What DO-HEALTH Actually Tested: Omega-3, Biomarker Aging, and Endpoint Limits

Evidence reviewed through: 2026-08-02

Related interpretation: use the trial-endpoint framework to separate endpoint roles and the human-translation guide to bound biomarker claims.

1. Evidence summary

DO-HEALTH was a three-year, multinational, randomized 2x2x2 factorial trial involving 2,157 generally healthy and active adults aged 70 or older. It tested vitamin D3, marine omega-3 and a strength-exercise program, separately and in combination. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

The parent trial evaluated six prespecified primary clinical outcomes and applied a multiplicity threshold of P<.01 with 99% confidence intervals. None of the three interventions produced a statistically significant benefit across that primary outcome family under the prespecified threshold. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

A later analysis studied DNA-methylation measures in 777 consenting Swiss participants who had samples at baseline and year three. That post hoc substudy reported small omega-3-associated changes in three of four measures, with standardized effects of 0.16 to 0.32 and age-clock equivalents of about 2.9 to 3.8 months over three years. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

These are two different evidence layers. The later biomarker result does not convert the parent trial's clinical outcomes into a positive primary efficacy result. Epigenetic-clock changes do not by themselves establish slower clinical aging, improved healthspan or extended lifespan. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

2. The parent DO-HEALTH trial

DO-HEALTH, registered as NCT01745263, was designed to test three accessible interventions in older European adults using clinical and functional outcomes rather than a single biological-age marker. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

The trial enrolled 2,157 participants across Austria, France, Germany, Portugal and Switzerland. Participants were generally healthy and active, which is important when considering whether findings generalize to frailer populations, other health systems or younger adults. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

3. Population and factorial design

Participants were aged 70 or older and assigned across eight groups in a 2x2x2 factorial design. This allowed each intervention to be compared with its control while also evaluating combinations. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

Factorial trials should not be narrated as if one combination arm were the only comparison. Each main-effect estimate pools across the assignments of the other interventions, subject to the trial's analytical assumptions and interaction assessments. (Sources: Bischoff-Ferrari et al., design paper,; Bischoff-Ferrari et al., JAMA primary report,.)

The population and design support inference about the randomized comparisons studied. They do not establish that the same effects would occur in all older adults, people with substantial frailty or populations outside the five countries. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

4. Vitamin D, omega-3 and exercise interventions

The active interventions were vitamin D3 at 2,000 IU per day, marine omega-3 at 1 gram per day, and a 30-minute strength-training program three times weekly. The omega-3 formulation provided 330 mg EPA and 660 mg DHA. The exercise comparator was a control flexibility program. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

These exact exposures define the tested contrasts. The results should not be generalized to other doses, formulations, exercise programs or populations without further evidence. This record describes trial evidence and does not recommend personal use. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

5. The six original primary outcomes

The parent trial specified six primary outcomes: systolic blood pressure, diastolic blood pressure, the Short Physical Performance Battery (SPPB), the Montreal Cognitive Assessment (MoCA), incidence of nonvertebral fractures and incidence of infections. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

This outcome family covered cardiovascular risk, physical function, cognitive function and clinical events. It did not include a DNA-methylation clock as an original primary endpoint. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

6. Prespecified multiplicity threshold

Testing several primary outcomes increases the probability of a chance-positive result. The parent analysis addressed six comparisons by reporting 99% confidence intervals and using P<.01 as the significance threshold. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

This prespecified threshold is essential to interpretation. A nominal P value below .05 would not necessarily meet the trial's primary-outcome standard. Readers should therefore evaluate the whole primary family under the planned rule rather than select an isolated favorable estimate. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

7. Primary clinical results

Under the prespecified multiplicity approach, vitamin D3, omega-3 and the strength-exercise program did not significantly improve the six primary outcomes over three years. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

The result is best described as an overall null primary outcome family, not proof that every possible effect is exactly zero. Confidence intervals, adherence, follow-up duration, population selection and outcome sensitivity still bound what the trial can exclude. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

8. What the parent trial did not show

The parent trial did not show a statistically significant benefit across its six primary clinical outcomes under the prespecified threshold. It did not establish that any of the interventions broadly improves “healthy aging.” (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

It also did not test a validated global endpoint for slower biological aging. Later secondary, exploratory and post hoc analyses can generate additional evidence, but they do not rewrite the prespecified primary result. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

9. The later Swiss DNA-methylation analysis

The later DNAm analysis used samples from 777 consenting Swiss participants with measurements at baseline and year three. It assessed several DNA-methylation algorithms as candidate measures of biological aging. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

Omega-3 assignment was associated with changes in PhenoAge, GrimAge2 and DunedinPACE, while the pattern was not uniform across every clock. An additive analysis reported an effect of all three interventions on PhenoAge, again without consistent change across every measure. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

10. Why the analysis is post hoc

The DNAm analysis was not the parent trial's prespecified primary endpoint family. It examined a selected subset, years after the main clinical report, using multiple clock measures and two sample time points. The source describes it as a post hoc analysis. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

Random assignment remains a strength for intervention comparisons within the substudy. Post hoc status still matters because selection into the sample, choice of measures, multiplicity and analytical decisions can increase uncertainty and limit confirmatory interpretation. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

11. The clocks studied

The analysis evaluated PhenoAge, GrimAge, GrimAge2 and DunedinPACE. These algorithms use DNA-methylation patterns but are constructed for different targets and are not interchangeable. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

No single clock is a gold-standard measure of biological age. Clock behavior can vary by cohort, assay, model design and outcome, and methodological corrections in the broader field reinforce the need for careful interpretation. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Skinner et al., clock limitations/correction,.)

PhenoAge, GrimAge and GrimAge2 express age-related predictions, while DunedinPACE estimates a pace-like quantity. Translating all of them into one statement that participants became “younger” would erase those distinctions. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

12. Reported effect sizes

The omega-3 comparisons produced standardized effect sizes reported between 0.16 and 0.32 across three responsive measures. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

For the age-type clocks, the paper expressed the difference as approximately 2.9 to 3.8 months over the three-year period. For DunedinPACE, the reported interpretation was roughly a 1% difference in pace. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

These are small biomarker-scale estimates. They are not months of additional life, months free of disease or a measured delay in clinical aging. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

13. Approximate 2.9-3.8-month interpretation

The 2.9-3.8-month language is a translation of differences in age-clock outputs over three years. It should be tied to the specific clocks and analysis rather than presented as a direct change in a person's actual age. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

The estimate does not mean that participants lived 2.9-3.8 months longer or avoided 2.9-3.8 months of disease. No such clinical endpoint was measured by this substudy. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

14. Statistical versus clinical significance

Statistical evidence addresses compatibility with a null model under specified analytical assumptions. Clinical significance asks whether an effect changes outcomes that matter to patients, such as function, disease or survival. The two are not equivalent. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

The DNAm analysis supplies a preliminary randomized biomarker signal. Its clinical meaning, persistence and relationship to healthspan or lifespan are unknown. The parent trial's null primary clinical family remains the controlling evidence for the original six outcomes. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

15. Replication and generalizability limits

The substudy included 777 Swiss participants rather than the full multinational cohort of 2,157. Participants required available consented samples at baseline and year three, creating a selected analysis population. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

Only two time points were used, so the analysis cannot establish the trajectory between them or persistence after year three. The reviewed packet identified no independent replication of the DO-HEALTH DNAm intervention result by the cutoff. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

Generalization to different ages, frailty levels, countries, omega-3 preparations or baseline nutritional states requires separate evidence. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

16. What headlines commonly overstate

“Omega-3 slowed biological aging” compresses a post hoc, multi-clock biomarker analysis into a broad clinical conclusion. A more accurate statement is that omega-3 assignment was associated with small changes in three DNAm measures in a selected Swiss subset. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

“DO-HEALTH proved a healthy-aging benefit” also conflicts with the parent trial's null six-outcome primary family. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)

Ancillary analyses of falls, prefrailty, invasive cancer and vertebral fracture have different prespecified, secondary or exploratory status, sometimes with small event counts and multiplicity concerns. They should not be presented as replicated primary efficacy. (Sources: Gagesch et al., prefrailty secondary outcome,; Bischoff-Ferrari et al., exploratory cancer analysis,.)

17. What this establishes

• The parent trial was a 2,157-person, five-country, three-year factorial RCT of vitamin D3, omega-3 and strength exercise. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)
• None of the interventions significantly improved the six primary outcomes under the prespecified P<.01/99% CI approach. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)
• A later post hoc analysis in 777 Swiss participants found small omega-3-associated changes in three DNAm measures. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)
• The biomarker signal is a legitimate hypothesis-generating result that warrants replication and clinical validation. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)

18. What this does not establish

• It does not establish that the original primary clinical trial was positive. (Sources: Bischoff-Ferrari et al., JAMA primary report,; JAMA full text,.)
• It does not establish that omega-3 made participants biologically younger. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)
• It does not establish slower clinical aging, improved healthspan or extended lifespan. (Sources: Bischoff-Ferrari et al., DNAm analysis,; Same DNAm PubMed record.)
• It does not turn secondary or exploratory signals into definitive replication. (Sources: Gagesch et al., prefrailty secondary outcome,; Bischoff-Ferrari et al., exploratory cancer analysis,.)
• It does not provide personal supplement, vitamin or exercise advice. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., design paper,.)

19. Evidence hierarchy table

20. Update triggers

Review this record annually and earlier if an independent randomized study replicates the DNAm signal; longer follow-up links the clock changes to prespecified clinical outcomes; a correction changes the 2025 analysis; NCT01745263 posts material results updates; or a consensus process validates a DNAm measure as a clinical surrogate. (Sources: ClinicalTrials.gov NCT01745263; Bischoff-Ferrari et al., DNAm analysis,.)

The current URL remains aligned with the subject and search intent. The final recommendation is KEEP_EXISTING_URL; no slug, canonical or redirect change is proposed. The malformed non-URL external anchor in the current article is not present in this draft.