Does Treating Insomnia Really Slow Biological Aging? What the CBT-I Trial Actually Shows

There is promising randomized evidence that treating insomnia with cognitive behavioral therapy can affect one measure of biological aging—but the study did not prove that patients became biologically younger or will live longer.

In a 2026 secondary analysis of a randomized trial, older adults assigned to cognitive behavioral therapy for insomnia, or CBT-I, had a significantly more favorable trajectory on DunedinPACE, a DNA-methylation measure designed to estimate the pace of biological aging, than adults who received sleep education.

That result is real. But the simplified claim that “CBT-I slows aging” leaves out several important details.

The CBT-I group’s own DunedinPACE decline was not statistically significant. Much of the separation occurred because DunedinPACE increased in the sleep-education group. Two other epigenetic measures, GrimAge and PCPhenoAge, did not show statistically significant treatment differences. The biomarker analysis also included only 92 people from a larger 291-person randomized trial.

Most importantly, no one in this study was shown to live longer, avoid dementia, prevent cardiovascular disease, or gain additional healthy years of life.

The strongest conclusion supported by the evidence is narrower: CBT-I may influence some biological processes associated with the pace of aging in older adults, but whether that translates into slower clinical aging or longer life remains unknown.

What the insomnia trial actually found

The study, Cognitive behavioural therapy for insomnia and epigenetic ageing: secondary analysis from a randomised controlled trial, was published in The Lancet Healthy Longevity in 2026.

It drew participants from an earlier randomized clinical trial of 291 adults aged 60 or older with insomnia. In that trial, 156 people were assigned to CBT-I and 135 to an active comparison treatment called sleep education therapy, or SET.

For the newer biological-aging analysis, researchers had complete paired DNA-methylation data for 92 participants: 47 in the CBT-I group and 45 in the sleep-education group. Their average age was about 69.5. Blood samples were collected before treatment and again an average of about 24.5 months later.

Researchers tested three measures:

That table exposes an important statistical distinction that is easy to lose in headlines.

Only DunedinPACE showed a statistically significant difference in trajectory between the randomized treatment groups.

It would be inaccurate to say that only one of the three measures “changed.” GrimAge did decline significantly within the CBT-I group. But because GrimAge did not differ significantly between CBT-I and the randomized control group, it does not provide the same level of evidence that CBT-I caused the difference.

The significant result was largely about the control group getting worse

The most important qualification is that the CBT-I participants did not demonstrate a statistically significant DunedinPACE decline from their own baseline.

Their estimated change over roughly two years was:

CBT-I: −0.01

The 95% confidence interval included zero, and the multiple-testing-adjusted p-value was .42.

Meanwhile, the sleep-education group changed:

SET: +0.03

That increase was statistically significant, with an adjusted p-value of .02.

The randomized comparison between those trajectories was significant.

In plain English, the result looks less like:

CBT-I caused people to start aging substantially backward or slower.

and more like:

People receiving CBT-I remained relatively stable on DunedinPACE while the comparison group moved toward a faster estimated pace of aging.

That can still represent a meaningful treatment effect. Preventing deterioration is a legitimate benefit.

But it is different from the mental picture created by saying therapy simply “slowed aging.”

What does the −0.02 DunedinPACE number actually mean?

It does not mean the participants became 0.02 years younger.

It does not mean CBT-I added a particular number of days to their lives.

And it cannot be translated into something like “two biological years reversed.”

DunedinPACE is designed differently from clocks that estimate someone’s biological age in years.

Researchers developed it using repeated measurements of 19 physiological indicators across multiple organ systems in participants from the long-running Dunedin Study. Those changes were combined into an estimate of how rapidly physiological deterioration was occurring and then translated into a DNA-methylation blood test.

The underlying scale is centered near 1.0, representing approximately one biological year of change for each chronological year in the population in which the measure was developed. A higher value represents a faster estimated pace and a lower value a slower one.

In the insomnia trial, however, the reported −0.02 treatment-by-time coefficient describes the difference in the trajectory of DunedinPACE per year of follow-up between treatment groups.

It is not a direct measurement of added lifespan.

Over roughly two years, the model estimated a change of approximately −0.01 in CBT-I and +0.03 in sleep education.

How many “biological years” did CBT-I save?

There is no scientifically justified conversion.

Any claim that these patients became a certain number of years younger, saved a certain number of days of aging, or extended their expected lifespan by a certain percentage would go beyond what the study measured.

The two groups also started at somewhat different DunedinPACE levels

There is another detail worth examining.

Before treatment, average DunedinPACE was:

CBT-I: 0.91

Sleep education: 0.87

The researchers reported no statistically significant baseline difference in the epigenetic measures, so this should not be described as a proven failure of randomization.

Still, the numerical pattern is relevant.

The group that began somewhat higher subsequently moved slightly downward, while the group that began lower moved upward. The groups therefore appear to have converged considerably over time.

That raises a legitimate statistical question about regression toward the mean—the tendency for unusually high or low measurements to move closer to typical values when measured again.

The published mixed-effects model accounted for treatment, time and their interaction and adjusted for factors including age, sex, BMI and education. The paper did not report the more conventional endpoint ANCOVA sensitivity analysis in which follow-up DunedinPACE is explicitly conditioned on baseline DunedinPACE.

That does not establish that regression toward the mean caused the finding. With randomized treatment assignment, baseline differences can arise by chance, especially in relatively small samples.

The appropriate conclusion is narrower:

Because CBT-I started numerically higher and the groups subsequently converged, an analysis explicitly conditioning the endpoint on baseline DunedinPACE would be a useful robustness check. The publicly reported results do not establish whether regression toward the mean contributed materially to the observed interaction.

That is an uncertainty in interpretation, not evidence that the study is invalid.

Why did DunedinPACE change when GrimAge and PCPhenoAge did not?

At first glance, one significant measure out of three looks troubling.

It does weaken the case for a broad claim that CBT-I changes “biological aging” generally.

But the three measures are not interchangeable tests of exactly the same thing.

DunedinPACE was built specifically to estimate the ongoing rate of aging-related physiological deterioration. GrimAge and PhenoAge were developed using different methods that place greater emphasis on accumulated mortality and health-risk information.

That distinction matters when measuring an intervention over only a few years.

There is also an important precedent from another randomized intervention.

In the CALERIE trial, researchers studied two years of caloric restriction in healthy adults. That intervention produced evidence of slower aging on DunedinPACE but did not significantly change PhenoAge or GrimAge. The authors argued that pace-of-aging measures could be more sensitive to short-term intervention than clocks representing accumulated biological aging.

That does not prove the insomnia result is correct.

It does show why saying “two clocks contradicted DunedinPACE” would be too simplistic.

A better interpretation is:

The evidence is narrow rather than internally impossible. One measure designed to detect changes in the pace of physiological aging responded; two differently constructed aging measures did not show significant randomized treatment effects.

The 92-person biomarker sample is an important limitation

Calling this simply “a randomized trial of 92 people” misses part of the story.

The original randomization involved 291 people.

The epigenetic analysis involved a much smaller subset with available paired blood samples. Because methylation assays were costly, the investigators selected a limited number of eligible paired samples for analysis.

The final study contained 47 CBT-I participants and 45 controls.

The authors themselves acknowledge that analyzing this subset could introduce selection bias, reduce precision and weaken the balance created by the original randomization.

There is evidence that the smaller group was not perfectly balanced.

For example, women made up 45% of the CBT-I group but 73% of the sleep-education group. Sex was included as a statistical adjustment. The subset was also predominantly White.

This does not erase the randomized origin of the trial, but it should lower confidence relative to a biomarker analysis performed in nearly every randomized participant.

The researchers’ post-hoc power analysis estimated 73% power to detect a treatment-by-time interaction in at least one of the three clocks after multiple-testing correction.

Replication in a larger, independently randomized sample would materially strengthen the evidence.

Insomnia remission may matter—but that result is exploratory

Full insomnia remission occurred in:

34% of CBT-I participants, 16 of 47

versus

13% of sleep-education participants, 6 of 45.

Researchers also divided participants according to both their assigned treatment and whether their insomnia remitted.

The clearest worsening occurred among sleep-education participants whose insomnia did not remit: their estimated DunedinPACE increased by about 0.04 over two years.

Participants whose insomnia remitted tended to show more favorable trajectories.

This is compatible with an intriguing possibility:

Successfully resolving insomnia itself might influence aging-related biology, regardless of exactly how remission is achieved.

But the study cannot establish that.

Only six people in the sleep-education group achieved full remission, and the remission analysis was exploratory rather than a sufficiently powered causal test.

The finding should therefore be treated as a hypothesis for future research, not proof that insomnia remission directly slows biological aging.

Does this result have any support beyond one epigenetic clock?

Yes, but it is important not to double-count the evidence.

A separate 2023 analysis of the same parent trial examined p16INK4a, a gene-expression marker associated with cellular senescence.

That analysis included 231 participants. p16INK4a expression increased over two years in the sleep-education group while remaining comparatively stable in the CBT-I group. Participants receiving CBT-I who achieved sustained insomnia remission showed a significant decline from baseline.

That gives the newer DunedinPACE result some biological coherence.

But it is not an independent replication.

The studies arose from the same randomized trial and the same research program. The p16 result should therefore be described as convergent evidence from another aging-related biomarker in the same experiment, not as a completely separate confirmation.

Was the biological-aging analysis invented after researchers saw the data?

There is no evidence supporting that allegation.

In fact, public records show that the aging hypothesis predates the 2026 publication by years.

UCLA researcher Judith Carroll received an NIA grant beginning in 2016 titled A Randomized Trial for Sleep Disturbances to Reverse Cellular Aging. The project’s stated exploratory aims included studying age-related epigenetic methylation patterns in DNA.

The exact clocks used in the 2026 analysis could not have been selected when the original insomnia trial began in 2012 because the modern versions of PhenoAge, GrimAge and DunedinPACE were developed later.

DunedinPACE itself was published in 2022.

The broader hypothesis that treating sleep disturbance might alter biological aging therefore clearly predates the current finding. What is not publicly established is exactly when the researchers fixed the final three-clock statistical analysis plan.

There is no evidence here of fabrication, misconduct or deliberate cherry-picking.

The DunedinPACE finding actually appeared before 2026

The complete peer-reviewed analysis is new, but its central DunedinPACE finding was publicly presented earlier.

A 2023 conference abstract by Judith Carroll reported an analysis of 49 CBT-I and 45 sleep-education participants and described a significant treatment-by-time DunedinPACE interaction.

The 2026 Lancet Healthy Longevity publication substantially improves on that preliminary disclosure by providing the peer-reviewed methods, final analytical sample, multiple epigenetic measures, multiple-testing correction, sensitivity analyses and detailed limitations.

So the accurate chronology is not that researchers suddenly discovered the effect in 2026. The full peer-reviewed evidence arrived in 2026 after the core DunedinPACE result had previously been reported at a scientific meeting.

Does this prove CBT-I will make people live longer?

No.

This is the largest gap between the biomarker result and the longevity headlines it can generate.

DunedinPACE has meaningful validation. Its developers found high test-retest reliability and associations with morbidity, disability and mortality in outside cohorts.

But an aging biomarker being associated with future health outcomes does not automatically make it a proven surrogate endpoint for those outcomes.

To establish that, researchers ultimately need evidence that when an intervention changes DunedinPACE, the intervention-induced change reliably predicts meaningful changes in things such as disease, disability or survival.

That has not been demonstrated by this insomnia trial.

The study did not show that CBT-I recipients:

• lived longer;
• developed dementia later;
• had fewer heart attacks or strokes;
• developed less cancer;
• became less frail;
• avoided disability;
• gained additional healthy years of life.

The CALERIE researchers made essentially the same distinction when caloric restriction changed DunedinPACE: biomarker modification is promising geroscience evidence, but long-term disease and mortality outcomes are needed to establish clinical anti-aging effects.

Therefore, the phrase “slows biological aging” is defensible only when it is immediately qualified as “as estimated by DunedinPACE.”

It should not be treated as synonymous with “extends human lifespan.”

Does treating insomnia reduce dementia risk?

This study does not answer that question.

No dementia endpoint was tested.

The same is true for cardiovascular disease and cancer.

Associations between poor sleep, aging biology and chronic disease provide plausible reasons to study those outcomes, but plausibility is not clinical proof.

The parent randomized trial did demonstrate another important clinical benefit: adults assigned to CBT-I were substantially less likely to develop incident or recurrent major depression during follow-up.

Major depression occurred in 12.2% of CBT-I participants versus 25.9% of the sleep-education group, corresponding to a hazard ratio of 0.51.

That is a useful reminder that CBT-I does not need a speculative longevity effect to have meaningful health value.

Would sleeping pills have the same biological-aging effect?

We do not know.

This trial did not randomize participants to sleeping medication.

Researchers performed sensitivity analyses accounting for sleep-medication use, and the main epigenetic result remained similar. But statistically controlling for medication use is not the same thing as conducting a randomized medication trial.

Nothing in this study establishes that sedative or hypnotic drugs reproduce the DunedinPACE result.

Current sleep-medicine guidance also treats CBT-I as a core evidence-based treatment rather than simply an interchangeable route to sedation. The American Academy of Sleep Medicine gives multicomponent CBT-I a strong recommendation for chronic insomnia in adults.

Whether different successful insomnia treatments produce similar effects on aging biomarkers remains an open research question.

Would CBT-I slow biological aging in younger adults?

Again, unknown.

Everyone in this analysis was at least 60 years old, with an average age close to 70. The study was also conducted at a single site in Los Angeles and the epigenetic subset was predominantly White.

Aging biology, immune responses, sleep architecture and the health consequences of chronic insomnia can differ by age.

The results therefore cannot simply be assumed to apply to someone in their 20s, 30s or 40s.

That requires replication.

What exactly is CBT-I?

Cognitive behavioral therapy for insomnia is a structured treatment aimed at changing the behaviors, sleep patterns and thought processes that perpetuate chronic insomnia.

Multicomponent CBT-I commonly combines cognitive therapy with sleep-regulation education, stimulus control and sleep restriction therapy, often alongside relaxation and sleep-hygiene strategies. Treatment commonly lasts four to eight sessions.

Importantly, sleep hygiene by itself is not equivalent to CBT-I. The American Academy of Sleep Medicine specifically recommends multicomponent CBT-I and suggests against using sleep hygiene alone as the sole treatment for chronic insomnia.

The academy’s evidence review identified dozens of randomized controlled trials and found clinically meaningful benefits for outcomes including insomnia remission and treatment response.

So the uncertainty exposed by the aging study is not whether CBT-I is a legitimate insomnia treatment.

It is whether one additional potential benefit—slowing aspects of biological aging—has now been established strongly enough to make a clinical longevity claim.

It has not.

So, does treating insomnia actually slow biological aging?

The evidence supports three different levels of conclusion.

Verified: In this 92-person secondary analysis of a randomized trial, CBT-I produced a significantly more favorable DunedinPACE trajectory than sleep education after adjustment for multiple comparisons.

Plausible but not yet established: Successfully treating or remitting chronic insomnia may influence biological processes involved in aging. The DunedinPACE result, an earlier p16INK4a finding from the same trial and analogous DunedinPACE results from another randomized intervention make that possibility scientifically credible.

Not proven: CBT-I makes humans biologically younger, prevents dementia or cardiovascular disease by slowing aging, adds healthy years to life, or increases lifespan.

There is also an important statistical nuance behind the significant result: the CBT-I group’s own DunedinPACE change was not significant, while the sleep-education group’s pace increased. And because CBT-I participants started with a numerically higher average DunedinPACE score before the groups converged, regression toward the mean remains a reasonable sensitivity question that cannot be resolved from the published aggregate data alone.

That does not make the study meaningless.

Quite the opposite.

A behavioral insomnia treatment producing a randomized signal on a credible pace-of-aging biomarker is interesting precisely because it creates a testable hypothesis: could effectively treating chronic insomnia alter aging biology enough to eventually change disease, disability or lifespan?

The 2026 trial provides a reason to ask that question seriously.

It does not yet provide the final answer.

References and Further Reading

Primary insomnia and aging research

• Cognitive behavioural therapy for insomnia and epigenetic ageing: secondary analysis from a randomised controlled trial — The Lancet Healthy Longevity — The primary 2026 study reporting DunedinPACE, GrimAge and PCPhenoAge outcomes after CBT-I versus sleep education.

• Prevention of Incident and Recurrent Major Depression in Older Adults With Insomnia: A Randomized Clinical Trial — JAMA Psychiatry — The original 291-person randomized clinical trial from which the epigenetic subset was drawn.

• Remission of insomnia in older adults treated with CBT-I reduces p16INK4a gene expression — GeroScience — Earlier aging-biomarker analysis from the same trial involving the cellular-senescence-associated marker p16INK4a.

• Cognitive Behavioral Treatment of Insomnia Slows the Pace of Biological Aging: Results From an RCT in Older Adults — Innovation in Aging — The 2023 conference abstract that publicly reported the core DunedinPACE finding before publication of the full 2026 paper.

Understanding DunedinPACE and epigenetic aging

• DunedinPACE, a DNA methylation biomarker of the pace of aging — eLife — The primary development and validation paper explaining what DunedinPACE measures and how its scale should be interpreted.

• Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial — Nature Aging — Independent randomized-intervention evidence in which DunedinPACE responded while GrimAge and PhenoAge did not, useful for understanding why different clocks can produce different findings.

CBT-I clinical evidence

• Behavioral and psychological treatments for chronic insomnia disorder in adults: AASM clinical practice guideline — American Academy of Sleep Medicine guideline giving multicomponent CBT-I a strong recommendation for chronic insomnia.

Study provenance

• Judith Carroll — UCLA Research Profile — Documents the NIA-funded 2016 project A Randomized Trial for Sleep Disturbances to Reverse Cellular Aging.

• Aging and Wellness — UCLA Cousins Center for Psychoneuroimmunology — Describes the project’s prespecified interest in cellular-aging markers and exploratory analysis of age-related epigenetic methylation.

Editorial currency note: This article reflects the peer-reviewed evidence and publicly available research record through September 9, 2026. The longevity interpretation could change if larger replications, individual-level reanalyses, longer clinical follow-up or new randomized trials connect insomnia-treatment-induced biomarker changes to disease, disability or mortality.