492 Adult Development and Aging
Robbins, & Schultz, 1987) and survival effects represent
two major problems encountered in cross-sectional studies
(M. Elias et al., 1990; Waldstein 1995, 2001). Consequently,
there is general agreement that longitudinal studies provide
the best paradigm for examining relations between hyperten-
sion, or any other cardiovascular disease risk factor, and cog-
nitive functioning.
Contemporary Longitudinal Studies
Findings of greater cognitive decline over a four-year
test/retest period (Tzourio, Dufouil, Ducimetière, &
Alpérovitch, 1999) and a six-year longitudinal period (Knop-
man et al., 2001) for middle-aged and elderly subjects are
consistent with the earlier “ndings (Wilkie & Eisdorfer,
1971), although both studies involved only two measure-
ments„baseline and follow-up. The study by Knopman et al.
(2001) involved an impressively large sample of subjects
(n10,963) and risk factors other than hypertension.
Diabetes and incident stroke, as well as hypertension,
were related to greater decline over the six-year study pe-
riod. However, neither of these studies followed subjects
over a signi“cant period of time, and neither involved a
cognitive test battery or a measure of general intellectual
functioning.
To meet these criteria, we need to turn to data published
from the Maine-Syracuse Longitudinal Study of Hyperten-
sion and Cognitive Functioning, which has followed subjects
over a 25-year period (1975 to 2001). This study involved an
extremely comprehensive battery of tests, including the orig-
inal version of the WAIS, as well as signi“cant numbers of
tests from the Halstead-Reitan Battery and the Wechsler
Memory Scale. The mean length of time between waves is
“ve years. The “rst wave of longitudinal data collection with
the “rst cohort took place in Syracuse, New York, in 1981
and 1982. Since then, four additional longitudinal-study
cohorts have entered the study. This is essentially a time-
lagged, cross-sectional, and prospective longitudinal design
(Dwyer & Feinleib, 1992). Longitudinal analyses make use
of the data from serial examinations. Cross-sectional analy-
ses are made possible by pooling data for an examination
across cohorts. Secular trends may be examined by compar-
ing subjects who entered the study at different times.
Multiple studies have evolved from this 25-year project.
Several studies illustrate the use of contemporary longitudi-
nal data analysis methods designed to deal with the prob-
lem of selective attrition, to control for potential confounds
related to comorbidity and hypertension-related diseases,
and to use all available data even though not every subject in
the study has completed the same number of longitudinal
examinations.
M. Elias, Robbins, Elias, and Streeten (1998b) employed
140 relatively healthy men and women taken from a
larger sample of individuals who had completed the WAIS.
Sample size was signi“cantly reduced because they restricted
the sample to persons who (a) completed the WAIS; (b) were
between 40 and 70 years of age at baseline; (c) free from
stroke, dementia, secondary forms of hypertension, and co-
existing diseases; and (d) free from treatment with antihyper-
tensive medications at baseline (M. Elias, Robbins, Elias, &
Streeten, 1998a), using a method of analysis that both ac-
counts for attrition and allows estimation of missing longitu-
dinal data (Willett, 1988).
An important feature of this longitudinal analysis is that it
allows estimates of decline in performance for a given num-
ber of years (e.g., 10 or 20). It does not require that all sub-
jects complete every longitudinal examination as long as at
least two examinations are completed at some point in the
longitudinal study. One signi“cant bene“t of this analysis is
that it adjusts for longitudinal attrition because data for
dropouts are not discarded from the analysis. This data has
been collected for persons who were enrolled in the study
from periods ranging from 5 to 20 years. In this study, the
predictors of decline on the WAIS were (a) ever-never hyper-
tensive status; (b) blood pressure over all examinations (dias-
tolic or systolic); and (c) most importantly, blood pressure at
baseline (examination). Crystallized ability (verbal abilities)
was unrelated to the blood pressure predictor measures, but a
measure of speed (digit symbol substitution) and a composite
measure of ”uid ability (visualisation-performance) were.
Figure 21.1 shows the estimated decline in a ”uid ability
composite score (picture arrangementobject assembly
picture completionblock design) per 20 years of longi-
tudinal study participation for persons de“ned as always-
normotensive or ever-hypertensive. Expressed in percent of
correct scores and adjusted for covariates (age, education,
occupation, anxiety, depression, cigarette smoking, alcohol
consumption), the estimated decline over 20 years was
12.1% greater for persons who were hypertensive at any ex-
amination versus those who were never hypertensive. For
both the ”uid V-P composite (shown in Figure 21.1) and
speed (digit symbol substitution scores), persons who were
hypertensive at baseline exhibited greater longitudinal de-
cline. This “nding with ”uid V-O was observed for each of
the BP predictor variables including untreated diastolic
and systolic blood pressure values at baseline. The higher
the BP, the greater the longitudinal decline in cognitive
functioning. All-exam (averaged) DBP was also associated