High blood pressure predicts hippocampal atrophy rate in cognitively impaired elders

Abstract Introduction Understanding relationships among blood pressure (BP), cognition, and brain volume could inform Alzheimer's disease (AD) management. Methods We investigated Alzheimer's Disease Neuroimaging Initiative (ADNI) participants: 200 controls, 346 mild cognitive impairment (MCI), and 154 AD. National Alzheimer's Co‐ordinating Center (NACC) participants were separately analyzed: 1098 controls, 2297 MCI, and 4845 AD. Relationships between cognition and BP were assessed in both cohorts and BP and atrophy rates in ADNI. Multivariate mixed linear‐regression models were fitted with joint outcomes of BP (systolic, diastolic, and pulse pressure), cognition (Mini‐Mental State Examination, Logical Memory, and Digit Symbol) and atrophy rate (whole‐brain, hippocampus). Results ADNI MCI and AD patients with greater baseline systolic BP had higher hippocampal atrophy rates ([r, P value]; 0.2, 0.005 and 0.2, 0.04, respectively). NACC AD patients with lower systolic BP had lower cognitive scores (0.1, 0.0003). Discussion Higher late‐life BP may be associated with faster decline in cognitively impaired elders.

heterogeneous and blood pressure (BP) may partially account for this variability.
Atrophy rate is a biomarker of AD progression, and an important outcome measure in clinical trials. 9 It is necessary to understand how BP associates with atrophy rates to enable effective planning of clinical trials in which they are used as outcome measures.
Targeting BP in the same way across the cognitive spectrum may not be appropriate. 10 Low BP is associated with poorer cognition in the elderly 11 and is common in AD patients. 11,12 Falling BP is also associated with worsening cognition in the hypertensive elderly. 13 Here, we aimed to understand relationships among late-life BP, cognition, and brain volume in 700 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI). We also investigated BP and cognition in a group of 8240 participants from a complementary dataset: National Alzheimer's Coordinating Center (NACC). We hypothesized that BP would be associated with brain volume change; changes in BP would be associated with changes in brain volume. Similarly, we hypothesized that baseline BP would be associated with baseline neuropsychology and change in neuropsychology. We also considered that BP changes would be associated with changes in neuropsychology.
Data are freely available for qualified researchers. Participants All participants were from the first phase of the study (ADNI1).
All MCI and AD patients had an amnestic impairment. AD patients met criteria for probable AD according to the NINCDS-ADRDA criteria. 14  were not enrolled in ADNI1 ( Figure S1 in supporting information).
Imaging data were unavailable for the majority of subjects; therefore, for NACC we investigated BP and cognition relationships only. Criteria on MCI diagnosis for ADCs was developed using consensus guidelines set by an expert panel. 17 MCI patients were included if they had memory problems. AD patients were demented with a diagnosis of probable or possible AD at first visit. 14 We excluded individuals with secondary causes of dementia (see section 1.1 in supporting information).

BP readings
Seated systolic blood pressure (SBP) and diastolic blood pressure (DBP) readings were measured at all visits for ADNI and at approximately annual visits for NACC (see section 1.2 in supporting information). Pulse pressure (PP) was calculated as SBP minus DBP.

Image analysis
The ADNI MRI protocol is detailed elsewhere. 18 After the acquisition, quality control was completed and ADNI image pre-processing was then applied, including gradient warping, 19 B1 non-uniformity, 20 and intensity non-uniformity correction. 21 Imaging data consisted of all available ADNI1 time points from baseline to 36 months (0-, 6-, 12-, 18-, 24-, and 36-month scans), with a T1-weighted volumetric scan acquired on a 1.5T scanner of sufficient quality. Internal visual quality control was performed, excluding images with severe blurring of tissue boundaries due to motion artefacts. Whole-brain and hippocampal volumes were estimated automatically from the 1.5T volumetric T1-weighted images using BMAPS 22 and HMAPS, respectively. 23 The boundary shift integral (BSI) was used to estimate change directly from scan pairs after segmentation, 24 the outcome representing change in volume of wholebrain or hippocampus (mL) during the scan interval. Total intracranial volume (TIV), a proxy for head size, was also calculated. 25

Cognition
We for processing speed. Examinations were administered at each visit, apart from LM in ADNI, which was administered annually.

Demographic information
Demographic differences between groups for continuous outcomes were tested using analysis of variance; Fisher's exact test was used for categorical outcomes. TIV was used as a covariate for brain and hippocampal outcomes.

Joint modeling of outcomes
We jointly modeled change in BP (using SBP, DBP, and PP in separate models), change in cognition, and change in brain/hippocampal volume. There was no "predictor" as with traditional regression models; to observe concurrent changes in BP, cognition, and atrophy, each of the three variables were jointly modeled as outcomes. Multivariate mixed linear-regression models were fitted with outcomes of BP (SBP, DBP, PP), atrophy (whole-brain, hippocampal change), and cognition (MMSE, LM, DSST). A random intercept (BP, atrophy rate, and cognition) and random slope were included for the participant level with unstructured covariance to allow for correlations among all random effects, apart from the random intercept for atrophy rate. The atrophy rate random intercept was included to account for the random measurement error at the first scan, which would be shared between all atrophy measures. Likelihood ratio tests were used to test random effects correlations between outcomes. The following correlations were examined: (1) baseline BP, change in brain volume; (2) change in BP, change in brain volume; (3) baseline BP, baseline neuropsychology; (4) baseline BP, change in neuropsychology; (5) change in BP, change in neuropsychology. Models were run separately in each diagnostic group.
No fixed effects intercept was included in the model for change in brain volume due to the assumption that the estimated atrophy rate over a scan interval of zero is zero. As no intercept, or measure of baseline volume change, is estimated using BSI (volume change) data, we separately modeled baseline brain and hippocampal volume in a regression model (see below). We used the multilevel modeling software MLWin version 2.36, February 2016 release. 26 For BP and cognitive models, age and sex were added as covariates to adjust for their associations with the baseline value of these outcomes, and for all outcomes these covariates were interacted with time, to adjust for their associations with changes in outcome. Apolipoprotein E (APOE) 4 status (binary covariate indicating presence or absence of an 4 allele) was included in the relevant ADNI models to adjust for associations with baseline BP, baseline MMSE, MMSE change, and brain volume change.
Fixed effects data are tabulated in supporting information. Due to lack of any detectable change over time in SBP and DBP for ADNI AD patients the random effect slope for change in blood pressure was removed to allow the model to converge. For PP in the ADNI AD group the random effect slope was permitted, but the random effect correlation between baseline PP and change in PP was omitted.
For NACC, models were also fitted with adjustment for antihypertensive medication use and also for APOE 4 (binary covariate indicating presence or absence of an 4 allele), see section 1.3 in supporting information.
In ADNI, to investigate correlations between baseline BP and baseline whole-brain/hippocampal volumes, regression models were fitted, with separate outcomes of whole-brain/hippocampal volume and baseline BP (either SBP, DBP, or PP) as the predictor. Age, APOE 4 status, TIV, and sex were included as covariates.

Analysis by hypertensive status in AD
AD patients were categorized by hypertensive status based upon initial BP and antihypertensive usage. BP cut-offs for hypertension were defined as = >140 mmHg for SBP and/or = >90 mmHg for DBP. 27 Hypotension was defined as SBP = < 90 mmHg and/or = <60 mmHg for DBP, as defined by the United States Institutes of Health for groups and fixed effects resulting from the models are tabulated in Table S7 in supporting information.
For information on figures and tabulation see section 1.4 in supporting information. For reference we ran all graphically presented effects without adjustment for covariates.

Demographic information
After quality control, 700 ADNI participants were included in this study (see Figure S1). MCI and AD patients had lower brain and hip-pocampal volumes, greater cognitive impairment, and were more likely to be an APOE 4 carrier than controls (Table 1).
A total of 8240 NACC participants were suitable for the present study. Controls were older and were more likely to be female, compared with MCI and AD groups.

Control results
From both ADNI and NACC, controls on average had no significant change in BP over time (see Table S1 in supporting information). ADNI controls had significant rates of brain and hippocampal volume loss, but did not decline in cognition over time. The F I G U R E 1 Relationships between baseline blood pressure (BP) and baseline psychology or hippocampal volume. Graphs to demonstrate relationships between baseline blood pressure and baseline psychology or hippocampal volume (

BP and atrophy
ADNI control participants with higher baseline DBP had lower baseline brain volumes, see Figure 1a,

BP and neuropsychology
In both cohorts, higher BP tended to be associated with worse baseline neuropsychology scores in controls, although this was not found across all measures. In ADNI, controls with higher baseline DBP had a worse baseline MMSE score (correlation coefficient, P value) (-0.24, 0.05) (see Figure 1B, Table S4 in supporting information). In NACC, controls with higher SBP had lower baseline LM (-0.13, 0.01) (see Table 2, Figure 1C). NACC controls with higher PP also had lower DSST scores (-0.12, 0.01). No correlations were found between changes in BP and changes in neuropsychology.

MCI results
In both ADNI and NACC, MCI participants had falling DBP over time (Table S1). ADNI MCI participants declined in MMSE and DSST over time, but not LM. NACC MCI participants showed significant rates of decline in all neuropsychology measures.

BP and atrophy
For ADNI MCI patients, higher baseline SBP was associated with higher hippocampal atrophy rates (correlation coefficient, P value) (0.20, 0.005) (see Table 3, Figure 2A) and those with higher PP also had TA B L E 2

BP and neuropsychology
In NACC, higher baseline SBP showed significant correlations with lower baseline MMSE (−0.11, 0.01) (see Table 2, Figure S2a  No correlations were found between BP and neuropsychology in ADNI (see Table S4).

AD results
The ADNI AD group did not show any significant change over time in BP, but falling SBP, DBP, and PP were all detected in the NACC AD group, see Table 1.

BP and atrophy
Higher baseline SBP was associated with greater subsequent hippocampal atrophy (0.23, 0.04) (see Table 3, Figure 2B); this relationship was mirrored by DBP (0.22, 0.05) (see Table 3, Figure S2d All ADNI relationships are corrected for APOE genotype, age, sex, and total intracranial volume. All NACC relationships adjusted for age and sex. Results for models without covariate adjustment are also shown.^Similar relationship exists for pulse pressure; + similar relationship exists for diastolic blood pressure between change in PP and either hippocampal or whole brain atrophy rate.

BP and neuropsychology
In ADNI, no significant correlations with neuropsychology were observed. However, in NACC, higher SBP pressure and higher PP were associated with higher baseline MMSE (0.09, 0.0003 [ Figure 1D) and 0.09, 0.0005 respectively). Falling SBP and declining PP were associated with faster declines in MMSE (0.19, 0.003 [ Figure S2e] and 0.14, 0.001, respectively).

Analysis of AD patients by hypertensive group and medication
We found low baseline BP predicted low baseline MMSE in the medicated hypotensives (see Table 4 There were no relationships between BP and cognition in the medicated normotensives, despite being the largest group.

Adjustment for APOE genotype and antihypertensive use
With adjustment for APOE 4 genotype in NACC the overall relationships between BP and cognition were not materially changed ( Graphically represented results were also tested without adjustment for covariates and the majority of findings were materially unchanged (see Figures 1 and 2 and Figure S2). The relationship between DBP and atrophy rate in AD participants changed to trend level without covariate adjustment ( Figure S2). Without covariate adjustment the relationship between DBP and brain volume in controls was no longer significant (Figure 1), and the relationship between change in MMSE and SBP in AD was no longer significant ( Figure S2); however, for both results the effect size and direction remained similar.

DISCUSSION
The central finding of this study is that higher baseline BP is associated with greater hippocampal atrophy rates in MCI and AD. This is the first demonstration, to our knowledge, of a potentially modifiable risk factor which is associated with hippocampal atrophy rate in clinical AD. In AD, BP in late-life has been inconsistently related to cognition and cognitive decline. 10 High BP has been identified as a risk factor for lower MMSE score 10 and smaller whole brain volume. 34 Conversely, low BP has been found to predict greater longitudinal atrophy in cognitively normal individuals with manifest arterial disease. 5 We did not find high BP predicted atrophy rate in controls, similarly to others 35-37 although some have detected relationships. 6,7 The duration of hypertension may explain why dementia is most consistently associated with mid-, rather than late-life BP.
SBP appeared to have the strongest influence on brain volumes and cognition across the cohorts. Many relationships between SBP and cognition in NACC were mirrored by similarly strong effects in PP. PP is considered an indirect measure of arterial stiffness, and has also been shown to negatively influence brain volumes and cognition. 38 We found the impact of BP on cognition was dependent on disease stage; low BP correlated with poor cognition in AD, whereas in controls and MCI high BP predicted lower cognition. Falling BP tracked with declining MMSE in both MCI and AD. Low BP has previously been found to relate to poorer cognition 12 and is frequently observed in old AD patients. 11,12 At the prodromal AD stage, high BP may be harmful, potentially lowering the threshold at which symptoms appear (perhaps through cerebrovascular disease); falling and subsequently low BP in symptomatic AD may then reflect disease stage. Studies investigating long-term BP trajectories have found that hypertension, followed by declining BP, and subsequent hypotension is associated with lower cross-sectional brain volumes 39,40 and is common in those who develop dementia. 41 One study showed that increases in BP from ages 36 to 43 years were associated with lower brain volumes at 70 years indicating that monitoring of BP needs to happen from early adulthood. 42 Low BP is likely to impair cerebral perfusion, which may be worse in those with histories of hypertension, due to arterial stiffening and impaired cerebro-autoregulation. 43 Elderly individuals may also require a higher BP to maintain adequate perfusion. 12 Conversely, low BP could also be a product of AD-related neurodegeneration in BP-regulating brain areas. 44 When the NACC AD group was categorized dependent on baseline hypertensive status, a pattern emerged which supports careful BP monitoring. In non-medicated hypertensives, high baseline SBP was detrimental for future cognition, whereas in hypotensives and non-medicated normotensives, low SBP was predictive of cognitive decline. Interestingly, those with successfully lowered BP were the only group without relationships between BP and cognition, despite being the largest group. Notably in this group, baseline SBP was more than 10mmHg higher than the non-medicated normotensives; a higher BP may be necessary to maintain adequate brain perfusion in old age. 12 Using data from two cohorts, our study was large, with a combined total of 8940 participants. Our models investigated longitudinal data from three outcomes, which enabled detection of joint patterns at an individual level. We also investigated relationships across the cognitive spectrum. While we ran a relatively large number of models, we opted not to correct for multiple comparisons, as our tests were hypothesis driven. Follow-up was relatively short in both cohorts. It may be that decades of follow-up are needed to fully understand the relationships among BP, brain volumes, and cognition. Both cohorts have limitations in generalizability because subjects enrolled in studies tend to be younger and fitter than those in the community. 45

CONFLICTS OF INTEREST
There are no reported conflicts apart from a note that Christopher Lane is now employed by Eisai Europe Ltd.

FUNDING INFORMATION
The ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer's