra­tio (STR) = 1.0], ad­justed me­dian sur­vival time was 15% lower (STR = 0.85, 95% CI 0.81-0.90) in matched in­di­vi­d­u­als with­out di­a­betes.[...]

We can ex­trap­o­late a Risk Ra­tio from this quote:

Unad­justed event rates [...], sur­pris­ingly, were lower in those treated with met­formin than in their matched con­trols (14.4 vs. 15.2 per 1000 per­son-years, re­spec­tively; p = 0.054).

This data trans­lates to around 1 year of more life for di­a­bet­ics than the non di­a­betic con­trol. The con­ver­sion from Risk Ra­tio to years added is ex­plained by Gw­ern in his ar­ti­cle.

There are some con­sid­er­a­tions to make:

  • The mean age of par­ti­ci­pants who had been given met­formin was 61, and it’s pos­si­ble that start­ing ear­lier would make the effect greater.

  • Since di­a­bet­ics lived more than the con­trol group it’s pos­si­ble that healthy peo­ple would live even more, es­pe­cially be­cause non di­a­betic con­trols had less mor­bidity. In the study, table 1 shows that more di­a­bet­ics had already been treated for di­s­or­ders and had slightly worse pa­ram­e­ters than the con­trol group. This could mean that their life ex­pec­tancy with­out met­formin would have been worse than the one of the con­trol group. On the other hand, at page 7 it is noted that: Due to the as­so­ci­a­tion be­tween type 2 di­a­betes and in­creased car­dio­vas­cu­lar risk, peo­ple with type 2 di­a­betes are more likely to be re­ceiv­ing ex­er­cise and lifestyle in­ter­ven­tions and close mon­i­tor­ing and con­trol of blood pres­sure and choles­terol lev­els. Hyper­ten­sion and hy­per­c­holes­tero­laemia are risk fac­tors for car­dio­vas­cu­lar dis­ease but are gen­er­ally asymp­tomatic. There­fore, these con­di­tions may be less well di­ag­nosed in the con­trol group.
    In the study it is also noted that the con­trol group could have more un­di­ag­nosed mor­bidi­ties. In fact an­other in­ter­est­ing find­ing is this: When com­pared with matched, non-di­a­betic con­trols, di­a­betic pa­tients with high co-mor­bidity who were treated with met­formin had par­tic­u­larly im­proved sur­vival (Char­l­son in­dex ≥3: STR=0.67, 0.59–0.77), and this pat­tern in­creased with in­creas­ing mor­bidity (Figure 3). Im­por­tantly, sur­vival was bet­ter with met­formin even in those peo­ple who had not re­ceived car­diac pro­phy­lac­tic med­i­ca­tions at baseline, but con­sis­tent sur­vival benefits were ob­served with met­formin when used in peo­ple with a prior his­tory of each pro­phy­lac­tic treat­ment sub­group.
    This could be due to less un­di­ag­nosed mor­bidi­ties lead­ing to more treat­ment or (most prob­a­bly) due to the fact that the STR, so the differ­ence in death rates, starts to show when di­a­bet­ics get older, that is the same time when their mor­bidity in­creases.

  • The study is a cor­rel­a­tive re­sult, al­though there are the­o­ries of why met­formin may have benefi­cial effects in healthy peo­ple. Met­formin could re­duce car­dio­vas­cu­lar events and could pro­tect from can­cer [7]; it could in­fluence var­i­ous metabolic path­ways linked to ag­ing and mimic caloric re­stric­tion.

  • Turchin cites a pos­si­ble life ex­tend­ing effect of met­formin of 3 years. This pre­sen­ta­tion (by Stephen B. Kritchevsky, PhD. Given at the OAIC An­nual Meet­ing April 19, 2016) at first glance seems to sup­port his figure, stat­ing a 30% effect size, that would trans­late to around 3 years. But this effect size re­gards stud­ies look­ing for the re­duc­tion of the risk of var­i­ous di­s­or­ders such as car­dio­vas­cu­lar dis­eases, de­men­tia and can­cer and not all cause mor­tal­ity, so we can’t use it in this anal­y­sis. The effect sizes of met­formin re­gard­ing var­i­ous di­s­or­ders have been shown to be higher than the effect size re­gard­ing all cause mor­tal­ity. More­over these stud­ies were done on di­a­bet­ics, not healthy peo­ple.

The first two con­sid­er­a­tion could make the 1 year figure a bit higher and the third one lower. The last one has no effect on it. So I’ll leave it at 1 year. If you are able to quan­tify the con­sid­er­a­tions above in an ob­jec­tive man­ner please com­ment.

B) Statis­ti­cal power of the TAME trial: worst case = 10%, av­er­age case = 47%, best case 84%

The ob­jec­tive of the TAME trial will be com­par­ing the risk of var­i­ous mor­bidi­ties among met­formin’s users and non users. It will not check for all cause mor­tal­ity, so I’ll take Gw­ern’s es­ti­mate of 10% - ex­trap­o­lated as if all cause mor­tal­ity was the end­point of TAME—as a worst case, since check­ing for all cause mor­tal­ity leads to a smaller effect size, that in­fluences the power of the trial nega­tively. On the other hand, the pro­jec­tions in Kritchevsky’s pre­sen­ta­tion seem to be too op­ti­mistic. They as­sume at worst an effect size of 20%, and a 30% effect size as a best case. I’m highly skep­ti­cal of effect sizes greater than 20%, since the stud­ies done till now on met­formin’s effect on var­i­ous mor­bidi­ties were done on di­a­bet­ics, who have already an in­creased risk of the mor­bidi­ties stud­ied. I’ll take the 20% effect size as a best case. Ac­cord­ing to the table in the pre­sen­ta­tion, if the effect size is 20%, the statis­ti­cal power could vary be­tween 58% and 84%. I’ll take 84% as the best case. The av­er­age be­tween 10% and 84% is 47%.

C) Num­ber of peo­ple dy­ing of ag­ing in a year near LEV: LEV in 2030: 65M, LEV in 2065: 88M, LEV in 2100s: 98M

I ex­pect LEV could hap­pen some­where be­tween 2030 and 2100: av­er­age 2065. The later LEV hap­pens, the more peo­ple will be dy­ing of ag­ing at that time, and so more would be saved by anti-ag­ing in­ter­ven­tions. Tak­ing an early date for LEV makes the es­ti­mate con­ser­va­tive. The num­ber of peo­ple dy­ing of ag­ing in 2030 will be a frac­tion of the num­ber of peo­ple born around 1950. At that time the world pop­u­la­tion was 2.52 billions and the crude birth rate—the to­tal num­ber of live births ev­ery 1,000 peo­ple in a year—was 36.876, so ap­prox­i­mately 93 mil­lions peo­ple were born. To­day, ag­ing is re­spon­si­ble for around 70% of all deaths. If this per­centage re­mains the same (in re­al­ity it will in­crease, but I will take a con­ser­va­tive ap­proach), in 2030 ap­prox­i­mately 65 mil­lions peo­ple will die of ag­ing. If LEV hap­pens around 2065, then, at that time, peo­ple born in the 80s will be dy­ing of ag­ing. In the 80s the world pop­u­la­tion was ap­prox­i­mately 4.5 billions and the crude birth rate was 27.863. So the num­ber of peo­ple dy­ing of ag­ing in 2065 will be ap­prox­i­mately (4,500,000,000/​1000)*27.863*0.7 = 88 mil­lions. If LEV hap­pens in 2100, then peo­ple born to­day will be dy­ing of ag­ing at that time. To­day’s world pop­u­la­tion is 7.6 billions and the crude birth rate is 18.5, so the num­ber peo­ple dy­ing of ag­ing in a year around 2100 will be (7,600,000,000/​1000)*18.5*0.7= 98 mil­lions.

D) Ad­di­tional share of the world pop­u­la­tion who would take met­formin con­sis­tently: worst case: 0.13%, av­er­age case: 1.3%, best case: 13%

I ex­pect that such an in­ter­ven­tion will be im­ple­mented only among the up­per-mid­dle in­come and high in­come shares of the world pop­u­la­tion. This amounts to 16% of the world pop­u­la­tion al­though this per­centage is in­creas­ing, so this is a con­ser­va­tive es­ti­mate. It also has to be taken into ac­count that only a per­centage of this pop­u­la­tion will take met­formin con­sis­tently enough to benefit from it. Ac­cord­ing to the CRN, in 2017, 76% of the US pop­u­la­tion took dietary sup­ple­ments in the pre­vi­ous 12 months and 21% of them for healthy ag­ing. It is also re­ported that 45% of those who do not take dietary sup­ple­ment might con­sider tak­ing them if a doc­tor recom­mended it.

About the sur­vey:

The sur­vey was con­ducted be­tween Aug. 24–28 by Ip­sos Public Af­fairs and was funded by CRN. The sur­vey was con­ducted on­line in English and in­cluded a na­tional sam­ple of 2,001 adults aged 18 and older liv­ing in the United States, in­clud­ing 1,528 among those who are con­sid­ered sup­ple­ment users. The sur­vey has been con­ducted an­nu­ally since 2000. The pre­ci­sion of Ip­sos on­line polls are mea­sured us­ing a cred­i­bil­ity in­ter­val. In this case, the poll has a cred­i­bil­ity in­ter­val of plus or minus 2.5 per­centage points for all re­spon­dents, and plus or minus 2.9 per­centage points for sup­ple­ment users (see Ip­sos’ on­line pol­ling method­ol­ogy for more info).

Ac­cord­ing to the study above, a ma­jor­ity of the US pop­u­la­tion uses sup­ple­ments, but this is not true for other de­vel­oped coun­tries, with differ­ences in prevalence of sup­ple­ments us­age up to ten­fold.

Tak­ing all this into con­sid­er­a­tion, the share of peo­ple tak­ing sup­ple­ments for healthy ag­ing in the world should be be­tween 0.16*0.76*0.21/​10 = 0.0025 (0.25%) and 0.16*0.76*0.21 = 0.025 (2.5%). Aver­age 1.3%.

This vari­able is about how many more peo­ple do not cur­rently take met­formin but will as a re­sult of the TAME trial. In or­der to im­prove the sense of scale about how wide­spread met­formin cur­rently is, I want to re­port a few statis­tics: Ac­cord­ing to the Amer­i­can Di­a­betes As­so­ci­a­tion, in 2015, 9.4% of the US pop­u­la­tion had di­a­betes. 96% of them have type 2 di­a­betes and 76% were di­ag­nosed.
In the UK, 57% of peo­ple with type 2 di­a­betes are treated with met­formin (Ban­nister et al., 2014, p.1). If met­formin us­age is similar among di­a­bet­ics in the US, that means that al­most 4% of the US pop­u­la­tion cur­rently uses met­formin. This is in ac­cor­dance with the high num­ber of pre­scrip­tions of this medicine in the US: 83.9 mil­lions in 2015. The num­ber is so high prob­a­bly be­cause of pre­scrip­tion re­newals and pre­scrip­tions for other con­di­tions other than di­a­betes.

I want to com­bine all these num­bers to­gether in three pos­si­ble sce­nar­ios:

Worst case: In this case met­formin’s pre­scrip­tion for healthy peo­ple re­mains not very wide­spread, maybe be­cause of side effects (al­though some of them can be miti­gated by the ex­tended re­lease ver­sion), or maybe be­cause met­formin has to be taken for mul­ti­ple years in or­der for it to have an effect on longevity, or maybe be­cause it will be seen as less “nat­u­ral” than sup­ple­ments like mul­ti­vi­tam­ins. In the worst case sce­nario, the num­ber of healthy peo­ple tak­ing met­formin is well be­low the num­ber of ill peo­ple tak­ing met­formin, and also well be­low the num­ber of peo­ple tak­ing sup­ple­ments for healthy ag­ing. There could be as much as a ten­fold de­crease from the 1.3% of peo­ple tak­ing sup­ple­ments for healthy ag­ing in the world, maybe more. So 0.13% of the world pop­u­la­tion. In the US this would trans­late to around 0.76*0.21/​10 = 0.01, so 1% of the US pop­u­la­tion, that is still a lot. I chose a cor­rec­tion of one or­der of mag­ni­tude be­cause this per­centage could mean that a minor­ity of older peo­ple (cur­rently 14.5% of the US pop­u­la­tion is older than 65, but in other coun­tries like Ja­pan and Italy the per­centage is around 30%) with­out di­a­betes would get a met­formin pre­scrip­tion, maybe in the face of risk fac­tors for car­dio­vas­cu­lar con­di­tions. There would be also a minor­ity of young healthy peo­ple tak­ing met­formin, prob­a­bly a big­ger share than to­day’s self-ex­per­i­menters.

There could be a case even worse than this, in which the TAME trial is mostly ig­nored by physi­ci­ans and maybe a cor­rec­tion of two or­ders of mag­ni­tudes would be re­quired. I be­lieve this is too un­likely, and it is coun­ter­bal­anced by a similar ra­tio­nale for the best case, so I will not in­clude it. The sum of the cases (even cases not in­cluded here), each mul­ti­plied by its prob­a­bil­ity, is prob­a­bly very close to the av­er­age case.

Aver­age case: In this case the nega­tive con­sid­er­a­tions of the worst case are com­pen­sated by the pos­i­tive ones in the best case. Or sim­ply met­formin be­comes one sup­ple­ment used for healthy ag­ing like any other to­day. In this case its us­age in the world would be around 1.3%, as calcu­lated ear­lier.

Best case: In this case met­formin be­comes re­ally wide­spread thanks to all the me­dia at­ten­tion it gets. This is partly be­liev­able, be­cause a surge of me­dia at­ten­tion for ag­ing re­search and in­ter­ven­tions is to be ex­pected in the next decades: This will oc­cur when the ba­sic sci­ence and the clini­cal tri­als be­ing done to­day, not only for met­formin, will trans­late into some­thing available for the pub­lic. At that point, medicine against ag­ing could be one of the most widely dis­cussed top­ics in pub­lic dis­course. Sup­ple­ment us­age in gen­eral could also con­tinue to rise in many coun­tries. In this case met­formin’s us­age could be as widely spread across the world as to­day’s sup­ple­ments for healthy ag­ing in the US. So 2.5% of the world pop­u­la­tion would use met­formin. I do not ex­pect more be­cause this already means that 15% (0.76 * 0.21 = 0.15) of the pop­u­la­tion of up­per-mid­dle in­come and high in­come coun­tries would take met­formin.

There could be an even bet­ter case, in which al­most all the el­der pop­u­la­tion and many young peo­ple take met­formin, re­sult­ing maybe in more than 30% of the pop­u­la­tion of de­vel­oped coun­tries us­ing met­formin. But I think this is too im­prob­a­ble and so it’s not in­cluded in the calcu­la­tion.

E) By what ex­tent other ther­a­pies will re­duce met­formin’s marginal effect be­fore Longevity Es­cape Ve­loc­ity is reached: 50%

Met­formin seems to have mul­ti­ple effects on mul­ti­ple metabolic path­ways that are ag­ing re­lated, al­though it re­mains un­clear if its main effect could be on just one of them, and if the path­way in­fluenced the most is known at all. Met­formin’s effect is up­stream enough in the metabolism-to-dam­age chain that the main­te­nance-type ther­a­pies be­ing de­vel­oped now, which af­fect the down­stream changes pro­duced by ag­ing, could all re­duce met­formin’s marginal effect. And even if met­formin some­how re­duced only one kind of ag­ing dam­age (down­stream), a more pow­er­ful ther­apy could ren­der met­formin less use­ful. By how much it de­pends on a va­ri­ety of fac­tors, among which in­clude:
- How close to­gether in time the first big im­pact ther­a­pies for ag­ing will be available.
- How many big im­pact ther­a­pies are needed in or­der to reach LEV.
In my in­tu­ition, the more abruptly LEV ar­rives the big­ger the marginal effect of met­formin could be. If only a sin­gle event is suffi­cient for LEV to ar­rive, met­formin’s effect is full. Other­wise, an­other ther­apy could ac­count for a gain of many years of healthy life but not bring LEV, de­creas­ing met­formin’s marginal effect by an ex­tent, and more of those kind of ther­a­pies could sub­se­quently come. This vari­able is tough to eval­u­ate and I’m not lean­ing for any of the two pos­si­bil­ities. My cor­rec­tion is 50%. I un­der­stand this es­ti­mate re­ally leaves some­thing to be de­sired, and I do not know if I could do much bet­ter us­ing only cur­rent ev­i­dence.

F) Chance the gov­ern­ment will not com­plete fund­ing the trial: 82%

This dis­count is nec­es­sary if we want to calcu­late only the marginal effect of dona­tions to TAME. In or­der to es­ti­mate this chance I con­tacted Nir Barzilai, one promi­nent pro­poser of the TAME trial and au­thor of Met­formin as a Tool to Tar­get Aging. He con­firmed they ap­plied for gov­ern­ment fund­ing, al­though it is still un­sure if the gov­ern­ment will fund the trial. Con­sid­er­ing that around 18% of the ap­pli­ca­tions to NIH re­sult in an award, the chance the gov­ern­ment will not fund the trial is 82%.

G) Ex­pected num­ber of years LEV will be ad­vanced by, con­sid­er­ing the ice­break­ing effect of TAME: 3.6

The TAME trial was con­ceived in 2014 and till till now AFAR was able to raise half of the $75 mil­lions needed (see H). I could pro­ject that there is a 50% chance that in the next four years the fund­ing will be com­pleted any­way. Split­ting chances evenly among the next four years, each year has a 12.5% chance of be­ing the one in which the trial is funded. Ex­tend­ing it four years fur­ther, con­sid­er­ing that eight years from now some­thing else could cause the ice­break­ing effect any­way, there’s a 12.5% chance per year for the next eight years that the ice­break­ing will hap­pen spon­ta­neously. So the ex­pected num­ber of years LEV will be ad­vanced is, let’s say, no more than: 0.125*8 + 0.125*7 + 0.125*6 + 0.125*5 + 0.125*4 + 0.125*3 + 0.125*2 + 0.125*1 = 4.5.

Will ad­vanc­ing the date of the ice­break­ing effect by 4.5 years re­sult in an ad­vance of LEV of 4.5 years? Prob­a­bly not, since the FDA not recog­nis­ing ag­ing as an in­di­ca­tion does not stop ag­ing re­search, and it‘s not clear if the ice­break­ing effect is a bot­tle­neck for achiev­ing LEV. Such an effect could, though, in­crease the bud­get of ag­ing re­search and make re­search in this field more fo­cused on trans­la­tion and on the hal­l­marks of ag­ing in­stead of sin­gle dis­eases, their symp­toms, or transver­sal top­ics (to get a sense of how the bud­get is al­lo­cated right now you can see it di­rectly on NIH’s web­site). If the ice­break­ing effect en­ables the fund­ing of one or more pro­jects which are bot­tle­necks to LEV, then LEV’s date is ad­vanced. The main prob­a­bil­ity at play here is if NIH would in­deed in­crease its bud­get on ag­ing or spend it bet­ter. If this does not hap­pen I do not an­ti­ci­pate other ac­tors would step in who wouldn’t oth­er­wise. Since the ex­pert opinion on this is that such a thing will hap­pen (it is the ob­jec­tive of TAME stated by AFAR and Barzilai), I will put this prob­a­bil­ity at 80%. 4.5*0.8 = 3.6.

Com­ing up with a good value for this vari­able is difficult, and the figure is not very ro­bust. It should be noted, though, that in the calcu­la­tion the num­ber of peo­ple dy­ing per year will have the most in­fluence and the size of this vari­able does not in­fluence the bot­tom line much.

H) Dol­lars nec­es­sary to com­plete the fund­ing of the TAME trial: $37.5M

In or­der to eval­u­ate this vari­able I con­tacted Nir Barzilai and Odette van der Willik, the di­rec­tor of the grant pro­grams at AFAR. They said that till now AFAR was able to raise half of the $75 mil­lions nec­es­sary to fund the trial. So the other half re­mains to be funded. $75M/​2 = $37.5M


Ac­count­ing for ev­ery­thing: H/​(A*B*C*D*E*F + G*C)

Worst case: 37500000/​(1*0.10*65000000*0.0013*0.5*0.82 + 3.6*65000000) = 0.16$
Aver­age case: 37500000/​(1*0.47*88000000*0.013*0.5*0.82 + 3.6*88000000) = 0.12$
Best case: 37500000/​(1*0.84*98000000*0.025*0.5*0.82 + 3.6*98000000) = 0.11$

Only ac­count­ing for the ice­break­ing effect: H/​(G*C))

Worst case: 37500000/​(3.6*65000000) = 0.16$
Aver­age case: 37500000/​(3.6*88000000) = 0.12$
Best case: 37500000/​(3.6*98000000) = 0.11$

Without ac­count­ing for the ice­break­ing effect: H/​(A*B*C*D*E*F)

Worst case: 37500000/​(1*0.10*65000000*0.0013*0.5*0.82) = 10824$
Aver­age case: 37500000/​(1*0.47*88000000*0.013*0.5*0.82) = 170$
Best case: 37500000/​(1*0.84*98000000*0.025*0.5*0.82) = 44$

Edit: Upon sug­ges­tion (see com­ments) I set up a Guessti­mate model (here). You can find wider con­fi­dence in­ter­vals (and prob­a­bil­ity dis­tri­bu­tions) for my fi­nal es­ti­mates there. You may want to com­bine my es­ti­mates with your own pri­ors on gen­eral in­ter­ven­tion’s effec­tive­ness and thereby po­ten­tially cor­rect for the high lev­els of un­cer­tainty in my model.

This re­sult shows that most of the cost-effec­tive­ness of TAME comes from its ice­break­ing effect. The vari­able G is not very ro­bust al­though it should be noted that the ma­jor­ity of in­fluence on the cost-effec­tive­ness come from the scope of the prob­lem (the vari­able C) so with even lower val­ues of G, the figure would be small. Over­all this fund­ing op­por­tu­nity is po­ten­tially very cost effec­tive, al­though there could be fur­ther cor­rec­tions I missed. Five cases out of six are at least one or­der of mag­ni­tude bet­ter than GiveWell’s recom­men­da­tions, al­though this eval­u­a­tion is prob­a­bly less ro­bust than GiveWell’s ones. A coun­ter­bal­ance to its lack of ro­bust­ness (al­though prob­a­bly very small) is that this anal­y­sis does not ac­count for DALYs averted and the eco­nomic gains of a com­pres­sion of the old age mor­bidity of a na­tion (the “longevity div­i­dend”), that could en­hance the cost-effec­tive­ness a bit more.

It should be noted that the worst case that does not ac­count for the ice­break­ing effect pos­si­bly acts like a cap. I do not ex­pect the real cost effec­tive­ness to be worse than that, or, if worse, not much worse.

Ac­count­ing for the ice­break­ing effect, in the de­nom­i­na­tor, I mul­ti­plied the ex­pected years LEV should be ad­vanced by TAME with the num­ber of peo­ple dy­ing of ag­ing per year. The re­sult of the mul­ti­pli­ca­tion re­mains ex­actly the same even in the face of a slow spread­ing of re­ju­ve­nat­ing ther­a­pies (and so LEV) across the world in many years. If 3.6 years seems too much it has to be con­sid­ered that even ap­proach­ing 0, it will gen­er­ate a re­sult always smaller than the ones not ac­count­ing for the ice­break­ing effect.

The or­der of mag­ni­tude of the re­sults that do not ac­count for the ice­break­ing effect is highly in­fluenced by the per­centage of peo­ple who will take met­formin con­sis­tently (vari­able D), mak­ing it prob­a­bly the most rele­vant fac­tor in the calcu­la­tion to­gether with the num­ber of peo­ple dy­ing of ag­ing in a year near LEV (vari­able B), al­though this one varies in a smaller in­ter­val.

QALYs saved and the longevity dividend

In this post I did not ac­count for the QALYs saved by met­formin with­out con­sid­er­ing LEV. I ex­pect this figure to be prob­a­bly higher than 1 for an in­di­vi­d­ual tak­ing met­formin, since the RRs for many di­s­or­ders is around 0.80 [7], sug­gest­ing met­formin com­presses the pe­riod of mor­bidity at the end of life. Ac­cord­ing to Barzilai this would save around 4 trillions of dol­lars. But this would be a com­pletely differ­ent calcu­la­tion and I do not ex­pect the re­sult to be com­pet­i­tive with other in­ter­ven­tions in Effec­tive Altru­ism if not summed with the ones in this anal­y­sis.

Fu­ture di­rec­tions in eval­u­at­ing the cause area of aging

Fi­nanc­ing the TAME trial may prove to be an effec­tive in­ter­ven­tion, but the effect of met­formin or other ex­ist­ing sim­ple in­ter­ven­tions is neg­ligible if com­pared to the po­ten­tial effect of trans­la­tional ba­sic re­search be­ing con­ducted now that di­rectly tar­gets the hal­l­marks of ag­ing. Or­gani­sa­tions work­ing in this area are more difficult to eval­u­ate, due to the un­cer­tainty in­her­ent to ba­sic re­search. Nonethe­less I think the work that some of them are do­ing is be­ing ter­ribly ne­glected de­spite their re­search be­ing a bot­tle­neck to the goal of cur­ing age re­lated dis­eases. Eval­u­at­ing those op­por­tu­ni­ties could have an im­pact too high to ig­nore. My ap­proach will prob­a­bly be to clas­sify the hal­l­marks of ag­ing by their cur­rent ne­glect­ed­ness, tractabil­ity and scope, try­ing to find all the com­pa­nies or non profit or­gani­sa­tions mak­ing re­search in each one of them and clas­sify­ing the differ­ent or­gani­sa­tions by how they ap­proach the prob­lem. This will be a much more difficult task than the pre­sent eval­u­a­tion. Other in­ter­est­ing ques­tions I will try to an­swer re­late to the im­pact of ad­vo­cacy and in­duc­ing policy changes in this area, po­ten­tially hav­ing a mul­ti­plica­tive effect on dol­lars donated.


[1]: I’m still in the pro­cess of learn­ing how to do good cost-effec­tive­ness analy­ses. I worked con­sid­er­ing the ex­am­ple of GiveWell’s CEAs in the sense that I tried to find all the dis­counts ap­pli­ca­ble, and con­sider coun­ter­fac­tu­als. Hav­ing said this I hope I’m not mis­rep­re­sent­ing what GiveWell does.

[2]: AFAR is the Amer­i­can Fed­er­a­tion for Aging Re­search and its mis­sion com­prises “Iden­ti­fy­ing and fund­ing a broad range of cut­ting-edge re­search most likely to in­crease knowl­edge about healthy ag­ing”. TAME has been un­der re­views through sev­eral fund­ing mechanisms and has re­ceived plan­ning fund­ing from AFAR (Barzilai et al., 2016, p. 1). Till now this or­gani­sa­tion raised half the fund­ing needed for the trial (source: email ex­change with Nir Barzilai and Odette van der Willik, the di­rec­tor of the grant pro­grams at AFAR).

[3] Longevity Es­cape Ve­loc­ity, or LEV, is reached when life ex­pec­tancy raises at least one year per year. This con­cept was first in­tro­duced by David Go­bel, co-founder of the Methuse­lah Foun­da­tion, and spear­headed by Aubrey de Grey in his book End­ing Aging. After the first treat­ments tar­get­ing ag­ing will be available to the pub­lic, there will be a point in which so­cieties be­come effec­tively non-ag­ing through reach­ing LEV be­fore the com­plete elimi­na­tion of ag­ing hap­pens, maybe much be­fore it. This be­cause of two rea­sons:

  • Mul­ti­ple in­ter­ven­tions are needed in or­der to tar­get all the as­pects of ag­ing, and at first they will be im­perfect but they will in­crease life ex­pec­tancy more and more as they im­prove.

  • Elimi­nat­ing the ag­ing dam­ages ac­cu­mu­lat­ing in a hu­man life­time is a differ­ent challenge than elimi­nat­ing the dam­ages ac­cu­mu­lat­ing in a longer than hu­man lifes­pan. There­fore be­ing able to elimi­nate dam­ages re­spon­si­ble for cur­rent hu­man death by ag­ing does not guaran­tee achiev­ing neg­ligible senes­cence, but ad­di­tional ther­a­pies may be needed. Given the fast rate at which tech­nol­ogy im­proves af­ter the first proof of con­cepts work, it’s very prob­a­ble that sub­se­quent ad­vances will go faster than the rates of death by ag­ing. Espe­cially be­cause at the point when a sec­ond wave of ther­a­pies will be needed, much more fund­ing will be available (the idea of elimi­nat­ing ag­ing will be some­thing re­ally main­stream) and re­searchers will have much bet­ter tools than to­day’s, while the challenge will be prob­a­bly similar in difficulty and gen­eral ap­proach (re­mov­ing dam­ag­ing molecules, re­pairing struc­tures in­side the cell, re­plac­ing or re­pairing tis­sue etc.).

[4]: As Richard Feyn­man puts it in his 1964 lec­ture at the Gal­ileo Sym­po­sium in Italy ti­tled “What Is and What Should Be the Role of Scien­tific Cul­ture in Modern So­ciety”:

It is one of the most re­mark­able things that in all of the biolog­i­cal sci­ences there is no clue as to the ne­ces­sity of death. If you say we want to make per­pet­ual mo­tion, we have dis­cov­ered enough laws as we stud­ied physics to see that it is ei­ther ab­solutely im­pos­si­ble or else the laws are wrong. But there is noth­ing in biol­ogy yet found that in­di­cates the in­evita­bil­ity of death. This sug­gests to me that it is not at all in­evitable, and that it is only a mat­ter of time be­fore the biol­o­gists dis­cover what it is that is caus­ing us the trou­ble and that that ter­rible uni­ver­sal dis­ease or tem­po­rari­ness of the hu­man’s body will be cured.

[5]: Se­ri­ous efforts about a com­pre­hen­sive strat­egy to bring ag­ing un­der med­i­cal con­trol started in the early 2000s, thanks to Aubrey de Grey’s early pro­pos­als (2002) to the biogeron­tol­ogy com­mu­nity and his land­mark book End­ing Aging (2007), pro­vid­ing a defi­ni­tion of the prob­lem that ren­ders it at­tack­able. At first it en­coun­tered some re­sis­tance, that, though, was not adopted as the main po­si­tion:

We need to re­mem­ber that all hy­pothe­ses go through a stage where one or a small num­ber of in­ves­ti­ga­tors be­lieve some­thing and oth­ers raise doubts. The con­ven­tional wis­dom is usu­ally cor­rect. But while most rad­i­cal ideas are in fact wrong, it is a hal­l­mark of the sci­en­tific pro­cess that it is fair about con­sid­er­ing new propo­si­tions; ev­ery now and then, rad­i­cal ideas turn out to be true. In­deed, these ex­cep­tions are of­ten the most mo­men­tous dis­cov­er­ies in sci­ence. (MIT Tech­nol­ogy Re­view, 2006).

With the hind­sight of to­day it is safe to say that de Grey’s ideas proved to be the kind that causes a paradigm shift. “The Hal­l­marks of Aging” (2013) is a cor­ner­stone in the field that guides to­day’s re­search and it con­tains the same ideas first spear­headed by de Grey, with ag­ing defined as a similar cat­e­gori­sa­tion of dam­ages, with a com­pre­hen­sive, di­vide et im­pera, view of the prob­lem.

Some par­tic­u­larly de­vel­oped or in­ter­est­ing branches to­day are stem cell re­search, in vivo re­pro­gram­ming us­ing Ya­manaka Fac­tors, senes­cent cells re­moval, pre­vent­ing epi­ge­netic er­rors. All this is pos­si­ble also thanks to the en­large­ment and up­grade of the toolbox re­searchers have at their dis­posal (very fa­mous ex­am­ple be­ing CRISPR/​Cas9).

It is in­creas­ingly clear to the sci­en­tific com­mu­nity that tar­get­ing ag­ing is the­o­ret­i­cally su­pe­rior to treat­ing in­di­vi­d­ual chronic dis­eases (Matt Kae­ber­lein et al., 2015) and that “treat­ing one dis­ease at a time, the most you can ex­pect is to ex­change one dis­ease for an­other” as Nir Barzilai points out in his TED Talk.

[6]: The pri­vate sec­tor in the area be­gan to flour­ish in the last three years with promi­nent ex­am­ples such as Oisin Biotech­nolo­gies, Ichor Ther­a­peu­tics, Co­va­lent Bio­sciences (All three spun from SENS re­search Foun­da­tion), Unity Biotech­nol­ogy, AgeX, In­sili­coMedicine. Many of them have tens of mil­lions in in­vest­ments. An overview of the pri­vate sec­tor in this area can be found in Jim Mel­lon’s book “Ju­ve­nes­cence: In­vest­ing in the Age of Longevity” (2017).

[7]: Here I or­ganise some rele­vant pa­pers cited in Gw­ern’s anal­y­sis, in the pa­per by Ban­nister et al. and in Kritchevsky’s pre­sen­ta­tion, plus oth­ers found in­dipen­dently. Note that some of them com­pare met­formin with other drugs that could be harm­ful. Dozens are left out.

All cause mor­tal­ity and differ­ent dis­eases of ag­ing
1. C.A. Ban­nister et al. 2014 Can peo­ple with type 2 di­a­betes live longer than those with­out? A com­par­i­son of mor­tal­ity in peo­ple ini­ti­ated with met­formin or sulpho­ny­lurea monother­apy and matched, non-di­a­betic con­trols.
2. Camp­bell et al. 2017, Met­formin re­duces all-cause mor­tal­ity and dis­eases of age­ing in­de­pen­dent of its effect on di­a­betes con­trol: A sys­tem­atic re­view and meta-anal­y­sis.
3. Stevens et al 2012, Cancer out­comes and all-cause mor­tal­ity in adults al­lo­cated to met­formin: sys­tem­atic re­view and col­lab­o­ra­tive meta-anal­y­sis of ran­domised clini­cal tri­als.
4. Saenz et al. 2005, Met­formin monother­apy for type 2 di­a­betes mel­li­tus.
5. Mor­gan CLl et al. 2012, What next af­ter met­formin? A ret­ro­spec­tive eval­u­a­tion of the out­come of sec­ond-line, glu­cose-low­er­ing ther­a­pies in peo­ple with type 2 di­a­betes.
6. Pan­talone KM, Kat­tan MW, Yu C et al. 2012, In­crease in over­all mor­tal­ity risk in pa­tients with type 2 di­a­betes re­ceiv­ing glipiz­ide, gly­buride or glimepiride monother­apy ver­sus met­formin: a ret­ro­spec­tive anal­y­sis.
7. Mor­gan CLl et al. 2014 As­so­ci­a­tion be­tween first-line monother­apy with sulpho­ny­lurea ver­sus met­formin and risk of all-cause mor­tal­ity and car­dio­vas­cu­lar events: a ret­ro­spec­tive, ob­ser­va­tional study.

Car­dio­vas­cu­lar dis­ease and mor­tal­ity
1. Selvin et al. 2008, Car­dio­vas­cu­lar out­comes in tri­als of oral di­a­betes med­i­ca­tions: a sys­tem­atic re­view.
2. La­manna et al. 2011, Effect of met­formin on car­dio­vas­cu­lar events and mor­tal­ity: a meta-anal­y­sis of ran­dom­ized clini­cal tri­als.
3. Kooy A, de Jager J, Le­hert P et al. 2009, Long-term effects of met­formin on metabolism and microvas­cu­lar and macrovas­cu­lar dis­ease in pa­tients with type 2 di­a­betes mel­li­tus.
4. Rous­sel R. et al. 2010, Met­formin use and mor­tal­ity among pa­tients with di­a­betes and atherothrom­bo­sis.
5. Aguilar D. et al. 2011 Met­formin use and mor­tal­ity in am­bu­la­tory pa­tients with di­a­betes and heart failure.
4. Roumie CL, Hung AM, Greevy RA et al. 2012, Com­par­a­tive effec­tive­ness of sulfony­lurea and met­formin monother­apy on car­dio­vas­cu­lar events in type 2 di­a­betes mel­li­tus: a co­hort study.
5. Mor­gan CLl et al. 2014 As­so­ci­a­tion be­tween first-line monother­apy with sulpho­ny­lurea ver­sus met­formin and risk of all-cause mor­tal­ity and car­dio­vas­cu­lar events: a ret­ro­spec­tive, ob­ser­va­tional study.
6. John­son JA et al. 2005 Re­duced car­dio­vas­cu­lar mor­bidity and mor­tal­ity as­so­ci­ated with met­formin use in sub­jects with type 2 di­a­betes.

1. Cur­rie CJ, Poole CD, Gale EAM. 2009, The in­fluence of glu­cose-low­er­ing ther­a­pies on can­cer risk in type 2 di­a­betes.
2. Stevens et al 2012, Cancer out­comes and all-cause mor­tal­ity in adults al­lo­cated to met­formin: sys­tem­atic re­view and col­lab­o­ra­tive meta-anal­y­sis of ran­domised clini­cal tri­als.
3. Fran­ciosi M et al. 2013, Met­formin ther­apy and risk of can­cer in pa­tients with type 2 di­a­betes: sys­tem­atic re­view.
4. Jacek Kasznicki, Ag­nieszka Sliwin­ska, and Józef Drze­woski, 2014 Met­formin in Cancer Preven­tion and Ther­apy.
5. Hyun Hee Chung et al. 2013, The Re­la­tion­ship be­tween Met­formin and Cancer in Pa­tients with Type 2 Di­a­betes.

Mechanisms of ac­tion
1. Nir Barzilai et al. 2016, Met­formin as a Tool to Tar­get Aging.
2. Kir­pich­nikov D, McFar­lane SI, Sow­ers JR. 2002, Met­formin: an up­date.