Sample Characteristics
The analytic sample is comprised of 2,311 adults who ranged between 50 to 98 in chronological age. As shown in Table 1, the epigenetic GrimAge score had a mean of 69.4 (SD = 8.7). The average epigenetic age for the sample was 2.2 years younger than the average chronological, or calendar age. The ADHD-PGS was standardized to a mean of 0 and standard deviation of 1, with scores ranging from − 3.6 to 3.7. More than half of the sample was women, and on average individuals reported experiencing less than 1 depressive symptom in the last week, and mean lifetime pack-years among ever smokers of 27.5. Additionally, at least one-third of the sample fell into the obese range when defined as a BMI of 30 or greater. About 46.7% respondents completed at least 16 years of school, or the equivalent of a college degree, and the mean income was over $80,000.
Table 1
Sample Characteristics for Participants from the U.S. Health and Retirement Study
Variable
|
Mean (Std. Dev.) or
n (%)
|
Epigenetic age in 2016: GrimAge (years)
|
69.4
|
(8.7)
|
Chronological age in 2016 (years)
|
71.6
|
(9.6)
|
ADHD Polygenic score (PGS)
|
0.0
|
(1.0)
|
Gender
|
|
|
Male
|
996
|
(43.6%)
|
Female
|
1289
|
(56.4%)
|
Depressive Symptoms
|
1.0
|
(1.6)
|
Smoking (lifetime pack years)
|
27.5
|
(26.2)
|
Body Mass Index (BMI) (kg/m2)
|
27.8
|
(5.6)
|
Cognition score
|
17.7
|
(3.6)
|
Education attained (years)
|
13.6
|
(2.4)
|
Income ($)
|
82,358.4
|
(100592.4)
|
Note. Range for depressive symptoms was 0 to 8, smoking pack-years was 0 to 156, BMI was 16.5 to 57.4, cognition score was 0 to 27, years of education was 0 to 17, income was 0 to 2.1 million.
ADHD-PGS and Epigenetic Aging
As shown in Table 2, we found support for hypothesis 1. The ADHD-PGS, with each 1-SD increase, was significantly associated with almost half a year older GrimAge (b = 0.49, p < .0001). Each additional chronological year of age was associated with a 0.76 year older GrimAge (b = 0.76, p < .0001). Men had a 3.43 year higher GrimAge (b = 3.43, p < .0001) compared to women. All variables accounted for 74.85% of the variation in GrimAge, with 4.82% of the variation accounted for by gender, ADHD-PGS and ancestral principal components. The ADHD-PGS alone accounted for 0.29% of the variation in epigenetic age. Bivariate correlations among variables are provided in the supplementary material (Table S1).
Table 2
Unstandardized Coefficients from Regression Models for the Direct Association between the ADHD Polygenic Score (ADHD-PGS) and Epigenetic Aging, as Indexed by the DNA Methylation Based Indicator of GrimAge
Parameter
|
beta coefficient
|
Std. Err.
|
p-Value
|
Intercept
|
74.79
|
0.30
|
< .0001
|
Age
|
0.76
|
0.01
|
< .0001
|
Gender: male
|
3.43
|
0.18
|
< .0001
|
ADHD-PGS
|
0.49
|
0.09
|
< .0001
|
Model R2
|
|
|
74.85%
|
Notes. Models are adjusted for ancestral principal components as covariates to adjust for population substructure.
Because a prior study [53] found no association between the ADHD-PGS and the first generation Horvath epigenetic clock often used to reflect biological aging, we tested the association in our sample to compare potential differences in performance of the ADHD-PGS in another clock. We also found no association between the ADHD-PGS and Horvath epigenetic clock (p = 0.56; details provided in the supplementary material, Table S2).
Indirect Effects Through Adverse Behavioral and Sociodemographic Mediators
To test hypothesis 2, the model predicting GrimAge was constructed with all mediators in the model, as depicted in Fig. 2, showing good fit (χ2(42) = 5159.42, p < .0001, RMSEA = 0.036, CFI = 0.998, TLI = 0.976). In this model, the ADHD-PGS directly associated with GrimAge, with each additional SD of the PGS associating with almost a quarter year older GrimAge (b = 0.22, p < .01). Smoking, depressive symptoms, and education mediated 51.7% of the effect of the ADHD-PGS on GrimAge as described below.
Indirect paths from the ADHD-PGS to GrimAge, shown in Table 3, were significant through smoking (b = 0.13, p < .001), depressive symptoms (b = 0.02, p < .011), and education (b = 0.06, p < .0001). Indirect paths were not through BMI (b = 0.009, p = .14), cognition (b = 0.01, p = .08), or income (b = 0.01, p = .07). Thus, mediational hypothesis 2 was partially supported.
All variables in the model accounted for 78.8% of the variation in GrimAge. This included direct effects, depicted in Fig. 2, in which the ADHD-PGS was positively associated with smoking pack years (b = 1.62, p < .001), which in turn, predicted older epigenetic age (b = 0.08, p < .0001). Similarly, higher ADHD-PGS was positively associated with depressive symptoms (b = 0.11, p < .001), which in turn, predicted older epigenetic age (b = 0.21, p < .0001). The ADHD-PGS was not related to BMI (b = 0.19, p = .09), but BMI did significantly predict older epigenetic age (b = 0.05, p < .001). Higher ADHD-PGS significantly and inversely associated with more educational attainment (b=-0.27, p < .0001) which, in turn, predicted younger GrimAge (b=-0.20, p < .0001). Higher ADHD-PGS significantly and inversely associated with higher income (b=-0.05, p < .01), which, in turn, significantly predicted GrimAge (b=-0.23, p < .05). Lastly, higher ADHD-PGS significantly and inversely associated with higher cognitive scores (b=-0.31, p < .0001), but cognitive scores did not predict GrimAge (b=-0.05, p = .07). Additional paths added for model fit and for covariates are provided in the supplementary material (Table S3).
Table 3
Unstandardized Direct and Indirect Effects for the Single Mediation Model
Pathway
|
Path a
|
Path b
|
Path a*b:
Indirect Effects
|
Path c:
Direct Effect
|
|
beta (SE)
|
beta (SE)
|
beta (95% CI)
|
beta (SE)
|
ADHD-PGS → GrimAge
|
--
|
--
|
--
|
0.222 (0.080)**
|
via Smoking
|
1.622 (0.482)***
|
0.077 (0.003)***
|
0.125
(0.064, 0.187) ***
|
|
via Depressive Symptoms
|
0.109 (0.033)**
|
0.214 (0.052)***
|
0.023
(0.009, 0.041)*
|
|
via Body Mass Index
|
0.194 (0.115)#
|
0.046 (0.015)***
|
0.009
(0.000, 0.020)
|
|
via Education
|
-0.269 (0.048)***
|
-0.204 (0.038)***
|
0.055
(0.032, 0.081)***
|
|
via Cognition
|
-0.306 (0.073)***
|
-0.047 (0.024)#
|
0.014
(0.001, 0.030)#
|
|
via Income
|
-0.052 (0.019)**
|
-0.231 (0.095)*
|
0.012
(0.002, 0.025)#
|
|
Total Indirect Effect
|
|
|
0.238
(0.165, 0.315)***
|
|
Notes. SE = standard error. 95% CI = 95% Confidence Interval, estimated from a bias-corrected bootstrap procedure with 5000 draws. # p < .10; * p < .05; ** p < .01; *** p < .001. Total effect on GrimAge: beta = 0.460, SE = 0.090, p < .0001.
Multi-Mediation Effects Through Education and Other Mediators
To test hypothesis 3, multi-mediational models were constructed, as shown in Fig. 3. The model fit was good (χ2(42) = 5159.42, p < .0001, RMSEA = 0.030, CFI = 0.998, TLI = 0.983) with similar or slightly better fit than the mediation model as indicated by the RMSEA and TLI. In this model, as shown in Table 4, the association between ADHD-PGS and GrimAge was the same as for the single mediation model (b = 0.22, p < .01) as was the percentage of effect being mediated, 51.6%. The effect of the ADHD-PGS through education to GrimAge was secondarily mediated through smoking (b = 0.04, p < .0001), depressive symptoms (b = 0.005, p < .01), BMI (b = 0.003, p < .05), and income (b = 0.008, p < .05), but not via cognition (b = 0.006, p = .09). Thus, hypothesis 3 was partially supported.
The ADHD-PGS, mediators, and covariates explained 80.5% of the variation in GrimAge. Direct effects from the ADHD-PGS include that it inversely predicted education (b=-0.27, p < .0001) and cognition (b=-0.18, p < .01), and positively predicted depressive symptoms (b = 0.09, p < .05) and smoking (b = 1.13, p < .05), but did not predict BMI (b = 0.13, p = .26), or income (b=-0.02, p = .31). Significantly and directly related to older GrimAge were smoking (b = 0.08, p < .0001), depressive symptoms (b = 0.21, p < .0001), BMI (b = 0.05, p < .001), less education (b=-.20, p < .0001), less income (b=-0.23, p < .05), but not cognition (b=-0.05, p = .07). Notably, there was a significant direct effect from education to GrimAge (b=-0.20, p < .0001); with each additional year of education attained, there was a fifth of a year decrease in GrimAge.
Overall, the total indirect effect of the ADHD-PGS on GrimAge was (b = 0.24, p < .0001) proportionally similar, but in opposite direction to the total indirect effect of education on GrimAge (b=-0.22, p < .0001). As shown in Table 4, the specific indirect effect from the ADHD-PGS to GrimAge was significant via education (b = 0.06, p < .0001), via smoking (b = 0.09, p < .05), and via depressive symptoms (b=-0.02, p < .05), whereas no significant were paths via BMI (b = 0.006, p = .30), via cognition (b = 0.008, p = .15), or via income (b = 0.004, p = .39). Specific indirect effects from education to GrimAge were significant via smoking (b=-0.13, p < .0001), via depressive symptoms (b=-0.02, p < .01), via BMI (b=-0.01, p < .01), and via income (b=-0.03, p < .05), but not via cognition (b=-0.02, p = .07) was not. Additional paths added for model fit and for covariates are provided in the supplementary material (Table S4).
Table 4
Unstandardized Direct and Indirect Effects for the Multi-Mediation Model
|
Path a1:
ADHD-PGS → Mediator
|
Path a2:
Education → Mediator
|
Path b:
Mediator → GrimAge
|
Path a1*b:
Indirect Effects:
ADHD-PGS → Mediator → GrimAge
|
Path a2*b:
Indirect Effects:
Education → Mediator → GrimAge
|
Path a1-a2-b:
Indirect Effects:
ADHD-PGS → Education → Mediator → GrimAge
|
Path c:
Direct Effects:
on GrimAge
|
Predictor or Mediator
|
beta (SE)
|
beta (SE)
|
beta (SE)
|
beta
(95% CI)
|
beta
(95% CI)
|
beta
(95% CI)
|
beta (SE)
|
ADHD-PGS
|
|
|
|
|
|
|
0.222 (0.080)**
|
Smoking
|
1.134 (0.477)*
|
-1.745 (0.202)***
|
0.077 (0.003)***
|
0.087 (0.026, 0.150)*
|
-0.134 (-0.163, -0.106)**
|
0.036
(0.024, 0.050)***
|
|
Depressive Symptoms
|
0.086 (0.033)*
|
-0.087 (0.014)***
|
0.214 (0.052)***
|
0.018 (0.005, 0.035)*
|
-0.019 (-0.029, -0.010)**
|
0.005
(0.002, 0.009)**
|
|
Body Mass Index
|
0.125 (0.115)
|
-0.258 (0.049)***
|
0.046 (0.015)***
|
0.006 (-0.002, 0.016)
|
-0.012 (-0.019, -0.005)**
|
0.003
(0.001, 0.005)*
|
|
Education
|
-0.269 (0.048)***
|
|
|
0.055 (0.032, 0.081)***
|
|
|
-0.204 (0.038)***
|
Cognition
|
-0.180 (0.069)**
|
0.470 (0.030)***
|
-0.047 (0.024)#
|
0.008 (0.000, 0.019)
|
-0.022 (-0.042, -0.002)#
|
0.006
(0.001, 0.012)#
|
|
Income
|
-0.018 (0.018)
|
0.133 (0.008)***
|
-0.231 (0.095)#
|
0.004 (-0.003, 0.012)
|
-0.031 (-0.052, -0.010)*
|
0.008
(0.002, 0.015)*
|
|
Total Indirect Effect for ADHD-PGS to GrimAge
|
0.237
(0.165, 0.313)***
|
|
Total Indirect Effect Education to GrimAge
|
-0.217 (-0.257, -0.178)***
|
|
|
Notes. SE = standard error. 95% CI = 95% Confidence Interval, estimated from a bias-corrected bootstrap procedure with 5000 draws. # p < .10; * p < .05; ** p < .01; *** p < .001. Total effect from ADHD-PGS to GrimAge: beta = 0.459, SE = 0.090, p < .0001. Total effect from Education to GrimAge: beta=-0.421, SE = 0.038, p < .0001.
Sensitivity Analysis for Education
Because it is also possible that an ADHD-PGS underlies cognitive challenges that would likely affect years of schooling pursued, we conducted analyses to assess whether the results found could be due to a differential effect of education among the lowest and highest quartiles of the ADHD genetic risk score. We constructed linear regression models to include an interaction term (ADHD-PGS*education), adjusted for the ADHD-PGS, education and covariates, and did not find evidence for the interaction (b=-0.04, p = .371) or when including only those in the lowest and highest quartiles of the genetic risk score.