Association of Early Adulthood 25-Year Blood Pressure Trajectories With Cerebral Lesions and Brain Structure in Midlife

Key Points Question Are blood pressure trajectories in early adulthood associated with brain structure and integrity in midlife? Findings In this cohort study of 853 adults aged 18 to 30 years who were followed up for 30 years, trajectories with a higher level of, or a gradual increase in, mean arterial pressure during early adulthood were associated with reduced brain health in midlife after adjusting for sociodemographics and cardiovascular risk factors. Meaning These results suggest that, starting in early adulthood, the longitudinal pattern in blood pressure levels may be an indicator of increased risk for future poor brain health; preventing blood pressure increases as early as young adulthood may be warranted.

This supplemental material has been provided by the authors to give readers additional information about their work.

CARDIA Brain MRI sub-study
The sample for the CARDIA Brain magnetic resonance imaging (MRI) sub-study was initially enrolled at Y25 exam from 3 of the CARDIA field centers: Birmingham, Minneapolis, and Oakland. Each of the 3 centers had a target sample size for the sub-study, with the aim of achieving a balance within 4 strata of ethnicity/race (black/white) and sex (men/women). When the target sample size was reached, enrollment was ended. Exclusion criteria included a contraindication to MRI or a body size that was too large to enable the MRI examination. At year 25, 719 individuals from the CARDIA cohort underwent brain MRI examination. This same baseline sample was re-invited for a second scan in the Y30 follow-up. As some participants did not have a second exam (n=231), we recruited new participants (n=175) giving a total of 663; 488 had repeat MRIs. We merged the sample with good quality MRI scans to increase the sample size (n=885).

Details on the MRI hardware and quality control
All brain MRI scans were performed per CARDIA protocol using 3-Tesla magnetic resonance scanners, and standardized across machines using a common machine head phantom (Oakland: Siemens [Munich, Germany] 3T Tim Trio/VB 15 platform; Minneapolis: Siemens 3T Tim Trio/VB 15 platform; and Birmingham: Philips [Best, the Netherlands] 3T Achieva/2.6.3.6 platform). 1 Structural [3D sagittal T1, T2, and fluid-attenuated inversion recovery] and pseudocontinuous arterial spin labeling sequences were acquired. Brain MRI imaging analyses, quality control checks, and atlas registration were performed at the CARDIA Brain MRI Reading Center (Department of Radiology, University of Pennsylvania, Philadelphia, PA) according to standardized protocols as described previously. 1 Brain microstructural tissue integrity was estimated from axial diffusion tensor images using previously described parameters. 1

Details on the combination of Y25 and Y30 MRI data
The brain magnetic resonance imaging (MRI) was conducted for a subset of CARDIA participants at exams of Y25 (n=719) and Y30 (n=663). Of the unique 885 subjects with MRI scans, 485 had repeated exams of good quality. To increase our sample size, we pooled the data from the 2 exams by calculating the weighted average for each brain measure. When combining the MRI measures (means), it seems appropriate to weight them according to the amount of noise (ex., due to machine error) associated with each. In addition, the MRI measures measured at Y30 exam may have larger variance than Y25 (biological portion). To get a weighted average, per sequence, we first estimated the noise and biological portions for each MRI outcomes based on the repeated data. The MRI measures at exams of Y25 and Y30 can be written as: 2 ~ N(0, σ 2 ) ; 1 ~ N(0, σ 2 /w) ; 2 ~ N(0, σ 2 /w) We define as the observed trait value of individual i at the j th exam (i.e., 1 is the MRI outcome at the exam Y25), as the real trait value of individual i, and as the noise component of individual i at the j th exam. A is the weight that represents the unequal variances at the different time points (biological portion). All are independent from and independent from one another. In addition, we assume is normal with mean 0 and variance σ 2 , and is normal with mean 0 and variance σ 2 /w. And the weighted average was calculated as: ⁄ To get w and A, we first estimated the slope of the regression model 2~1 : (1 + ) ⁄ And then, we calculated the ratio of variances ( ): ( 2 ) ( 1 ) ⁄ = ( 2 2 + 2 ⁄ ) ( 2 + 2 ⁄ ) ⁄ = ( 2 + 1) (1 + ) ⁄ In short, we solved the quadratic equation to get w and A: (1 + ) ⁄ = ( ) ( 2 + 1) (1 + ) ⁄ = ( ) And w and A are equal to: The weighted average was calculated as the following formulas: • Subjects with repeated MRI measures:

Correcting volume measures for intracranial volume differences between Y25 and Y30
There was a within person change in total intracranial volume (ICV), which should be constant over time. To account for this, the tissue volume needs to be benchmarked against the same standard. Therefore, all volume measures at the exam of Y30 were corrected by the following equation before calculating the weighted average.

Details of group-based trajectory modeling
Group-based trajectory modeling (GBTM, SAS Proc Traj, v9.4) was used to identify unique polynomial functions that define both intercepts and slopes for each group trajectory. We identified 5 distinct trajectories for each blood pressure trait, including systolic blood pressure (SBP), diastolic blood pressure (DBP) and mean arterial pressure (MAP), across 25 years. The CARDIA has measured blood pressure at each of exam, with up to 8 BP measures. Blood pressure measured during pregnancy was removed from GBTM analysis. Subjects with at least 3 repeated BP measures from Y0 to Y25 were included in the trajectory modeling.
As reported in Allen et al.'s JAMA study, 2 we hypothesized 5 trajectory groups for each blood pressure trait. We then investigated models with a varying number of trajectory groups in a stepwise manner, starting with the highest polynomial (ex., cubic function) for each group trajectory. If the highest polynomial is not statistically significantly different from 0 (p > .05), we replaced it with a lower polynomial function (ex., quadratic or linear function). 3 The final number of trajectories were determined by the Bayesian Inclusion Criteria (BIC), with larger negative values for the BIC signifying better fit.
To ensure that Proc Traj program accurately assigned each subject to the appropriate trajectory group, four diagnostic criteria were used to evaluate the adequacy of the selected models: 3 1) the average posterior probability (AvePP) calculated for each trajectory group exceeded 0.7 threshold; 2) the estimated probability of group membership (π) should correspond closely to the proportion of the sample assigned to the group (number of people in that group / total number of subjects); 3) the confidence intervals around estimated group memberships (π) should be reasonably narrow; 4) the odds of correct classification (OCC) for each trajectory group exceeded the minimum threshold of 5, where OCC is defined by [(AvePP/(1-AvePP)] / [π/(1-π)]. Figure S1.      a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: For the volume of total brain, gray matter and hippocampus, n= 853; normal-looking whiter matter and abnormal white matter volume, n=846; white matter fractional anisotropy and mean diffusivity, n=777; and cerebral blood flow n=753. c: All outcomes were standardized to Z-score values. d: All models were adjusted for age, sex, race, education, and field center. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added. Model 3 was additionally adjusted for antihypertensive medication. Tissue volume measures models additionally adjusted for total intracranial volume; white matter fractional anisotropy and gray matter cerebral blood flow models additionally adjusted for total brain volume. e: Abnormal white matter volume was log-transformed before standardizing to Z-score values. f: Adjusted means based on the estimated model when intracranial volume/total brain volume was standardized, and age and physical activity level centered. eTable 7. The association of diastolic blood pressure (DBP) trajectory with brain outcomes (reference DBP trajectory: Low-stable, overall n=853) a,b,c,d Lowstable (n=181 ) CI=confidence interval. a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: For the volume of total brain, gray matter and hippocampus, n= 853; normal-looking whiter matter and abnormal white matter volume, n=846; white matter fractional anisotropy and mean diffusivity, n=777; and cerebral blood flow n=753. c: All outcomes were standardized to Z-score values. d: All models were adjusted for age, sex, race, education, and field center. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added.

DBP Trajectory Group
Model 3 was additionally adjusted for antihypertensive medication. Tissue volume measures models additionally adjusted for total intracranial volume; white matter fractional anisotropy and gray matter cerebral blood flow models additionally adjusted for total brain volume. e: Abnormal white matter volume was log-transformed before standardizing to Z-score values. f: Adjusted means based on the estimated model when intracranial volume/total brain volume was standardized, and age and physical activity level centered. eTable 8. The association of systolic blood pressure (SBP) trajectory with cognitive function outcomes at Y30 exam (reference SBP trajectory: Low-stable, overall n=2,736) a,b,c,d Lowstable (n=705) Moderate-gradual (n=1,225) Abbreviations: RAVLT=Rey Auditory Verbal Learning Test; DSST=Digital symbol substitution test; CI=confidence interval. a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: All outcomes were standardized to Z-score values. c: All models were adjusted for age, sex, race, education, and field center. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added. Model 3 was additionally adjusted for antihypertensive medication. d: Adjusted means based on the estimated model when intracranial volume/total brain volume was standardized, and age and physical activity level centered. eTable 9. The association of diastolic blood pressure (DBP) trajectory with cognitive function outcomes at Y30 exam (reference DBP trajectory: Low-stable, overall n=2736) a,b,c,d Lowstable (n=587) Moderate-gradual (n=1,205)   Abbreviations: RAVLT=Rey Auditory Verbal Learning Test; DSST=Digital symbol substitution test; CI=confidence interval. a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: All outcomes were standardized to Z-score values. c: All models were adjusted for age, sex, race, education, and field center. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added. Model 3 was additionally adjusted for antihypertensive medication. d: Adjusted means based on the estimated model when intracranial volume/total brain volume was standardized, and age and physical activity level centered. eTable 11. The association of systolic blood pressure (SBP) trajectory with brain outcomes after adjustment of Y25/Y30 BP measure (reference SBP trajectory: Low-stable, overall n=853) a,b,c,d Lowstable (n=207)  a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: For the volume of total brain, gray matter and hippocampus, n= 853; normal-looking whiter matter and abnormal white matter volume, n=846; white matter fractional anisotropy and mean diffusivity, n=777; and cerebral blood flow n=753. c: All outcomes were standardized to Z-score values. d: All models were adjusted for age, sex, race, education, field center, and Y25/Y30 measure. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added. Model 3 was additionally adjusted for antihypertensive medication. Tissue volume measures models additionally adjusted for total intracranial volume; white matter fractional anisotropy and gray matter cerebral blood flow models additionally adjusted for total brain volume. e: Abnormal white matter volume was log-transformed before standardizing to Z-score values. f: Adjusted means based on the estimated model when intracranial volume/total brain volume and Y25/Y30 BP measure were standardized, and age and physical activity level centered. a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: For the volume of total brain, gray matter and hippocampus, n= 853; normal-looking whiter matter and abnormal white matter volume, n=846; white matter fractional anisotropy and mean diffusivity, n=777; and cerebral blood flow n=753. c: All outcomes were standardized to Z-score values. d: All models were adjusted for age, sex, race, education, field center, and Y25/Y30 measure. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added. Model 3 was additionally adjusted for antihypertensive medication. Tissue volume measures models additionally adjusted for total intracranial volume; white matter fractional anisotropy and gray matter cerebral blood flow models additionally adjusted for total brain volume. e: Abnormal white matter volume was log-transformed before standardizing to Z-score values. f: Adjusted means based on the estimated model when intracranial volume/total brain volume and Y25/Y30 BP measure were standardized, and age and physical activity level centered. eTable 13. The association of mean arterial pressure (MAP) trajectory with brain outcomes after adjustment of Y25/Y30 BP measure ( a: q-value was calculated using Benjamini-Hochberg method for the adjustment of multiple comparison. q ≤0.1 was considered statistically significant. b: For the volume of total brain, gray matter and hippocampus, n= 853; normal-looking whiter matter and abnormal white matter volume, n=846; white matter fractional anisotropy and mean diffusivity, n=777; and cerebral blood flow n=753. c: All outcomes were standardized to Z-score values. d: All models were adjusted for age, sex, race, education, field center, and Y25/Y30 BP measure. In Model 2, BMI, diabetes, physical activity level (log-transformed), current smoking status, and alcohol use were added. Model 3 was additionally adjusted for antihypertensive medication. Tissue volume measures models additionally adjusted for total intracranial volume; white matter fractional anisotropy and gray matter cerebral blood flow models additionally adjusted for total brain volume. e: Abnormal white matter volume was log-transformed before standardizing to Z-score values. f: Adjusted means based on the estimated model when intracranial volume/total brain volume and Y25/Y30 BP measure were standardized, and age and physical activity level centered.