Current and Remote Blood Pressure and Cognitive Decline FREE

High blood pressure (BP) is an important risk factor for vascular disease, including stroke, in the elderly.^1- 2 Because vascular disease can lead to dementia,^3- 4 this suggests a possible association between elevated BP and increased risk of dementia. However, the findings of cross-sectional, population-based studies of BP and cognitive function in the elderly have varied greatly. Some studies found higher rates of cognitive impairment associated with elevated BP,⁵ others with low BP,⁶ and other studies found little or no association.⁷ These results mirror the conflicting associations between BP and cardiovascular disease mortality found in observational studies of the elderly.⁸

Interpretation of cross-sectional relationships between BP and cognitive function is problematic because dementia itself may affect BP through its effects on diet⁹ and weight loss.¹⁰ Studies of the relationships of midlife BP with late-life cognitive function assessed at a single time point, in the Framingham¹¹ and Honolulu¹² heart studies and in an Uppsala study,¹³ found higher proportions with cognitive impairment among those with elevated BP during midlife. Two small longitudinal studies found decreased performance on intelligence tests¹⁴ or higher rates of dementia¹⁵ associated with elevations in diastolic BP at baseline. Two BP treatment trials reported conflicting results of antihypertensive treatment on cognitive test performance.^2,16

In the present analysis, we used prospective data to examine whether elevated BP predicted level of or change in cognitive function among persons 65 years or older residing in East Boston, Mass. More than 2000 of the 3809 East Boston participants in the Established Populations for the Epidemiologic Study of the Elderly (EPESE) project had BP measurements 9 years before the beginning of this study as part of the Hypertension Detection and Follow-up Program (HDFP). This allowed us to examine the relationships of both earlier and concurrent BP measurements with level of and change in cognitive function.

In 1982 and 1983, as part of the National Institute on Aging's EPESE project, a census identified 4497 noninstitutionalized residents, 65 years or older, of East Boston. A total of 3809 (85%) were interviewed in their homes, either in person or through a proxy. Blood pressure was measured 3 times at intervals of 30 seconds, while the subject was seated, according to the protocol of the HDFP.¹⁷

The interviewer also administered a 6-item memory test (the East Boston Memory Test¹⁸ [EBMT]) and a 9-item version of the Pfeiffer Short Portable Mental Status Questionnaire¹⁹ (SPMSQ) (the question, "What is the name of this place?" was omitted from these home interviews). The SPMSQ has been shown to be valid and reliable,^18- 19 and the sensitivity and positive predictive value of the EBMT in detecting clinically diagnosed cognitive impairment are at least comparable to 2 other commonly used brief cognitive tests.^18,20 For this study, we restricted analyses to the 3657 participants with baseline BP measures; 3586 of these completed the SPMSQ and 3576 completed the EBMT.

The interviewer also collected information about other characteristics potentially related both to BP and cognitive function. Participants were asked their number of years of formal schooling, their present and past use of cigarettes and alcohol, and whether they had a history of myocardial infarction or stroke. Angina was identified by the Rose questionnaire.²¹ All medications used in the past 2 weeks were identified by direct inspection.

Nine years before this baseline assessment, 2068 of the study participants had their BP measured in their homes by the same protocol, as part of the population-based screening visit (completed by 89% of enumerated residents) for the HDFP clinical trial.¹⁷ Because the trial had an upper age limit, all of these participants were 81 years or younger at the baseline EPESE evaluation. Within the age range of 65 to 81 years, 67% of participants had BP measurements taken 9 years previously. We linked data from these individuals to allow for evaluation of the relationship of earlier BP measurements with level and change in cognitive function from 9 to 15 years later.

In-home, follow-up interviews occurred 3 and 6 years after baseline and included the same measures of cognitive function described above. Of the 3133 surviving cohort members at the time of the 3-year follow-up, 2736 completed the SPMSQ and 2679 completed the EBMT. Of the 2502 participants alive at the time of the 6-year follow-up, 1994 completed the SPMSQ and 1970 completed the EBMT. Additionally, 69 participants had missing data on their number of years of education. These individuals were excluded from multivariate analyses controlling for education. Hence, multivariate analyses of SPMSQ scores at all 3 times included 8235 observations in 3588 people; comparable analyses of memory scores included 8142 observations. Further information about demographic characteristics of the population,²² quality control of BP measurements,²³ and participation in assessment of cognitive function²⁴ has been published elsewhere.

The outcomes of interest were the numbers of errors in the SPMSQ and the EBMT, and changes in these numbers over time. For both analyses of baseline BP and BP measured 9 years earlier, we calculated the systolic and diastolic levels of each subject as the averages of the 3 measurements, respectively. In initial graphical analyses we explored the relationships of systolic and diastolic BP, both at baseline and 9 years earlier, with the baseline numbers of errors on both tests, using the square roots of these numbers to stabilize their variance. These analyses used robust, local-weighted regression techniques²⁵ to fit smooth curves without assumptions about the parametric form of the relationships between BP and numbers of errors. The graphs suggested that relationships between BP and cognitive function were nonlinear, with some elevations in numbers of errors among those with low or high BP. In subsequent analyses, we classified participants into 5 groups based on their systolic BP or 4 groups based on their diastolic BP. Categories used were the same as those in a previous analysis of BP and mortality in this population,⁸ except that those with diastolic BP between 70 and 79 mm Hg were grouped in the present analysis. Because of the possible U-shaped relationships between BP and cognitive function at baseline, the reference group for systolic BP had values between 130 and 139 mm Hg and for diastolic BP, we chose the group with values between 70 and 79 mm Hg.

To control for potential confounding variables and to simultaneously examine cognitive function scores at all 3 evaluations, we fitted multivariate regression models with a dependent variable of either the number of errors on a cognitive test or the rate of change in the square root of this number over time. When the dependent variable was the number of errors on a cognitive test at baseline, 3-year, or 6-year follow-up, we used a Poisson regression model to relate categories of BP to this number. Parameters in this model are interpreted as the logarithm of the relative number of errors comparing individuals in 1 category of BP with the referent category. Also included in the model was a scale term for overdispersion because errors were not random among individuals.²⁶ We accounted for the correlation between the repeated measures of cognitive function in the same persons by using the generalized estimating equation approach described by Liang and Zeger.²⁷ We fitted models using SAS statistical software and made no assumptions about the values of the intercorrelations between repeated measures.²⁸ Potentially confounding variables included in all models were age (centered at 75 years), age squared, age cubed (all 3 of these variables were significantly related to numbers of errors), education (in years), sex, and the relative effects of evaluation time (3-year and 6-year follow-up compared with baseline). To examine the relationship of BP with change in cognitive function over time, we added interaction terms between BP categories and time of evaluation (scored as 0 for baseline, 1 for the 3-year follow-up, and 2 for the 6-year follow-up).

To examine rates of change in numbers of errors on cognitive tests we first obtained a slope for each person, based on a univariate linear regression, that described the rate of change over time in the square root of the number of errors. Because the distribution of slopes was skewed and because slopes were strongly related to baseline scores on cognitive tests, we used a normal scores transformation to adjust for regression to the mean and obtain a measure of change independent of baseline score.²⁹ We did this by first ranking the individual slopes separately within each level of baseline score. We then transformed these ranks to the value, or normal score, that corresponded to the percentile of a normal distribution with mean 0 and variance 1.³⁰ These normal scores were then used as dependent variables in multiple linear regression models that included the same independent variables as the Poisson regression models except for the indicators of evaluation time.

We repeated multivariate analyses within strata to examine potential modification of relationships by comorbidity or use of antihypertensive medications; another analysis included tobacco and alcohol intake. One analysis of people with diseases potentially affecting cognitive function was restricted to those who reported at baseline a history of myocardial infarction or stroke, had angina according to the Rose questionnaire, or were using digoxin, loop diuretics, or hypoglycemic agents. Another analysis considered only those with none of these conditions. We also performed 2 separate analyses stratified according to whether participants were taking antihypertensive medications (1) at baseline, or (2) at the time of BP assessment 9 years before baseline.

Because nonparticipation in cognitive assessments can be related to actual cognitive abilities,³¹ we also fitted several models after multiple imputation³² of missing scores on the SPMSQ, based on several alternative assumptions about the patterns of missing data. However, we noted little change in the effects of BP on cognitive function. Thus, these alternative analyses are not presented.

Baseline systolic and diastolic BP show U-shaped associations with baseline SPMSQ scores with a tendency for more errors among those with extreme values of BP (Figure 1). However, associations of BP 9 years previously with SPMSQ scores were somewhat different and suggested a possible relationship between higher systolic BP and an increased number of errors.

When BP was related to the square root of the number of errors on the EBMT (Figure 2), little association existed between systolic BP and errors on the EBMT. Low diastolic BP was possibly associated with a greater number of errors, although in this analysis potentially confounding variables were not accounted for.

Several characteristics known to be related to the risk of lower scores on cognitive tests differed in their distributions across categories of BP. Participants with elevated systolic BP tended to be older and to have less education (Table 1). Conversely, those with low diastolic BP were older and had less education than those with both intermediate and elevated levels of diastolic BP (Table 2). After adjustment for age and sex, the prevalence of several conditions that may modify the association between BP and cognitive function also varied across categories of BP at baseline. A history of myocardial infarction and use of digoxin and loop diuretics were more common among those with low systolic BP, whereas drug treatment for diabetes and a history of stroke were more common among those with elevated systolic BP. Participants with diastolic BP lower than 70 mm Hg had elevated rates of self-reported myocardial infarction and stroke and were more likely to use digoxin, loop diuretics, and hypoglycemic drugs, compared with those with high diastolic BP.

When baseline BP was compared with the average number of errors on cognitive tests at all 3 times, different associations were identified for the 2 cognitive tests (Figure 3). Errors on the SPMSQ during the 6-year follow-up appeared to have a small U-shaped association with both systolic and diastolic BP. Compared with the referent group (130-139 mm Hg) for systolic BP (Figure 3), persons with systolic BP lower than 130 mm Hg had, on average, 9% more errors (95% confidence interval [CI], 1%-17%; P=.02) and those at 160 mm Hg or higher had 7% more errors (95% CI, 0%-15%; P=.05). For diastolic BP (Figure 3), those at 90 mm Hg or higher averaged 11% more errors (95% CI, 4%-19%; P=.003) than the referent group, but the group with low BP had approximately the same number of errors as the referent. In contrast, no association existed between either systolic or diastolic BP (Figure 3) and performance on the EBMT.

When the BPs measured 9 years before baseline were compared with average number of errors over the 6 years of follow-up (Figure 4), no association was found between performance on the EBMT and either systolic or diastolic BP. For the SPMSQ and systolic BP, we again found a slightly increased error rate among those with elevated BP; the subgroup with BP of 160 mm Hg or higher averaged 14% more errors than the 130 to 139 mm Hg referent group (95% CI, 4%-25%; P=.007). For the SPMSQ and diastolic BP, the results, as with the baseline BP data, suggested a small U-shaped relationship. The subgroup with diastolic BP lower than 70 mm Hg 9 years previously had 12% more errors (95% CI, 2%-24%; P=.02) than the 70 to 79 mm Hg referent group, and those with BP of 90 mm Hg or higher had 10% more errors (95% CI, 1%-20%; P=.04).

Overall, BP either at baseline or 9 years earlier did not strongly predict change in cognitive function over a 6-year period of observation for either the SPMSQ (Table 3) or the EBMT (Table 4). One exception was that those with elevated systolic BP (≥160 mm Hg) 9 years before baseline had a significantly greater increase in numbers of errors over time (P=.02), relative to those with systolic BP of 130 to 139 mm Hg (Table 3). When we added interaction terms between BP categories and time of cognitive assessment to the analyses of BP and level of cognitive function, we also found no evidence that BP, measured at baseline or 9 years earlier, was associated with cognitive decline, based on either measure of cognitive function.

We found little evidence that relationships between BP and cognitive function differed by antihypertensive medication use, tobacco or alcohol use, or the presence of potentially modifying diseases. For example, when we ran the analyses summarized in Figure 3 separately among those taking and not taking antihypertensive medications, those taking antihypertensive medications with diastolic BP of 90 mm Hg or higher had 12% more errors in the SPMSQ (P=.03) than those in the referent group, while those not taking antihypertensive medications in the same group had 10% more errors (P=.03) than the referent. Among participants without cardiovascular disease or diabetes, those with diastolic BP of 90 mm Hg or higher at baseline had 14% more errors (P=.001) relative to those in the referent category of diastolic BP, while among those with these diseases, participants with diastolic BP of 90 mm Hg or higher had 5% more errors (P=.48) than the referent group. None of the stratified analyses revealed any substantial associations between BP and performance on the EBMT.

The analyses presented herein examine the relationship between BP and cognitive function in several ways. Blood pressure was measured both at the time cognitive testing was done and, for a major fraction of the cohort, 9 years previously. Both level of cognitive performance and change in performance over time were examined, and 2 different brief tests of cognition were used. Overall, results did not demonstrate a linear association between BP and cognition. The results do, however, raise the possibility of more complex associations.

These associations were fairly small in magnitude and varied somewhat according to whether level of cognitive function or change in cognitive function was assessed and according to which brief test was used to measure cognition. There was little evidence for an effect of BP on change in cognitive function with either test; only elevated systolic BP measured 9 years previously predicted an increasing number of SPMSQ errors in 1 of the analyses of change. For analyses of BP and level of cognition, the results varied according to which test was used, but within each test were reasonably consistent. For SPMSQ performance, persons with elevations either in systolic or diastolic BP, either at baseline or 9 years earlier, had consistent and statistically significant elevations in error rates compared with those in the referent categories of BP. Specifically, those with systolic BP of 160 mm Hg or higher, either at baseline or 9 years earlier, and those with diastolic BP of 90 mm Hg or above at either time committed between 7% and 14% more errors on the SPMSQ, compared with the referent groups with systolic BP between 130 and 139 mm Hg or diastolic BP between 70 and 79 mm Hg, respectively. Furthermore, there was a suggestion of a U-shaped association between performance on this measure and both systolic BP at baseline and diastolic BP 9 years previously, because persons in the lowest BP category also made more errors than those in the referent category. In contrast to these results for the SPMSQ, there was little association between BP at either time and level of performance on the EBMT.

The magnitude of the observed associations between low and high levels of BP and increased error rates on the SPMSQ can be understood in the context of the observed associations of age and education with performance on this test. Both the effects of age and education were highly significant (P<.001). In the model fitted to the 8235 available SPMSQs, for which the BP effects are summarized in Figure 3, the predicted error rate in persons aged 78 years, relative to the referent group of persons aged 75 years, was 11% greater. Similarly, relative to participants with any given educational level, those with 2 fewer years of education were predicted to have a 15% greater error rate. It is also worth noting that in the entire population both age and education had similarly strong effects on error rates on the EBMT, which provides a frame of reference for the small observed effects of BP on this measure.

Previous studies have suggested an association between lower cognitive function in late life and elevations in BP measured some years earlier. In the Longitudinal Population Study of persons aged 70 years in Göteborg, Sweden, the 76 individuals who were not demented on examination at age 85 years had significantly lower mean systolic and diastolic BP at age 70 years compared with 18 individuals who developed dementia between ages 70 and 85 years.¹⁵ In the Framingham Study, higher levels of both systolic and diastolic BP measured during an interval between 12 and 22 years before administration of a cognitive test battery predicted lower scores on the cognitive tests.¹¹ In the Honolulu Heart Program, elevated levels of systolic BP 9 to 18 years before administration of the Cognitive Abilities Screening Instrument predicted poorer performance on this test; however, diastolic BP had no relationship with performance.¹² Conversely, the recent Uppsala study¹³ found an association between elevated diastolic BP at age 50 years and poorer cognitive performance at age 70 years. The first 3 of these studies found little association between BP measured at the time of cognitive assessment and level of cognitive function. This is consistent with our finding of a substantial elevation in errors on the SPMSQ associated with elevated systolic BP 9 years before baseline but an apparent U-shaped relationship between systolic BP at baseline and errors.

Strengths of the study include the large population, with multiple longitudinal measurements of BP and cognitive performance. Perhaps the greatest limitation is that measures of cognitive function are restricted to 2 brief tests. These tests have been used in previous population studies of cognition,^{18- 24,29,33} but it is possible that more detailed testing of cognitive function would have provided more definitive results. Other limitations include the absence of BP measurements during the 9 years between the HDFP screening and the baseline EPESE evaluation, and the possibility that the HDFP experience led to more intensive treatment of BP in East Boston. With regard to the last point, a comparison between East Boston and 2 other EPESE populations found comparable baseline rates of use of antihypertensive drugs across sites, and slightly higher mean systolic BP and prevalence of elevated systolic BP (≥160 mm Hg) in East Boston.³⁴

Our prior hypotheses were that higher BP would be linearly associated with decline in cognitive function over the 6-year period of cognitive testing, and that decline would be more strongly predicted by BP measured 9 years previously. Overall, our results do not support these hypotheses, but they do not unequivocally reject an association between high BP and cognitive decline. It is noteworthy that neither set of BP measurements predicted change in cognitive function on either test (except that systolic BP ≥160 mm Hg predicted decline in performance on the SPMSQ). Individual change in cognitive test performance may be a more meaningful outcome than level of performance at a single point in time since education and cultural factors also influence test performance. Individual change in test performance, on the other hand, is likely to provide an unbiased measure of cognition.²⁹ Whether our findings are due to chance alone or reflect a more complex relationship between BP and cognition than previously appreciated remains uncertain. The relationship of BP to cognitive function continues to deserve detailed investigation because of the great public health importance of decline in cognition in later life and the need to identify potentially modifiable risk factors for this decline. More definitive examination of this issue will likely require longitudinal population studies that include BP measurement at multiple points, as did this study, but also incorporate substantially more detailed cognitive testing at multiple points.