health
Medical intervention extends typical healthy human lifespan by multiple decades by 2126.
Last revised Sep 12, 2026.
Current estimate
probability this has happened by each date
Twenty additional healthy years at the frontier, against the 2026 baseline, by 31 December 2035. Nine years. Japan's HALE moved 2.29 years in the twenty-one years to 2021 (hf-evidence-0117), so the required rate is roughly twenty times the best sustained frontier rate on record. No drug trial with an ageing or composite healthspan primary endpoint is enrolled anywhere (hf-evidence-0113), and the one reprogramming trial in humans completes in 2032 (hf-evidence-0114). This is an error bar around zero rather than a forecast.
The same test by 31 December 2050. Twenty-four years. The binding constraint here is arithmetic rather than biological and is the master agent's own reasoning, not a citation: healthy life expectancy is a period measure computed across the whole age distribution, so even an intervention approved in the 2030s and deployed universally would move the series only as fast as it reached and benefited every age group. The largest population life-expectancy gain ever documented from one intervention class is 11.3 years over eight years (hf-evidence-0124), and that was an epidemic-depressed population recovering, with near-total deployment urgency behind it.
The same test by 31 December 2100. Seventy-four years, and this horizon carries most of the probability mass. It is long enough for the reprogramming class now entering humans, or a successor, to be validated, approved, deployed and to show in a period measure - and long enough for more than one mechanism class to be tried, which is the main reason the figure is not lower. It is also the horizon Olshansky's deceleration analysis addresses directly and rejects (hf-evidence-0126).
Healthy life expectancy in at least one of the longest-lived populations exceeds its 2026 value by at least twenty years, shown by replicated population evidence and attributable to medical intervention.
Resolution criteria
- 2035
- Twenty additional healthy years at the frontier, against the 2026 baseline, by 31 December 2035. Nine years. Japan's HALE moved 2.29 years in the twenty-one years to 2021 (hf-evidence-0117), so the required rate is roughly twenty times the best sustained frontier rate on record. No drug trial with an ageing or composite healthspan primary endpoint is enrolled anywhere (hf-evidence-0113), and the one reprogramming trial in humans completes in 2032 (hf-evidence-0114). This is an error bar around zero rather than a forecast.
- 2050
- The same test by 31 December 2050. Twenty-four years. The binding constraint here is arithmetic rather than biological and is the master agent's own reasoning, not a citation: healthy life expectancy is a period measure computed across the whole age distribution, so even an intervention approved in the 2030s and deployed universally would move the series only as fast as it reached and benefited every age group. The largest population life-expectancy gain ever documented from one intervention class is 11.3 years over eight years (hf-evidence-0124), and that was an epidemic-depressed population recovering, with near-total deployment urgency behind it.
- 2100
- The same test by 31 December 2100. Seventy-four years, and this horizon carries most of the probability mass. It is long enough for the reprogramming class now entering humans, or a successor, to be validated, approved, deployed and to show in a period measure - and long enough for more than one mechanism class to be tried, which is the main reason the figure is not lower. It is also the horizon Olshansky's deceleration analysis addresses directly and rejects (hf-evidence-0126).
- 2126
- Healthy life expectancy in at least one of the longest-lived populations exceeds its 2026 value by at least twenty years, shown by replicated population evidence and attributable to medical intervention.
Probability history
| Revised | State | p(2035) | p(2050) | p(2100) |
|---|---|---|---|---|
hf-forecast-0202 | researched | 1% no prior revision | 5% no prior revision | 16% no prior revision |
hf-forecast-0017 | researched | — | — | — |
hf-forecast-0008 | initial prior | — | — | — |
Strongest evidence for
- Partial epigenetic reprogramming entered human trials for the first time in March 2026, which cycle 3 could not record at all; NCT07290244 is recruiting (hf-evidence-0114).
- An anti-IL-11 antibody given only from 75 weeks of age raised mouse median lifespan 22.5 to 25 percent, so a large effect is achievable by intervening after most of life has elapsed - the situation any human geroprotector faces (hf-evidence-0122).
- Two randomised human trials slowed the DunedinPACE pace-of-ageing biomarker, one of them by caloric restriction in healthy adults (hf-evidence-0125).
- The EU-27 aggregate healthy-life-years series recovered above its pre-pandemic level by 2024, 65.1 to 65.5 years for women, so the pandemic dip was not permanent at the aggregate level (hf-evidence-0118).
- The horizon is a century and the forecast needs only one success within it; the strongest optimistic demographic reading still has the frontier rising 1.45 years per decade with no visible ceiling (hf-evidence-0126).
Strongest evidence against
- Every geroscience trial that reached enrolment and reported a primary endpoint has missed it: PEARL partial eta squared 0.001 p = 0.942, RTB101 numerically worse at 1,024 randomised, dasatinib plus quercetin p = 0.611, foselutoclax beaten by its active control (hf-evidence-0113).
- The mouse assay that supplies the field's headline effect sizes may be measuring cancer suppression: 72 percent of UM-HET3 mice die of neoplasia against roughly 17 to 20 percent of humans, and several life-extending agents are directly oncostatic (hf-evidence-0119).
- The same drugs shrink monotonically as species complexity rises - rapamycin 57 percent in yeast to 13 percent in mice - and humans are already about a threefold allometric outlier, so headroom arguments from a two-year mouse do not transfer (hf-evidence-0120).
- Where lifespan and healthspan are measured in the same animals, the frailty-adjusted gain is smaller than the raw lifespan gain in both sexes (hf-evidence-0121).
- The HALE-to-LE gap widened in all five frontier countries between 2000 and 2021, by 0.65 to 2.04 years, so added years are still accruing disproportionately to unhealthy ones (hf-evidence-0117).
- No regulator recognises ageing as an indication: zero matches across the entire 2,787-document FDA guidance corpus, and the nearest artefact explicitly disclaims being a validated surrogate endpoint (hf-evidence-0115).
- TAME has still not registered or enrolled anyone eleven years after its FDA endpoint agreement (hf-evidence-0116).
- The latest replication-grade mouse cohort returned three compounds, all male-only and all single-digit, with one trending toward shortening female lifespan (hf-evidence-0123).
- The largest population life-expectancy gain ever attributed to one intervention class is 11.3 years, about half what this criterion requires, and it was recovery toward a prior level rather than frontier extension (hf-evidence-0124).
Present uncertainties
- Two uncertainties now dominate and they are of different kinds. The first is unchanged from cycle 3: whether human translation, regulatory recognition and population-scale attribution can all occur inside a century. The second is new and is about the seed's own instruments rather than about the world - whether healthy life expectancy at the frontier can be measured well enough, by 2126, for this criterion to resolve at all.
Next discriminating observation
The first trial in which an ageing-directed intervention reports a POSITIVE prespecified primary endpoint in humans - not a biomarker, not a secondary, not a sex-stratified subgroup. This cycle found that enrolment was the wrong event to watch for, because five trials enrolled and reported and all five missed. Secondarily, and as a precondition for the criterion rather than for the claim: whether any statistical agency publishes a healthy-life-expectancy series in which Switzerland does not report more than three times Bulgaria's burden of unhealthy years.
Sources
- Every geroscience trial that reached enrolment and reported a primary endpoint has missed it
Four registered trials of ageing-directed interventions have reported primary outcomes and all four are negative. PEARL (NCT04488601, AgelessRx, Phase 2, 129 enrolled and 114 completed, 48 weeks of weekly compounded rapamycin at 5 or 10 mg) missed its primary endpoint of change in visceral fat by DXA, partial eta squared 0.001, p = 0.942; the widely reported 6.2 percent lean-mass gain is a sex-stratified secondary in 10 mg females at 48 weeks, p = 0.018, with no significant change in males. RTB101 (dactolisib, NCT04668352, resTORbio, Phase 3, 1,024 randomised) found clinically symptomatic respiratory illness in 125 of 510 on placebo versus 134 of 511 on drug, so the drug arm was numerically worse. Dasatinib plus quercetin (NCT04313634, Mayo, Phase 2, 74 enrolled) found median percent change in serum CTX at 20 weeks of minus 4.1 for D+Q, plus 13.7 for fisetin and minus 7.7 for untreated control, p = 0.611. UBX1325 foselutoclax (NCT06011798, 52 enrolled) gained 3.7 ETDRS letters against 5.1 for the aflibercept active control.
Limit: These are four trials of three mechanism classes, not a census of the field, and three of the four tested a disease or body-composition endpoint rather than ageing itself - a missed visceral-fat endpoint is not the same as a failure to slow ageing. The correct reading is that no human efficacy signal has yet been demonstrated for any of these classes, not that any has been refuted as a geroprotector.
The four registry records were queried through the ClinicalTrials.gov API v2 by the retrieval lane and the posted resultsSection values quoted directly; the PEARL effect sizes were read from the PMC full text. The master did not independently re-query the registry for these four, which is a weaker verification than the population series in this cycle received.
- Partial epigenetic reprogramming reached a first-in-human trial in March 2026, with safety-only primary endpoints
NCT07290244, sponsored by Life Biosciences, tests ER-100, a single intravitreal dose of an AAV-delivered doxycycline-inducible epigenetic therapy, in open-angle glaucoma and non-arteritic anterior ischaemic optic neuropathy. It is Phase 1, recruiting, 18 participants estimated, actual start 2 March 2026, first posted 18 December 2025, primary completion estimated May 2027 and study completion estimated March 2032. All 22 primary outcome measures are safety measures - treatment-emergent adverse events, dose-limiting toxicities, laboratory values, intraocular pressure, best-corrected visual acuity, OCT and pattern ERG. Sponsor-field searches for Altos Labs, NewLimit, Turn Biotechnologies and Retro Biosciences each returned zero registered records, so Life Biosciences is the only reprogramming company with a human trial.
Limit: This is the strongest genuinely new supporting evidence in the cycle and it is also thin: an 18-person Phase 1 with no efficacy primary endpoint, in two eye diseases, completing in 2032. It demonstrates that the reprogramming class has cleared preclinical safety review far enough to enter humans. It demonstrates nothing about rejuvenation, nothing about systemic delivery, and nothing about healthspan.
The registry record was retrieved through the ClinicalTrials.gov API v2 by the retrieval lane, which quoted the status, enrolment, dates and the count of safety primary outcomes. The four zero-result sponsor searches were run as explicit negative queries. The master did not re-query the registry directly. The second URL is a cross-reference to the PEARL record queried in the same session rather than a second source for this claim, so this record is effectively single-sourced.
- The entire FDA guidance corpus contains no ageing document, and the nearest artefact disclaims being one
The complete FDA guidance-document dataset, 2,787 records, was downloaded and matched with word-boundary regular expressions over eight terms - aging, ageing, healthspan, geroscience, longevity, senescence, gerotherapeutic, intrinsic capacity - returning zero matches. A naive substring pass had returned 28 hits, all of them from the words packaging and imaging, so the first pass would have supported the opposite conclusion. The nearest regulatory artefact is a consensus statement dated 18 May 2026 prepared by an ARPA-H convened working group, proposing a five-domain intrinsic-capacity construct as a trial-grade endpoint. It is addressed TO the FDA and is not agency guidance. It disclaims its own reading in terms worth quoting: it does not claim that intrinsic capacity is a validated surrogate endpoint for mortality or disability, nor that a single global composite is currently ready as a universal primary endpoint for drug registration, and it concedes that no universally accepted trial-grade scoring algorithm has been established. It advises sponsors to use intrinsic capacity as a secondary endpoint.
Limit: The measurement covers guidance titles and metadata, not full document text, and the lane could not enumerate the CDER/CBER drug-development-tool qualification database because the ddt.fda.gov API was unreachable. A pending or non-public biomarker qualification submission for an ageing endpoint therefore cannot be excluded on this evidence. This is an absence established over one corpus, not proof that no regulator anywhere is moving.
The guidance dataset was downloaded and both the naive and word-boundary counts reproduced by the retrieval lane, which reported the command and the counts. The ARPA-H PDF was read directly from the saved binary and the two disclaiming sentences are verbatim quotes from its sections 1.2 and 2.1.2.
- The flagship metformin ageing trial has still never registered or enrolled anyone, eleven years after its FDA endpoint agreement
No ClinicalTrials.gov record exists for TAME. Three independent search strategies were used - the term TAME metformin, the title field for Targeting Aging which returned 20 unrelated results, and the sponsor field for Albert Einstein College of Medicine combined with metformin, which returned three records, none of them TAME: NCT03757910, NCT00618072 and the 16-participant MILES trial NCT02432287. AFAR's own page still lists under what is needed visionary donors to help launch the trial. The design is a multi-year, multi-site study of more than 3,000 adults aged 65 to 79 across 14 institutions over six years. It is not launched, has no secured funding, and has enrolled zero participants. The widely cited 2015 FDA agreement on a composite endpoint has produced no registered protocol in the eleven years since.
Limit: A trial that does not launch is evidence about funding and institutional appetite, not directly about biology - the seed already recorded that reading in hf-day-0003-takeaway-02. What is new here is duration: an eleven-year gap between a regulator's endpoint agreement and any registered protocol converts a reading of immature into a reading of structurally stuck, which is a stronger claim and the one this record supports.
Registry absence was established by three distinct queries reported by the retrieval lane, and the AFAR funding language was read from the sponsor's own page. The master did not re-run the registry queries. Cycle 3 recorded the same non-launch through hf-evidence-0008, so this is the second observation of the same continuing state rather than a new independent occurrence.
- Healthy life expectancy at the demographic frontier fell after 2019 and its gap against total life expectancy widened in all five leading countries
WHO Global Health Estimates healthy life expectancy at birth, both sexes, for the five longest-lived populations: Japan 71.11 in 2000, 73.58 in 2019, 73.40 in 2021; Singapore 69.41, 73.80, 73.65; Republic of Korea 66.55, 72.50, 72.45; Switzerland 68.35, 71.53, 71.15; Spain 68.89, 71.69, 71.11. HALE fell between 2019 and 2021 in every one of the five. The gap between life expectancy and healthy life expectancy widened in every one of the five between 2000 and 2021: Japan 10.42 to 11.07 years, Singapore 9.11 to 10.21, Korea 9.31 to 11.35, Switzerland 11.36 to 12.18, Spain 10.20 to 11.56. Korea widened most at 2.04 years. The series ends at 2021; no WHO HALE exists for 2022 to 2024. GBD 2021 independently reports global HALE falling from 63.6 years in 2019 to 62.2 in 2021.
Limit: WHO HALE is a model estimate, not a measurement: it applies the Sullivan method to GBD-derived years-lived-with-disability prevalence, so it inherits the assumptions of the disability weights. The 2019-to-2021 fall is substantially a pandemic effect and should not be read as a trend. The load-bearing part of this record is the twenty-year gap widening, which predates the pandemic and appears in all five countries.
The master pulled both WHO indicators directly from the GHO OData API in this session and computed every value and gap above first-hand, rather than accepting the retrieval lane's figures. The lane's independently derived numbers matched to two decimal places. One trap was found and corrected in the process: the sex dimension code is SEX_BTSX, not BTSX, and the wrong code returns an empty result set rather than an error.
- The only healthy-life series published past 2021 is too noisy at country level to resolve the effect this forecast is about
Eurostat healthy life years is published through 2024, dataset hlth_hlye, last updated 17 July 2026, and is the only healthy-life series anywhere extending past 2021. The EU-27 aggregate for women recovered above its pre-pandemic level: 65.1 years in 2019, 62.8 in 2022, 63.3 in 2023, 65.5 in 2024. The frontier did not move: the 2019 maximum was Malta at 73.5 and the 2024 maximum is Italy at 73.8, a gain of 0.3 years in five years. Underneath that flat frontier the country rankings churn implausibly. Sweden's women fell 72.7 to 66.0 and Spain's went 70.4 in 2019, 60.6 in 2022, 68.6 in 2024. Italy gained 4.2 years in the single step from 2023 to 2024. The 2024 ranking is internally implausible on its face: Bulgaria ranks second in Europe at 71.4 healthy years against a life expectancy of 79.5, giving 8.1 unhealthy years, while Switzerland reports 56.5 healthy years against a life expectancy of 85.9, giving 29.4 unhealthy years. Switzerland reports 3.6 times Bulgaria's burden of unhealthy years.
Limit: This is a record about an instrument, not about human health, and it does not show that European health is or is not improving. Healthy life years applies the Sullivan method to a single self-reported activity-limitation question from EU-SILC, so cross-country differences partly measure what respondents in different countries call a limitation. The Bulgaria-Switzerland inversion is strong evidence of that and weak evidence about anything else. It does not invalidate the EU aggregate trend, which is computed over a stable population and moves smoothly.
The master queried the Eurostat dissemination API directly in this session, decoded the JSON-stat payload, and computed every value, ranking and unhealthy-year difference above first-hand. The 17 July 2026 update timestamp was read from the live payload. Two dimension traps were hit and resolved: the indicator dimension is hlth_hle rather than indic_he, and the indicator code is HLY_Y0 rather than HLY.
- Seventy-two percent of the mice used to test lifespan drugs die of cancer, against roughly a fifth of humans
Full-necropsy pathology of the NIA Interventions Testing Program's UM-HET3 mice found neoplasia to be the cause of death in 72 percent of animals - 77 percent of females and 65 percent of males, rising to 85 and 72 percent among those with an identified cause. Cancer accounts for roughly 17 to 20 percent of deaths in high-income human populations, which are dominated instead by cardiovascular and neurodegenerative disease, and mice do not spontaneously develop human-type atherosclerosis or Alzheimer's disease. Several of the agents the programme has found to extend lifespan - rapamycin, trametinib, canagliflozin - have direct anti-tumour activity.
Limit: This does not show that rapamycin fails to slow ageing, and a drug that suppresses cancer in a cancer-prone species may still slow ageing. What it shows is that the assay's headline output is confounded: an intervention that is purely oncostatic and does nothing to ageing would score well on this endpoint. It bears on how much weight the mouse effect sizes can carry, which is the specific role they played in this forecast's prior update.
The necropsy figures were read from the fetched GeroScience paper by the retrieval lane. The master did not re-fetch this paper. The second URL is the cross-species shrinkage review used in hf-evidence-0120 and is a corroborating source for the translation critique generally, not a second measurement of the neoplasia fraction, so the 72 percent figure itself rests on one source.
- The same drugs produce monotonically smaller lifespan gains as species get more complex, and humans are already a threefold allometric outlier
A cross-species review reports the same compounds declining in effect as organism complexity rises. Resveratrol: about 70 percent lifespan extension in yeast, about 41 percent in flies, about 30 percent in worms, about 26 percent in rodents, and no statistically significant effect in humans. Rapamycin: about 57 percent in yeast, about 29 percent in flies, about 25 percent in worms, 13 percent in mice. Metformin: about 50 percent in yeast, negligible in humans. The authors attribute this to pathway redundancy and species-specific metabolism, and recommend healthspan rather than lifespan as the realistic human target. A separate allometric point compounds it: by body mass alone humans would be predicted to live about 27 years, so human longevity is already roughly a threefold outlier above the mammalian allometric prediction, and the headroom argument that applies to a two-year mouse does not obviously transfer.
Limit: The review is a synthesis of heterogeneous studies run under different protocols, so the percentages are not directly comparable to one another and the monotonic pattern is partly an artefact of which studies exist. The allometric argument is a heuristic, not a mechanism, and species that are allometric outliers in one direction are not thereby barred from moving further. The claim this supports is directional - expect attenuation on translation - not a specific predicted human effect size.
The cross-species table was read from the fetched Aging Cell full text by the retrieval lane. The master did not re-fetch it. The second URL corroborates the general translation critique rather than the specific percentages.
- Where lifespan and healthspan are measured in the same animals, the frailty-adjusted gain is smaller than the raw lifespan gain
Frailty-adjusted analysis of calorie-restricted C57BL/6J mice, using the FAMY measure of years lived in perfect health and the 22-item GRAIL health assessment, found median lifespan raised 16 percent in males and 39 percent in females while FAMY rose only 15.6 and 31.9 percent - so in both sexes the healthy-year gain was smaller than the raw lifespan gain, and restricted mice dying at equivalent ages were only marginally healthier. Frailty removed roughly 16 to 20 percent of maximum possible healthspan across the lifespan. The decoupling runs in both directions: protein restriction improved GRAIL healthspan with no lifespan extension, while isoleucine restriction in males gave 33 percent lifespan against 29 percent FAMY. Independent cross-sectional work in C57BL/6Nia mice at 4, 20, 28 and 32 months likewise concluded that lifespan and healthspan are not necessarily correlated.
Limit: These are single mouse strains under dietary rather than pharmacological intervention, and FAMY and GRAIL are constructed measures whose relationship to human healthy life expectancy is an analogy rather than a mapping. The gap between lifespan and healthspan gain here is small - roughly half a percentage point in males, seven in females - so this evidence weakens the assumption that lifespan gains transfer fully to healthy years; it does not show that they mostly fail to.
Both papers were fetched by the retrieval lane and the figures quoted from their text. The lane flagged that the first was retrieved as a preprint whose peer-reviewed version exists in Nature Aging, and that the numbers should be re-pulled from the journal version before publication. The master did not re-fetch either; this record carries that flag forward rather than resolving it.
- An antibody given only from late middle age extended mouse median lifespan by 22.5 to 25 percent
An anti-IL-11 antibody administered from 75 weeks of age to death raised median lifespan from 120 to 155 weeks, 22.5 percent in males and 25 percent in females, with reduced cancer deaths, less fibrosis and slower metabolic decline. Germline deletion of Il11 extended lifespan 24.9 percent as a mean across both sexes. The result matters disproportionately for this forecast because dosing begins after most of life has already elapsed, which is the situation any human geroprotector would face, and because the target already has human safety data from fibrosis programmes.
Limit: This is one paper, in mice, not yet replicated by the NIA Interventions Testing Program, which is the field's own replication machine and has failed to reproduce lifespan extension for metformin, resveratrol and nicotinamide riboside. It is also subject to the two discounts recorded in hf-evidence-0119 and hf-evidence-0120: reduced cancer deaths in a species where 72 percent of deaths are neoplastic, and an expected attenuation on translation.
The paper was located and its figures reported by the adversarial retrieval lane, which also ran a targeted search for PubPeer or research-integrity concerns on it and found none - reported as absence of contrary evidence rather than as endorsement. The master did not re-fetch the paper. The second URL is the ITP rapamycin paper, which corroborates the mechanism class's effect magnitude rather than this specific result.
- The most recent replication-grade mouse cohort found three new life-extending compounds, all male-only and all single-digit
The NIA Interventions Testing Program's C2021 cohort, run across three sites in genetically heterogeneous UM-HET3 mice, found three compounds significantly extending median lifespan, all in males only: epicatechin about 5 percent, p = 0.037, 141 males; halofuginone about 9 percent, p = 0.002, 154 males; mitoglitazone about 9 percent, p = 0.015, 149 males. No compound extended female lifespan - epicatechin p = 0.6, halofuginone p = 0.75 - and mitoglitazone trended toward shortening it, p = 0.07. Epicatechin and halofuginone also raised 90th-percentile survival about 6 to 7 percent in males. 2BAct, dichloroacetate and forskolin showed no significant effect in either sex. The paper states that fourteen of the sixty agents previously tested by the programme significantly increased lifespan in one or both sexes.
Limit: Single-digit, single-sex effects in the field's most rigorous assay are a weaker signal than the 23 to 26 percent rapamycin result that anchored this forecast's prior update, but they are not evidence that larger effects are unavailable - rapamycin was found by the same programme. The sex-specificity is the more durable point: an intervention that works in one sex halves its population effect before translation is even considered.
The master fetched this paper directly and read the per-compound percentages, per-sex sample sizes and p-values first-hand. Doing so corrected the retrieval lane, which had reported the programme's cumulative total as fifteen agents plus two combinations; the paper's own text says fourteen of sixty previously tested agents.
- The largest documented population life-expectancy gain from a single intervention class is 11.3 years, and it was recovery rather than frontier extension
In a population cohort of over 101,000 individuals in rural KwaZulu-Natal, South Africa, adult life expectancy was 49.2 years in 2003, the year before antiretroviral therapy became available in the public-sector health system, and 60.5 years by 2011 - an 11.3-year gain in eight years. A comparable gain of about ten years was independently observed in rural Malawi between 2005 and 2014. This is the fastest and largest population life-expectancy gain attributable to one identifiable intervention class in the documented record, and it is approximately half the twenty healthy years this forecast's resolution criterion requires.
Limit: Two limits decide how this may be used, and both cut against reading it as an encouraging precedent. It is life expectancy, not healthy life expectancy - no HALE equivalent was published for this cohort - so it does not speak directly to the quantity this forecast measures. And it is a population recovering from an epidemic-depressed baseline back toward its own prior level, not a frontier population being pushed beyond the best value ever achieved. The forecast requires the second kind, and this record is the best available instance of the first kind.
The master fetched the Science abstract directly through the NCBI eutils API and read the 49.2, 60.5, 11.3 and 101,000 figures first-hand from the primary abstract text, after science.org returned HTTP 403 and PubMed's own page returned only a cookie wall.
- Two randomised human trials slowed a validated pace-of-ageing biomarker, while static epigenetic clocks did not move
CALERIE-2 randomised 220 non-obese adults to 25 percent caloric restriction or ad libitum eating for two years and found DunedinPACE slowed by roughly 2 to 3 percent, Cohen's d 0.3 at 12 months and 0.2 at 24 months, p below 0.01 at both points, while PhenoAge and GrimAge showed no significant change. A separate 32-week double-blind randomised trial of semaglutide in HIV-associated lipohypertrophy, 84 participants randomised 45 to drug and 39 to placebo, found DunedinPACE lower by 0.09 units, about 9 percent slower, p = 0.01, with PhenoAge lower by 4.9 years, p = 0.004, PCGrimAge by 3.1 years, p = 0.007, and OMICmAge by 2.2 years, p = 0.009.
Limit: A biomarker is not an outcome, and this is the weakest kind of human evidence in the cycle. DunedinPACE is a measure of the rate of decline calibrated against a birth cohort; no trial has yet shown that slowing it produces additional healthy years, and the divergence in CALERIE-2 between DunedinPACE and the static clocks shows the biomarker family does not agree with itself. The semaglutide trial is also in a specific patient population with a metabolic abnormality, so its effect may be treatment of that abnormality rather than an ageing effect.
Both trials were located and their figures reported by the adversarial retrieval lane. The master did not re-fetch either. The lane flagged that the semaglutide result was retrieved as a preprint whose peer-reviewed version exists in Nature Communications, and that the figures should be re-pulled from the journal version before publication; this record carries that flag forward rather than resolving it.
- The published case that radical life extension this century is implausible stands unrebutted in the peer-reviewed literature
Olshansky and colleagues analysed national vital statistics from 1990 to 2019 for the eight longest-lived populations plus Hong Kong and the United States and found life-expectancy improvement decelerating, resistance to further gains rising, lifespan inequality falling and mortality compressing, concluding that survival to age 100 is unlikely to exceed 15 percent for females and 5 percent for males and that radical life extension this century is implausible without markedly slowing biological ageing. A targeted search for peer-reviewed rebuttals found none. The one located reply, by Vinicius and Migliano, argues that Olshansky tested country-level trends rather than the best-practice series of the annual world-leading population, and reports that best-practice female life expectancy rose from 84.85 to 87.75 years over the first two decades of this century, a linear 1.45 years per decade with no visible ceiling. It is a bioRxiv preprint posted 2 May 2025 and is not peer-reviewed.
Limit: The rebuttal's own number works against the claim even if it is accepted in full. A best-practice improvement rate of 1.45 years per decade, sustained without deceleration, produces about 14.5 years of total life expectancy over the century to 2126 - and this forecast requires twenty additional HEALTHY years, against a gap that hf-evidence-0117 shows is widening. The strongest published optimistic case for the frontier is therefore still short of the threshold, which is the load-bearing point of this record.
Both were located by the population retrieval lane, which fetched the preprint abstract after two HTTP 429 responses on the full-text paths and reported the absence of any peer-reviewed rebuttal as an explicit negative search result. The master did not re-fetch either.