Do Ozempic and Mounjaro Reverse Biological Aging? What the Aging Clocks Actually Measured

What did Novo Nordisk and Eli Lilly actually report?

Two presentations at one conference, backed by very different amounts of published material.

Both companies spoke at the same meeting. The Aging Research and Drug Discovery conference ran October 1 to 3, 2026 at the David Rubenstein Treehouse at Harvard University. The program lists Kevin Duffin, Lilly's vice president for aging research, on "Modulation of Aging Clocks by the GLP-1/GIP dual agonist, Tirzepatide" on October 2, and Alejandro Aguayo-Orozco, a Novo Nordisk senior scientific director, on "Semaglutide and Healthspan: Decoding the Emerging Evidence. From clinical outcomes to organ-specific biological age" on October 3.⁴

Novo's talk has a paper behind it. The analysis sits in a preprint posted to Research Square on September 22, 2026, under the title "Semaglutide reduces proteomic organ-specific biological age across randomized clinical trials".¹ Its abstract states the design and the result plainly:

"We analyzed proteomics from 10,052 participants across five placebo-controlled semaglutide trials using externally trained organ-specific mortality clocks, proteomic frailty scores and disease-risk models."

"Semaglutide reduced heart biological age by approximately 2 to 4 years versus placebo, with reductions of 1 to 4 years across kidney, brain, pancreas, lung and multiorgan clocks."

The same abstract then draws the line itself, in its closing sentence. The findings support the clocks' "potential as treatment-response readouts without establishing biological rejuvenation, causal mediation or surrogate validity."¹

Lilly's talk has no paper behind it. No trial name, no participant count, no comparator arm, no clock names and no effect size appear in the conference program or in any report of it that could be read for this page. What exists is Duffin's description to reporters.

	Novo Nordisk	Eli Lilly
Drug	Semaglutide	Tirzepatide
Clock type	Proteomic, organ-specific	Epigenetic (DNA methylation)
Participants	10,052	Not stated anywhere
Comparator	Placebo, in all five trials	Not stated anywhere
Written source	Preprint, September 22, 2026	None. Conference presentation only
Peer reviewed	No	No
Effect size given	Heart 2 to 4 years, other organs 1 to 4	None

Sources: the Research Square preprint and the conference program, both read at source, plus the reporting described below.¹⁴

Where does the "two to three years" figure come from?

An interview, not the analysis.

The preprint gives a range of roughly 2 to 4 years for the heart clock and 1 to 4 years across five other clocks. It does not report a single headline number for the body as a whole. The round figure that traveled came from an interview. Nikolaj Roed, Novo's global project leader, put the difference at "two to three years" when speaking to MIT Technology Review for its October 6 report.²

By October 11 the framing had reached a general audience as "GLP-1 drugs can 'reverse biologic aging' by two to three years", with the copy describing "a decreased age of around two or three years" in overweight or diabetic patients.⁵

So the chain runs: a preprint with a spread of organ-specific ranges, then a spokesperson's summary figure in an interview, then that figure as a headline with the word "reverse" attached. Each step is defensible on its own. The end of the chain says something the start of it does not.

Novo's own scientist was more careful than his company's headline figure. Aguayo-Orozco told the same reporter, "I think it's pretty clear that the organ age is shifting", and framed the question as "When you intervene with semaglutide, does it actually move the clocks in any direction? And the answer is yes."² Moving a clock in some direction is a narrower claim than reversing aging. It is also the claim the data supports.

What is a proteomic aging clock measuring?

A protein pattern that predicts death, scaled into years.

The preprint never names the clocks it used, beyond calling them externally trained, so which specific models produced its numbers is not public. The clearest published account of how a clock of that kind gets built came out in Nature in December 2023, from Hamilton Se-Hwee Oh, Jarod Rutledge, Daniel Nachun and colleagues at Stanford.⁶ Of roughly 5,000 plasma proteins measured in more than 5,600 adults, almost 900 turned out to be enriched in a single organ. Machine learning models were then trained on those proteins to estimate an age for each of 11 organs: fat tissue, arteries, brain, heart, immune tissue, intestines, kidneys, liver, lungs, muscle and pancreas.

What those models predict is risk. People with hypertension and diabetes had "older" kidneys than same-aged peers. Older hearts tracked with atrial fibrillation and heart attacks. For most organs, accelerated aging carried a 15 to 50% greater risk of death from any cause.⁶

Novo describes its clocks as "externally trained organ-specific mortality clocks."¹ Mortality is the training target. So a heart biological age three years younger means one thing. The blood now carries a protein pattern closer to that of people who died later. It does not mean a cardiologist would find a younger heart.

The National Institutes of Health summary of the Stanford work added one caveat. The technique "will need to be tested on many more people before it can be used in the clinic."⁶ That was about measuring organ age at all, before anyone tried to move it with a drug.

Who was actually in those five trials?

People with serious cardiovascular or kidney disease. Not the people who read headlines about anti-aging.

The preprint names its five trials only in its keyword list, not in the abstract: SELECT, STEP-HFpEF, STEP-HFpEF DM, FLOW and SOUL.¹ Those are outcome trials in cardiovascular disease, heart failure with preserved ejection fraction, and chronic kidney disease in type 2 diabetes. Entry to them required established illness or substantial risk of it. The preprint reports changes "during up to 156 weeks of follow-up", so the longest of these observations runs about three years.¹

That matters for reading the result in either direction. Take a drug that lowers a death-linked protein signature in people whose hearts and kidneys are already failing. It may well be doing something real and specific to that illness. Whether it does the same in a healthy forty-year-old is a separate question. Nobody has run that study.

Two researchers quoted in the original reporting made exactly this point. Steve Horvath, the UCLA professor who built several of the best-known clocks, called it "a pretty strong effect". He then said, "I did not yet see sufficient evidence that they will benefit skinny people", and that he is not ready to take the drugs himself.⁷ Vadim Gladyshev, the Harvard biologist who assisted Novo with its molecular measurements, put it more bluntly still: "But in a healthy population, no one knows."²

What did the Lilly tirzepatide analysis find?

Nobody outside the room can say.

Duffin's two public statements are the whole of it. "All the clocks are telling a consistent story that we're seeing a reduction in age", and "It's not like we're going to reverse age by 30 years or something, but it's a significant reduction."²⁷ MIT Technology Review described the Lilly work only as "smaller" than Novo's, and reported that molecular time "moved more slowly for people on the drug" without a number attached.²

A note on something circulating about this analysis. It has been described in places as a subgroup of SURMOUNT-5, small and single-arm. None of the sources read for this page states which trial the samples came from, how many people were in it, or whether there was a comparator. The conference program gives a title and a time slot. Until Lilly publishes, treat the trial name and the design as unconfirmed.

Duffin's second quote is the useful one, because it cuts against his employer's interest. A vice president of aging research volunteering that the effect is not thirty years is managing expectations. That is a reasonable thing to do when the headline risk runs the other way.

Is there any peer-reviewed evidence?

One randomized trial, published in May, in 84 people.

Nature Communications published "Semaglutide slows epigenetic aging in a randomized trial of HIV-associated lipohypertrophy" on May 19, 2026.³ This is the best-designed evidence in the story, and it drew the least coverage. That is the usual pattern when a careful result lands in a small population.

The parent trial was a 32-week, randomized, double-blind, placebo-controlled phase 2b study in adults with HIV-associated lipohypertrophy, a disorder of abnormal fat accumulation. Its numbers, from the methods: 154 people screened, 108 randomized at 54 per arm, 92 completing both the week 0 and week 32 assessments, and 84 with paired samples available for the epigenetic work. That left 45 on semaglutide and 39 on placebo. Participants received once-weekly semaglutide, titrated over eight weeks and then held at 1.0 mg for 24 weeks, which is the trial's protocol and not a recommendation.³

The analysis calculated 17 DNA methylation clocks plus 11 system-specific ones. The significant results, quoted as the paper reports them:

Clock	Result	p
PhenoAge	−4.9 years/year	0.004
PCGrimAge	−3.1	0.007
GrimAge V2	−2.3	0.009
OMICmAge	−2.2	0.009
RetroAge	−2.2	0.030
DunedinPACE	−0.09 units, 9% slower	0.01
Intrinsic Capacity	0.003 IC units, 95% CI −0.003 to 0.009	0.312

From the paper, read at source.³ Note the unit on the first row. PhenoAge is reported in years per year, a rate of change, not a total of 4.9 years knocked off anyone's age. Note also the last row, a system-specific clock that moved in the other direction and did not reach significance.

The authors state that "epigenetic aging was not pre-specified", and they list the limits themselves: "The post hoc design, modest sample size, HIV-specific cohort, and 32-week follow-up limit generalizability."³

Why do different clocks give different answers?

Because they were trained on different things, and the field has measured how much they disagree.

Look again at the table above. The same blood, from the same 84 people, at the same two timepoints, produced −4.9 on one clock and −2.3 on another, and no significant movement at all on a third. Any of those numbers could be a headline.

Nature Medicine published the systematic version of this problem on August 21, 2026.⁸ Researchers took 51 longitudinal interventional studies, 3,128 paired pre- and post-intervention samples, and calculated 16 prominent epigenetic clocks plus 94 other DNA methylation biomarkers for every one of them.

What they found about agreement is the part to carry away. Reliability-optimized clocks agreed with each other far more often than the originals they were built from, with PCGrimAge agreeing with other clocks about 90% of the time against roughly 60% for GrimAge. First-generation clocks agreed least, and biomarkers such as Horvath1, Horvath2 and Hannum "sporadically increased or decreased" across interventions with no clear trend.⁸

They also found which clocks move most readily. Clocks "trained to predict mortality or pace of aging show the strongest responses across all interventions", with DunedinPACE showing the largest average decrease.⁸ Both the Novo analysis and the HIV trial lean on that family. It is a reasonable choice, and it is also the family most likely to move.

How much of a clock reading is just noise?

Enough that a meal can move it.

The part of the field that undercuts every number above comes from the people who build the clocks. A preprint posted in October 2025 by Raghav Sehgal, Daniel Borrus, John Gonzalez, Yaroslav Markov and Albert Higgins-Chen split clock reliability into two questions that are usually treated as one.⁹ Higgins-Chen built the principal-component clocks that keep appearing in this story, so this is not an outside critic.

The first question is technical. Run the same blood twice and do you get the same number? Mostly yes. "The majority of clocks achieved excellent technical reliability (ICC > 0.9)."⁹

The second question is biological, and the answer is worse. Sample the same person again a short time later, with nothing done to them, and the readings move. "Most clocks achieved only 'moderate' to 'good' reliability", the paper reports, "with no clock reaching the 'excellent' range." The intraclass correlations behind that sit at roughly 0.4 to 0.7.⁹

Those two properties do not travel together. The authors report that "technical and biological reproducibility are largely independent (r = 0.0168)", and that "some of the most technically robust clocks, such as GrimAgeV2 and DunedinPACE, were among the most biologically fragile."⁹ Both of those clocks carry headline results in the 84-person trial above.

Then there is breakfast. After a meal, reliability "dropped markedly, with many clocks performing only in the 'moderate' or 'poor' range."⁹ The authors draw the conclusion themselves: if a meal can dramatically affect a score, an intervention study without fasting control "may detect changes that are mistakenly attributed to the intervention."⁹

None of the three GLP-1 analyses states whether its samples were drawn fasting. That is not evidence that any of them got it wrong. It is a question none of the coverage asked, and the answer would change how much a two-year shift is worth.

Treat this preprint as a preprint. It has not been peer reviewed, and the version read for this page carries two internal inconsistencies: an altitude sub-study described as seven timepoints over 21 days in the results and four over seven days in a figure legend, and a discussion citing brain-volume findings that do not appear in the results. Those do not touch the reliability figures quoted here, and they are a reason to weigh the paper rather than lean on it.

Does moving a clock mean you will live longer?

Nobody has shown that, and three separate 2026 papers say so in their own words.

The Nature Medicine review states that "it is still unclear whether short-term biomarker changes correspond to long-term improvements".⁸ It then names the specific gap. "The absence of an empirically defined minimal clinically important difference for DNAm clocks remains a key barrier" to using them as surrogate endpoints.⁸ Nobody has agreed how many years on a clock would count as a real benefit.

Nature Biotechnology published an attempt to use proteomic clocks as trial readouts on September 7, 2026, and reached a similar place.¹⁰ Its authors report that all six clocks they tested predicted lower biological age in treated arms. They then write that "proteomic clocks alone cannot deconvolute aging- and disease-specific effects". They add that "Establishing causality remains a challenge". On the standing of the endpoint itself, clock readings "could be prespecified as exploratory biomarkers, with clinical claims requiring conventional clinical endpoints or a qualified biomarker."¹⁰

The HIV trial's authors are the most direct of the three. Their limitations section calls the tools what they are: "epigenetic clocks are statistical predictors rather than mechanistic measures of aging". On what a reading means, it says "changes in clock-derived age acceleration should not be interpreted as direct evidence of altered lifespan". On the tissue itself: "These clocks should not be interpreted as definitive evidence of tissue-level aging or rejuvenation."³

Set that beside the Novo abstract and the position is the same across every group involved. None of that caution comes from outside critics. It is in the papers.

Who has a stake in the answer?

The companies whose drugs are being tested. The preprint's declarations run long.

The Research Square declarations name several authors as "employees of and hold shares in Novo Nordisk A/S", with one more an employee and shareholder of Novo Nordisk Research and Development UK Limited and another a former employee and current shareholder.¹ Among the academic co-authors, Helen M. Colhoun declares advisory work for Bayer, Novo Nordisk and Roche along with stock in Eli Lilly and Roche, and Vadim Gladyshev declares advisory panels including Retro Biosciences, Dior, LongeVC and Thorne. Donna Ryan's disclosure runs to more than twenty companies.¹ What the page does not carry is a funding statement, so what is on the record is employment and shareholding rather than who paid for the work.

Lilly's own vice president for aging research presented its result.

None of this makes the findings wrong. Drug companies run the trials that produce this kind of data, and the alternative is nobody measuring it. It does mean something narrower. The first full account of these numbers is being written by people with a financial stake in the answer, before peer review. That is a reason to wait, not a reason to dismiss.

Has this story run before?

Yes, in August 2025, and with a bigger-sounding number.

The same HIV lipohypertrophy trial circulated as a medRxiv preprint in 2025. Headlines then had Ozempic shaving three years off biological age, with one outlet giving the figure as 3.1 years.¹¹ That figure is the PCGrimAge row in the table above. One clock out of 17, in 84 people with a specific HIV-related condition. The work then went through peer review and published in Nature Communications in May 2026.

So anyone meeting this subject in October 2026 is seeing three separate things presented as one event. An 84-person randomized trial, first reported in 2025 and published in May. A 10,052-person preprint posted in September. Then an unpublished conference talk from a second company. They involve different drugs, different clock families, different populations and very different evidence quality.

What would actually settle this?

No trial here was built to answer the question.

Every analysis above is secondary work on trials run for something else. The Novo preprint uses stored samples from cardiovascular and kidney outcome studies. The HIV trial's epigenetic analysis was "not pre-specified" by the authors' own statement.³ Lilly's remains unpublished.

From what the three 2026 papers identify as missing, the list is reasonably clear. Aging endpoints prespecified before the trial starts, not examined afterwards. A population without established cardiovascular or kidney disease, since that is where the open question sits. A clinical outcome measured alongside the clock, so the two can be compared. Then a defined threshold for how much clock movement would count, which Nature Medicine names as a key barrier.⁸

The HIV trial's authors put the same thought in one sentence: "Prospective trials are needed to determine whether GLP-1 receptor agonists can be repurposed as gerotherapeutics."³

Until one runs, one statement holds up. Semaglutide shifts protein and methylation patterns linked to earlier death, in people already at high risk of dying, by an amount that depends on which model you ask.

Common questions

Does Ozempic make you biologically younger?

What does a two-year reduction in biological age actually mean?

Where did the two-to-three-year figure in the headlines come from?

Has any of this been peer reviewed?

Is tirzepatide better or worse than semaglutide for this?

Were the people in these trials healthy?

Why do different clocks give different numbers for the same person?

What is the difference between biological age, chronological age and lifespan?

Does this mean peptides or GLP-1s are longevity drugs?

Did weight loss cause the effect?

Is the 3.1-year figure I saw last year the same research?

Should healthy people take a GLP-1 to slow aging?

Can I test my own biological age and see this?

Is there any aging-clock data for retatrutide?

When will we know more?

Sources that could not be read for this page

• CNN, "Could GLP-1 drugs be anti-aging medications?", October 13, 2026. Refused automated access. Its existence confirms the story was still running as this page was written, but nothing from it is used.
• The New York Post article of October 11, 2026. Could not be located. The headline wording is cited from a syndicated reproduction at footnote 4, and the attribution to the Post is not confirmed here.
• ScienceDirect, the parent trial of the HIV lipohypertrophy study in Lancet Diabetes & Endocrinology. Refused automated access. Its participant numbers are reported at footnote 7 from the Nature Communications methods instead.
• PubMed and PubMed Central. Both returned a browser-check page rather than content. The Oh et al. citation at footnote 5 was confirmed through the Crossref registry instead, and the Nature papers were read on the publisher's own site.

Sources

1. Dermit M, Colhoun HM, Goeminne LJE, Ryan D, et al. Semaglutide reduces proteomic organ-specific biological age across randomized clinical trials. Preprint, Research Square, posted September 22, 2026, version 1. doi:10.21203/rs.3.rs-11082091/v1. Not peer reviewed. Abstract read at source; the full text was not available on the page, so methods detail beyond the abstract could not be checked. Quoted verbatim: "We analyzed proteomics from 10,052 participants across five placebo-controlled semaglutide trials using externally trained organ-specific mortality clocks, proteomic frailty scores and disease-risk models"; "Semaglutide reduced heart biological age by approximately 2 to 4 years versus placebo, with reductions of 1 to 4 years across kidney, brain, pancreas, lung and multiorgan clocks"; "Proteomic frailty and cardiometabolic disease-risk scores improved during up to 156 weeks of follow-up"; "Effects were not fully mediated by weight loss, glycated hemoglobin or C-reactive protein"; "Exploratory mediation analyses suggested that heart biological-age changes may account for 34% to 49% of effects on selected outcomes"; and the closing qualification that the findings support the clocks' "potential as treatment-response readouts without establishing biological rejuvenation, causal mediation or surrogate validity". The five trials SELECT, STEP-HFpEF, STEP-HFpEF DM, FLOW and SOUL appear in the keyword list, not in the abstract text. The author list is shown on the page as four names plus "and 8 more", so it is reproduced here as given. Declarations read at source and name multiple authors as employees of and shareholders in Novo Nordisk A/S, one as an employee of and shareholder in Novo Nordisk Research and Development UK Limited, and one as a former employee and current shareholder; Helen M. Colhoun declares advisory panels for Bayer, Novo Nordisk and Roche and stock in Eli Lilly and Roche; Vadim N. Gladyshev declares advisory panels for Dior, LongeVC, Retro Biosciences and Thorne. No funding statement appeared on the page. The page does not state whether the primary analyses were prespecified.
2. Regalado A. Weight-loss drugs show signs of slowing biological aging, say drugmakers. MIT Technology Review, October 6, 2026. technologyreview.com. Read at source. The original reporting on both presentations. Source of the "two to three years" figure, attributed to Nikolaj Roed. Reports the Novo analysis as 10,052 people with half on drug and half on placebo, blood drawn at baseline and again months later, and a slowdown of up to 4 years on a heart-protein clock. Describes the Lilly study only as "smaller" and reports no trial name, participant count, comparator arm, clock names or effect size for it. Quotes Kevin Duffin: "All the clocks are telling a consistent story that we're seeing a reduction in age" and "It's not like we're going to reverse age by 30 years or something, but it's a significant reduction." Quotes Aguayo-Orozco: "I think it's pretty clear that the organ age is shifting" and "When you intervene with semaglutide, does it actually move the clocks in any direction? And the answer is yes." Quotes Gladyshev: "But in a healthy population, no one knows." Quotes Horvath: "I did not yet see sufficient evidence that they will benefit skinny people." Notes the effect "varied widely, depending on which type of clock was used, and what organ was tested."
3. Semaglutide slows epigenetic aging in a randomized trial of HIV-associated lipohypertrophy. Nature Communications 2026;17:6606, published May 19, 2026. nature.com. Peer reviewed. Read at source. A post hoc exploratory epigenetic analysis of a 32-week randomized, double-blind, placebo-controlled phase 2b trial in adults with HIV-associated lipohypertrophy. Methods state: "A total of 154 individuals were screened, 108 were randomized (54 to each group)"; 92 participants completed both week 0 and week 32 assessments, and "of these, 84 had paired samples available for epigenetic analysis", giving semaglutide n = 45 and placebo n = 39. Once-weekly semaglutide, titrated over 8 weeks then 1.0 mg for 24 weeks. Seventeen DNA methylation clocks across three generations plus 11 system-specific clocks were calculated. Results quoted as reported: PhenoAge "( − 4.9 years/year, p = 0.004)", PCGrimAge "( − 3.1, p = 0.007)", GrimAge V2 "( − 2.3, p = 0.009)", OMICmAge "( − 2.2, p = 0.009)", RetroAge "( − 2.2, p = 0.030)", DunedinPACE "( − 0.09 units, 9% slower, p = 0.01)", and the Intrinsic Capacity clock at "0.003 IC units [95% CI: −0.003 to 0.009], P = 0.312", not significant. The abstract states "epigenetic aging was not pre-specified". Limitations quoted: "The post hoc design, modest sample size, HIV-specific cohort, and 32-week follow-up limit generalizability"; "epigenetic clocks are statistical predictors rather than mechanistic measures of aging"; "changes in clock-derived age acceleration should not be interpreted as direct evidence of altered lifespan"; "These clocks should not be interpreted as definitive evidence of tissue-level aging or rejuvenation". Abstract conclusion: "Prospective trials are needed to determine whether GLP-1 receptor agonists can be repurposed as gerotherapeutics." The parent trial is cited in its reference list as Eckard AR et al., Lancet Diabetes Endocrinol 2024;12:523-534, which was not read for this page.
4. Aging Research and Drug Discovery 2026, conference program and recap. Aging Pharma. agingpharma.org. Read at source. Confirms the meeting ran October 1 to 3, 2026 at the David Rubenstein Treehouse, Harvard University. Lists Kevin Duffin, VP Aging Research, Eli Lilly, presenting "Modulation of Aging Clocks by the GLP-1/GIP dual agonist, Tirzepatide" on October 2 from 10:40 to 11:00; Alejandro Aguayo-Orozco, Senior Scientific Director, Diabetes Insights, Novo Nordisk, presenting "Semaglutide and Healthspan: Decoding the Emerging Evidence. From clinical outcomes to organ-specific biological age" on October 3 from 12:10 to 12:30; and Nikolaj Roed, Global Project Leader, Novo Nordisk, presenting "GLP-1 and healthy ageing: potential evidence and unmet needs" on October 1. The program gives titles and times only. It carries no trial names, participant counts or results for any of the three talks.
5. Syndicated news copy, October 11, 2026, headlined "GLP-1 drugs can 'reverse biologic aging' by two to three years". newsbeep.com. Read at source, as a syndicated reproduction. Gives "a decreased age of around two or three years" for overweight or diabetic patients and says the drugs slowed aging "as much as four years in the case of a clock that looked at heart proteins." This copy has been described elsewhere as originating with the New York Post. The page carries no byline and names no originating publication, and the New York Post article itself could not be located, so the attribution is not confirmed here and the headline wording is cited from this reproduction only.
6. Oh HS, Rutledge J, Nachun D, et al. Organ aging signatures in the plasma proteome track health and disease. Nature 2023;624(7990):164-172. doi:10.1038/s41586-023-06802-1. Citation confirmed against the Crossref registry. The paper itself was read via the National Institutes of Health research summary, "Tracking organ aging and disease", dated January 9, 2024, at nih.gov, not at the journal, so the figures below are as that summary reports them. Almost 900 of roughly 5,000 plasma proteins measured were enriched in a single organ; machine learning models were trained on those proteins to estimate biological age across 11 organs, listed as fat tissue, arteries, brain, heart, immune tissue, intestines, kidneys, liver, lungs, muscle and pancreas, using data from more than 5,600 people across the adult lifespan. People with hypertension and diabetes had older kidneys than same-aged peers and older hearts were associated with atrial fibrillation and heart attacks. Quoted: "For most organs, accelerated aging led to a 15-50% greater risk of death from any cause", and the caveat that the technique "will need to be tested on many more people before it can be used in the clinic." Senior author Tony Wyss-Coray, Stanford University.
7. Brooks D. Ozempic and Zepbound Makers Report Slower Aging Clocks, but Unreviewed Data Do Not Prove Patients Age Slower. Medical Daily, October 8, 2026. medicaldaily.com. Read at source. Independent confirmation of the Duffin quotes and of Horvath calling the results "a pretty strong effect" while saying "I did not yet see sufficient evidence that they will benefit skinny people" and that he is not ready to take the drugs himself. States that the Lilly results "were presented at a conference rather than in a published paper", that Novo's analysis "has not been peer-reviewed", that "Several Novo authors are company employees who hold shares", and that "The evidence comes from after-the-fact analyses of trials designed for other purposes." Like every other source read for this page, it names no trial, participant count, arm structure, clock names or effect size for the Lilly analysis.
8. Responsiveness of epigenetic aging biomarkers to longevity interventions in humans. Nature Medicine 2026;32:3477-3490, published August 21, 2026. nature.com. Peer reviewed. Read at source. Analysed 51 longitudinal interventional studies with 3,128 paired pre- and post-intervention samples, calculating 16 prominent epigenetic clocks and 94 further DNA methylation biomarkers for each. Quoted: "clocks trained to predict mortality or pace of aging show the strongest responses across all interventions"; "pharmacological and lifestyle interventions drive the strongest responses from DNAm biomarkers"; "clocks with multiple subscores (that is 'explainable clocks') provide specificity and greater mechanistic insight"; "it is still unclear whether short-term biomarker changes correspond to long-term improvements"; and "the absence of an empirically defined minimal clinically important difference for DNAm clocks remains a key barrier". DunedinPACE had the largest average decrease, falling significantly in 16 interventions and rising in one, and never disagreed with other clocks, though it was the only significant clock in 17.6% of the interventions where it responded. PCGrimAge produced the most statistically significant results and agreed with other clocks about 90% of the time against roughly 60% for GrimAge. First-generation biomarkers including Horvath1, Horvath2 and Hannum "sporadically increased or decreased" with no clear trend.
9. Sehgal R, Borrus D, Gonzalez J, Markov Y, ADNI, Higgins-Chen A. Biological versus Technical Reliability of Epigenetic Clocks and Implications for Disease Prognosis and Intervention Response. Preprint, bioRxiv, posted October 14, 2025, version 2. doi:10.1101/2025.10.13.682176. Not peer reviewed. Title, author list and posting date confirmed against the Crossref registry; the text was read from the version 2 full PDF, whose captured content began at the Introduction, so the abstract itself was not read. Albert Higgins-Chen is the senior author and developed the principal-component clocks referenced elsewhere on this page. Quoted: "the majority of clocks achieved excellent technical reliability (ICC > 0.9)"; "Most clocks achieved only 'moderate' to 'good' reliability (ICC ~ 0.4-0.7), with no clock reaching the 'excellent' range"; "technical and biological reproducibility are largely independent (r = 0.0168)"; "some of the most technically robust clocks, such as GrimAgeV2 and DunedinPACE, were among the most biologically fragile"; that after a meal reliability "dropped markedly, with many clocks performing only in the 'moderate' or 'poor' range"; and the authors' warning that if a meal can dramatically impact a score, an intervention study without fasting control "may detect changes that are mistakenly attributed to the intervention". PC-based clocks and SystemsAge performed somewhat better on biological reliability but still fell short; GrimAge I and II, DNAmEMRAge and DunedinPACE clustered near the lower end. Author limitations quoted: "Our analyses remain constrained to specific populations and conditions"; "Causal mechanisms underlying biological instability remain unclear"; "Longitudinal studies spanning years are required to determine whether biological instability resolves over longer intervals". Two internal inconsistencies were noted in the captured text and are disclosed on the page: an altitude sub-study given as seven timepoints over 21 days in the results and four over seven days in a figure legend, and a discussion citing brain-volume findings absent from the results.
10. Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment. Nature Biotechnology, published September 7, 2026. nature.com. Peer reviewed. Read at source. Quoted: "Drugs for aging-related diseases may modulate aging itself, but standard clinical trial designs cannot detect such effects"; "We measure the variance between the clocks and find that all six clocks consistently predicted lower biological age in treated arms"; "However, proteomic clocks alone cannot deconvolute aging- and disease-specific effects"; "Establishing causality remains a challenge, as the statistical trends they capture may be associated with both compensatory and harmful aging processes"; "Full disentanglement of the two signals is not achievable within an IPF cohort and requires validating the drug or its underlying mechanism in healthy volunteers"; "Proteomic aging clocks still carry the caveats common among other omics-based clocks"; and that clock or senescence endpoints "could be prespecified as exploratory biomarkers, with clinical claims requiring conventional clinical endpoints or a qualified biomarker". The trial studied was in idiopathic pulmonary fibrosis and did not involve a GLP-1 drug; it is cited here only for what its authors state about the standing of proteomic clocks as endpoints.
11. Ozempic Shaves Three Years Off People's Biological Age in Study. SingularityHub, August 12, 2025. singularityhub.com. Read at source, cited as evidence of the 2025 coverage cycle rather than for its figures. Reports the HIV-associated lipohypertrophy semaglutide trial from a medRxiv preprint, describing 108 patients with half on weekly semaglutide for 32 weeks and half on placebo. It does not name the clock behind the three-year figure. The 108 figure is the parent trial's randomized total; the epigenetic analysis covered 84 participants with paired samples, per the peer-reviewed paper at footnote 7.