There is growing evidence that vitamin D supplementation is associated with slower “biologic” aging—the age reflected by molecular biomarkers rather than the number of birthdays on the calendar. In this analysis I’ll use data from the Precision Aging Network (PAN) to examine three established DNA‑methylation–based “epigenetic clocks”—Horvath 2013, AltumAge, and DunedinPACE.
Background/Tools
Epigenetic clocks: DNA‑methylation signatures that act as molecular clocks. We use three complementary models:
Horvath 2013 — 353 CpGs, pan‑tissue; returns an epigenetic age in years. Age acceleration = EpigenAge − ChronAge.
AltumAge — ≈21 000 CpGs, deep‑learning model trained on >140 data sets; interpreted like Horvath.
DunedinPACE — 38 CpGs; outputs the pace of aging (1.00 = average, >1 = faster, <1 = slower).
Why three clocks? Horvath and AltumAge give a snapshot of cumulative aging, while DunedinPACE acts like a speedometer, telling us how rapidly aging processes are unfolding now. Using all three offers both level and rate perspectives on biological aging.
PAN dataset: Community‑dwelling older adults with phenotypic and lifestyle data including supplement use.
Quick Interpretation Cheat Sheet
Clock
Output
“Higher” value implies
Horvath 2013
years (epigenetic age)
Biological age older than calendar age (accelerated aging)
AltumAge
years (epigenetic age)
Same as Horvath 2013 but with higher precision
DunedinPACE
ratio (bio‑years per chrono‑year)
Aging faster than average (> 1.0)
Key R packages:tidyverse for data wrangling, table1 for publication‑ready descriptive tables, ggplot2 and ggpubr for displaying data.
Literature Review
1. Vitamin D & Telomere Preservation in the VITAL Trial
Source: Medical News Today, May 28 2025
Vitamin D’s anti-aging promise first caught popular attention in a secondary analysis of Harvard’s large VITAL randomized trial, recently profiled by Medical News Today.
Design & cohort. VITAL enrolled 25,871 generally healthy U.S. adults (women ≥ 55 y; men ≥ 50 y) who were randomized to 2,000 IU/day vitamin D3, 1 g/day marine omega-3s, both, or placebo for roughly five years. In a nested lab sub-study (~1,000 participants; >2,500 leukocyte samples), investigators measured telomere length at baseline, 2 years and 4 years .
Key finding. Participants receiving vitamin D showed minimal telomere shortening, whereas placebo groups exhibited the expected attrition. Omega-3s alone had no significant effect. Translating the telomere signal, the authors estimate vitamin D may confer the biological equivalent of ~3 years less aging over four years .
Sub-group nuances. Benefits were strongest in non-White participants and in those not taking statins; BMI did not modify the effect .
Limitations. The sub-study was post-hoc, older and mostly White, with substantial missing samples by year 4, so findings are hypothesis-generating rather than definitive .
Relevance to our project. Although telomere length is a cruder marker than DNA-methylation clocks, the VITAL data reinforce the plausibility that vitamin D can slow cellular aging—aligning with our epigenetic-clock hypothesis and justifying its inclusion as a mechanistic link.
2. DO-HEALTH Bio-Age Trial: Epigenetic-Clock Outcomes after 3 Years
Source: Bischoff-Ferrari et al. Nature Aging (2025)
DO-HEALTH is the largest factorial RCT to date testing daily vitamin D (2,000 IU), omega-3s (1 g), and a home exercise program (SHEP) in 2,157 European adults aged ≥ 70. A Swiss Bio-Age sub-study (n = 777) assayed four next-generation DNA-methylation clocks at baseline and 3 years .
Headline results.
Omega-3 alone significantly slowed PhenoAge, GrimAge2 and DunedinPACE.
Additive benefit: combining omega-3 with vitamin D and exercise yielded a further reduction in PhenoAge (–0.24 to –0.32 SD ≈ 2.9–3.8 months younger) .
Vitamin D alone did not reach significance on any clock but amplified omega-3 when combined.
Clinical context. The same three-way regimen previously lowered prefrailty (–39 %) and invasive cancer (–61 %) in the parent trial, suggesting molecular and clinical signals converge .
Limitations. Only two time points (baseline, 3 y) raise measurement-error concerns; Swiss subset may not generalize to less-healthy or non-European elders; clock changes were modest in magnitude .
Implications for our study. DO-HEALTH’s epigenetic evidence supports our choice of Horvath-style clocks and underscores the potential synergy between vitamin D and co-interventions. It also cautions that vitamin D effects may surface only in concert with other lifestyle factors—an angle we can test with interaction terms in our PAN dataset.
Research Question
Is self‑reported vitamin D supplementation associated with lower (“younger”) values in any three epigenetic clock measures after accounting for chronological age?
Creating Analysis Dataset
Below we reproduce the setup and data‑loading chunks exactly as in the original file; the only goal is to annotate what they are doing
Data Preparation
The code chunk below reads the PAN dataset (saved as keys2025data.csv) and drops any participants with missing information on the stratification variable vitd_supplement. This mimics the filtering step in the diabetes tutorial where rows with missing outcome data were excluded.
Code
# Read PAN dataset and create analysis subsetraw_pan <-read_csv("keys2025data.csv")# Keep only variables required for analysis, renaming on the flypan <- raw_pan %>%select( hml_id, age,vitamin_d = vitd_supplement, # renamed here dose_mg, horvath2013, altumage, dunedinpace, sex,education = edu_yrs_hml, # renamed here bmi, health_medical_cancer ) %>%drop_na(age, vitamin_d, horvath2013, altumage, dunedinpace) # updated names used herepan$cancer<-factor(pan$health_medical_cancer,levels =c(0, 1),labels =c("no", "yes"))# Save subset for reproducibilitywrite_csv(pan, "pan_subset.csv")# Convert from wide to long for optional analyses (wide → long)pan_long <- pan %>%pivot_longer(cols =c(horvath2013, altumage, dunedinpace),names_to ="clock",values_to ="epigenetic_age" )### this code will enable prettier output using the rms packagepand <-datadist(pan)options(datadist='pand')
Table 1
Table 1 describes the study cohort stratified by vitamin‑D supplementation status (“no” vitamin D vs. “yes” vitamin D) and for the overall sample. What to look for: Compare the means/medians in each column. If vitamin‑D users systematically differ (older age, higher educational attainment, lower BMI) those variables could confound any relationship between vitamin D and the epigenetic clocks. Why it matters: The more similar the two columns are, the more confident we can be that the analyses isolate the effect of vitamin D rather than underlying demographic or health differences.
Table 1. PAN Demographics and Summary Statistics
Code
# --- Create the descriptive table ------------------------------------------tab <-table1(~ horvath2013 + altumage + dunedinpace + age + sex + education + bmi + cancer | vitamin_d,data = pan,overall ="Overall",render.continuous ="Mean(SD)",render.missing =NULL,digits =3,res =300)tab
no (N= 369)
yes (N= 59)
Overall (N= 428)
horvath2013
64.3(7.35)
65.7(6.66)
64.5(7.27)
altumage
60.9(5.81)
62.1(4.75)
61.1(5.69)
dunedinpace
1.01(0.110)
1.04(0.116)
1.01(0.111)
age
64.2(7.70)
66.0(6.71)
64.5(7.59)
sex
Female
238 (64.5%)
54 (91.5%)
292 (68.2%)
Male
131 (35.5%)
5 (8.5%)
136 (31.8%)
education
16.1(2.30)
15.9(2.15)
16.0(2.27)
bmi
26.9(5.47)
29.1(7.20)
27.2(5.78)
cancer
no
338 (91.6%)
53 (89.8%)
391 (91.4%)
yes
31 (8.4%)
6 (10.2%)
37 (8.6%)
Vitamin D with VITAL
As a comparison, we show the vital demographic data, labeled, Table 2. VITAL Study Demographics and Summary Statistics.
and another study (DO-HEALTH) showed that 3-year supplementation showed a slight improvement in duninPACE, but the effect was greater in fish oil supplementation than vitamin D (which is at odds with the VITAL study).
Code
# Read in .csv ----------vital_dat <-read.csv("vital_data_2019.csv")# Create labeled levels for categorical variables ---------------vital_dat$vitamin_d <-factor(vital_dat$vitdactive,levels =c(1, 0),labels =c("Yes", "No"))vital_dat$fish_oil <-factor(vital_dat$fishoilactive,levels =c(1, 0),labels =c("Yes", "No"))vital_dat$sex_factor <-factor(vital_dat$sex,labels =c("Male", "Female"))vital_dat$race_factor <-factor(vital_dat$race,labels =c("Non-Hispanic White", "Black", "Hispanic", "Asian", "Native American or Alaskan", "Others/Unknown"))vital_dat$malcancer_factor <-factor(vital_dat$malca,levels =c(1, 0),labels =c("Malignant Cancer", "No Malignant Cancer"))vital_dat$majorcvd_factor <-factor(vital_dat$majorcvd,levels =c(1, 0),labels =c("Has Major CVD", "No Major CVD"))# Create nice labels for variables used in table -----------label(vital_dat$ageyr) <-"Age"label(vital_dat$sex_factor) <-"Sex"label(vital_dat$race_factor) <-"Race"label(vital_dat$bmi) <-"BMI"label(vital_dat$vitamin_d) <-"Randomization to Active Vitamin D"label(vital_dat$fish_oil) <-"Randomization to Active N-3 Fatty Acids"label(vital_dat$malcancer_factor) <-"Cancer"
equal male:female (50:50) while PAN had more females
DO-Health
older age (75 vs 64)
fewer females (60% vs 68%)
lower BMI (25.7 vs 27.3)
lower education (13.5yrs vs 16yrs)
similar cancer percentages
1/3 participants vitamin D-deficient at baseline
Determining Correlation between Clocks in PAN (epigenetic vs chronologic age)
Below we visualize how each epigenetic clock (Horvath 2013, AltumAge, and DunedinPACE) relates to participants’ chronological age.
Each scatter plot is faceted by self‑reported Vitamin D supplement use (“Yes” / “No”), and includes the linear regression equation, the \(R^2\) statistic, and the p‑value for the age–clock association.
PhenoAge(r = 0.60), GrimAge (r = 0.92), GrimAge2 (r = 0.71) were similar
PAN vs DO-Health (3rd gen clock):
PAN DunedinPACE correlation was r = 0.19
DO-Health DunedinPACE correlation was r = 0.19
Statistical Analysis
Methods
We use linear regression to examine the both the relationship between biologic and chronologic age as well as evaluate whether or not there is evidence that vitamin D supplementation slows down biologic aging. We compare these results to two large studies that have come out recently. We don’t see a relationship between Horvath biologic aging and vitamin d supplementation, either as a univariate or multivariable analysis adjusting for other variables (two plots show this, from the RMS modeling).
Results
Output shows that there is a statistically linear relationship between biologic and chronologic aging, with the \(R^2\) of 0.585 showing good predictive capacity (for human cohorts an \(R^2\) of 0.2 is typical for prediction). Interestingly, the Horvath epigenetic clock is higher than chronologic age (by a little over 1 year on average), basically showing the PAN population shows slightly higher biologic age compared to their chronologic age.
Code
multivariable.mod <-ols(horvath2013 ~ sex + education + bmi + cancer + vitamin_d, data=pan)plot(anova(multivariable.mod), res=300, margin="P")
Code
multivariable.mod <-ols( horvath2013 ~ sex + education + bmi + cancer + vitamin_d,data = pan)# Panelled partial-effect plots with a custom y-axis titleplot(Predict(multivariable.mod),res =300,ylab ="Horvath Clock"# <- new Y-axis label)
Phenotype Significance Values with Horvath
Sex is statistically associated with epigenetic age using Horvath clock
Vitamin D wasn’t significant in lowering epigenetic age
Specific Phenotypes with Horvath
Male sex is associated with higher epigenetic age compared to females
Vitamin D supplementation interestingly shows increase in epigenetic age, though again, not statistically significant
Having cancer increases epigenetic age, though its not statistically significant
Dunedin Pace Analysis
Code
multivariable.dunedin <-ols( dunedinpace ~ sex + education + bmi + cancer + vitamin_d,data = pan)plot(anova(multivariable.dunedin), res=300, margin="P")
Code
# Fit model for DunedinPACEmultivariable.dunedin <-ols( dunedinpace ~ sex + education + bmi + cancer + vitamin_d,data = pan)# Panelled partial-effect plots with a custom y-axis titleplot(Predict(multivariable.dunedin),res =300,ylab ="DunedinPACE Clock")
Phenotype Significance Values with DunedinPACE
There is a statistically significant association between epigenetic age and BMI, education, and sex
Vitamin D supplementation didn’t seem to impact epigenetic age
Specific Phenotypes with DunedinPACE
Higher BMI and male sex are associated with higher epigenetic age
Higher education is associated with lower epigenetic age
Both vitamin D supplementation and cancer increase epigenetic age, though not statistically associated
Discussion
Degree of correlation between epigenetic aging using Horvath clock, but not Dundenin PACE Clock
Very similar to what was seen in the DO-HEALTH Study
Precision Aging Network data did not confirm epigenetic/biologic age differences with vitamin D supplementation
Unlike both the VITAL and DO-HEALTH Studies, PAN is not a randomized controlled trial - it is a cross-sectional study
Both VITAL and DO-HEALTH discussed epigenetic aging after 3 years of treatment/supplementation – we don’t know how long PAN participants were taking vitamin D
PAN is a longitudinal study, future direction of research is to include duration of supplementation
