Journal of Gerontology: MEDICAL SCIENCES
1997, Vol. 52A, No. 3, M177-M183
In the Public Domain
Age-Associated Changes in Blood Pressure
in a Longitudinal Study of Healthy Men and Women
Jay D. Pearson, Christopher H. Morrell, Larry J. Brant, Patricia K. Landis, and Jerome L. Fleg
Gerontology Research Center, National Institute on Aging, Baltimore.
Background. Current knowledge of age-associated increases in blood pressure is based primarily on unscreened
population studies that may not be representative of healthy men and women. We examined longitudinal patterns of
change in blood pressure in healthy male and female volunteers from the Baltimore Longitudinal Study of Aging (BLSA).
Methods. Longitudinal mixed-effects regression models are used to estimate the age-associated changes in blood
pressure in 1307 men (age 17-97) and 333 women (age 18-93) who have been followed for up to 32 years (mean: 8.4
years for men and 3.4 years for women) and who have been screened for health problems or medications that affect blood
pressure.
Results. On average, systolic pressure is relatively stable in men and women until approximately age 45, increases at
5-8 mm Hg per decade in middle age, then accelerates in men and stabilizes in women. Diastolic pressure increases at 1
mm Hg per decade at all ages in men, whereas in women the rate of change in diastolic pressure increases in middle age
and then plateaus and may decline after age 70. Additional findings include: (a) BLSA cross-sectional and longitudinal
findings are more similar than has been observed in studies of unscreened samples; (b) there is no evidence of a gender
cross-over in this group of healthy men and women; and (c) compared to previous studies of unscreened samples, healthy
BLSA men and women show a weaker association between baseline blood pressure and subsequent rate of blood pressure
change.
Conclusions. These findings suggest that several previously described age-associated patterns of blood pressure
change partially reflect the effects of hypertension and its treatment, rather than intrinsic age changes in the blood pressure
of healthy individuals.
is generally believed that systolic blood pressure inITcreases
sharply with age in most industrialized societies
METHODS
and that diastolic pressure rises modestly and plateaus or
declines after middle age (1-4). A number of studies also
indicate that average systolic and diastolic pressures are
higher in women than men after age 45-55 (1-4). To a great
extent, this view of age-associated differences in blood
pressure has been derived from cross-sectional studies of
populations which included individuals with known hypertension and other cardiovascular diseases. In contrast to
cross-sectional studies, longitudinal data have produced inconsistent results regarding gender- and age-associated
changes in blood pressure (3,5-14). However, the longitudinal studies have varied widely in inclusion criteria and in
control for historical variations in the treatment of hypertension. The fact that the increase in blood pressure with age
does not occur uniformly or universally (15—19) indicates
that an increase in blood pressure with advancing age is not
inevitable. Thus, the age-associated increase in blood pressure observed in unscreened population studies may not be
representative of blood pressure changes in healthy men and
women. Therefore, we have examined age-associated crosssectional and longitudinal trends in blood pressure of 17- to
97-year-old male and female volunteers from the Baltimore
Longitudinal Study of Aging (BLSA), who have been rigorously screened to exclude health problems or medication
usage that might affect blood pressure.
Study population. — Participants are male and female
volunteers in the BLSA, an open-panel multidisciplinary
study of normal human aging that is conducted by the
National Institute on Aging (20). Participants are predominantly White (95%) and well educated (over 75% have a
bachelor's degree or higher). The participants are scheduled to visit the Gerontology Research center at approximately 2-year intervals where they stay for 2.5 days of
evaluation and testing including a physical examination by
a health care provider (i.e., physician, physician's assistant,
or nurse practitioner), inventory of medications used, and
an intensive set of medical, physiological, and psychological evaluations.
Between 1958 and 1991, 1,478 men and 645 women had
blood pressure measurements taken during one or more of
their BLSA visits. Because blood pressure can be affected by
various diseases or medications, we censored the data of 231
men and 97 women for visits following the diagnosis of any
of the following conditions: probable or definite myocardial
infarction, angina pectoris, or Q-wave abnormalities on
resting ECG (Minnesota Code 1:1 or 1:2); stroke; aortic
valvular disease; untreated thyroid disease; renal disease;
diabetes; or congestive heart failure. Blood pressure measurements were also excluded at visits where the participants
were taking the following medications: diuretics, antihyperM177
PEARSONETAL.
M178
tensive or cardiovascular medications, bronchodilators, insulin, oral hypoglycemic agents, nonsteroidal antiinflammatory drugs, central nervous system stimulants (2005 visits in
688 men; 976 visits in 420 women), or had other conditions
that might affect blood pressure (hemoglobin 2= 17.0 g or
serum creatinine s* 1.5 mg/dL; 206 visits in 154 men, no
visits in women). In addition, any blood pressure measurements taken within 2 years of death (49 visits in 44 men and
5 visits in 5 women) were excluded to avoid any bias due to
occult diseases that might have affected blood pressure. The
censoring and visit exclusions resulted in the exclusion of all
data for 441 men and 312 women. Thus, the final screened
male study group consists of a total of 4,422 observations on
1,037 men (70% of all BLSA men) whose beginning age in
the study was between 17.2 and 96.7 years, and the screened
female study group consists of a total of 749 observations on
333 women (52% of all BLSA women) whose beginning age
in the study was between 18.6 and 92.7 years (Table 1).
Procedures. — On the first day of each biennial visit,
BLSA volunteers undergo an extensive medical history and
physical examination. A seated blood pressure is taken in
both arms with a manual sphygmomanometer appropriately
sized to the arm of the participant. Systolic and diastolic
pressure are defined by the first and fifth Korotkoff sounds,
respectively. We use the mean of the blood pressures from
the two arms at each visit. In males, we use biennial physical
examination blood pressure measurements collected from
1958 to 1991. Because women did not enter the BLSA until
1978, we use measurements collected from 1978 to 1991 for
women. Height and weight are measured in undergarments
and without shoes. Body mass index (BMI = weight in kg/
height in m2) is calculated from anthropometric measurements taken at each visit. Individuals are classified as current
cigarette smokers if they report smoking any cigarettes on a
regular basis at that visit.
Statistical methods. — The mean blood pressure at baseline visit ( ± 9 5 % confidence intervals) is reported for each
5-year age group in all BLSA participants and in the
screened group to determine the baseline comparability of
the two groups.
Linear mixed-effects regression models are used to analyze the longitudinal data (21,22). Mixed-effects models
allow estimation of the group-average blood pressure curve
and also allow each subject's estimated longitudinal change
to deviate from the group average. Details of the linear
mixed-effects model are provided in the Appendix.
To determine whether the longitudinal 10-year rate of
change in blood pressure differs depending on baseline
blood pressure level, we use mixed-effects models to predict
the change in blood pressure levels. Linear regression
models are then used to examine the association between
estimated 10-year rate of change in blood pressure and
baseline estimated blood pressure level in individuals with at
least 6 years of follow-up (n = 462 men and 88 women). A
two-tailed p-value of less than .05 was required for statistical
significance in all analyses.
RESULTS
The number of participants is fairly evenly distributed
across the age range, although the number of participants
over age 80 is limited (Table 1). On average, men have
longer follow-up and more visits than women because
women were not recruited into the BLSA until 1978. Approximately 50% of the men have 5 or more years of followup and 30% have 10 or more years. Approximately 30% of
the women have 5 or more years of follow-up and 11 % have
10 or more years. The screened groups represent a total of
8,711 person-years of follow-up in the men and 1,132
person-years of follow-up in the women. There is no evidence of selective attrition from the BLSA with respect to
blood pressure. Blood pressures are not significantly differ-
Table 1. Characteristics of Longitudinal Follow-up in the Screened BLSA Group
Age at First Visit
30-39.9
40-49.9
50-59.9
60-69.9
70-79.9
80 +
Total
Group
166
234
204
166
128
115
24
1037
7.9
(0,25.5)
10.4
(0,32.5)
12.0
(0,32.4)
8.2
(0,30.7)
5.7
(0,21.9)
3.7
(0,19.0)
2.7
(0,13.4)
8.4
(0,32.5)
3.6
5.5
4.4
(1,17)
2.4
(0.8,19.4)
3.7
(1,14)
1.8
(0.9,15.3)
3.2
(1,11)
1.7
(0.8,8.0)
2.5
(1,10)
3.0
(0.9,15.9)
4.5
(1,18)
2.9
(0.9,19.4)
4.3
(1,20)
2.6
(0.8,20.9)
48
1.8
(0,12.3)
68
3.6
(0,12.4)
55
4.1
(0,12.1)
1.6
(1,4)
2.8
(1.7,7.2)
2.1
(1,7)
3.3
(1.8,10.0)
61
4.2
(0,12.5)
2.6
(1,7)
<30
Men
n
Length of
follow-up (yr)
Number of visits
(1,16)
Interval between
visits (yr)
Women
n
Length of
follow-up (yr)
Number of visits
Interval between
visits (yr)
(1,20)
2.6
(0.8,20.9)
39
4.7
(0,13.1)
2.4
(1,6)
3.3
(1.0,13.1)
2.6
(0.9,8.0)
Notes: Values shown are means; values in parentheses are minima and maxima.
2.7
(1,6)
2.5
(1.3,7.7)
1.9
(1.0,9.4)
44
18
2.4
(0,12.9)
2.2
(1,8)
2.0
(0.7,4.3)
1.5
(0,5.9)
1.7
(1,4)
2.1
(1.0,4.0)
333
3.4
(0,13.1)
2.2
(1,8)
2.7
(0.7,13.1)
AGE-ASSOCIATED CHANGES IN BLOOD PRESSURE
ent in individuals who died or withdrew from the BLSA
compared to participants who remained active (p = .89 and
p = .91, respectively, for systolic and diastolic pressure in
men; p = .98 and p = .49, respectively, in women). An
analysis to identify birth cohort effects did not identify any
consistent effects (data not shown).
Table 2 shows the proportion of participants who were
hypertensive at their first examination and the distribution of
changes in blood pressure observed over approximately 10
years (mm Hg per decade assessed at the examination nearest
to 10 years after first exam). The proportion of hypertensives
increases with age in both men and women but is lower in
women in each age group. With increasing age, the proportion of men who exhibit increases in systolic blood pressure
greater than 5 mm Hg per decade increases from 34% for men
aged less than 40, to 48% for men aged 40-59.9, and 54% for
men aged 60 and above, whereas 35^40% of women in each
age group exhibit such increases (Table 2). For diastolic
blood pressure, 35-45% of men and women exhibit increases
greater than 5 mm Hg per decade regardless of age. However,
men and women age 60 and above both show an increased
proportion of individuals with decreases greater than 10 mm
Hg per decade compared to younger ages.
Effect of screening. — Figure 1 shows the means ( ± 95%
confidence intervals) of the observed blood pressure levels at
baseline visit by 5-year age groups for all BLSA participants
and for rigorously screened men and women. The confidence intervals in Figure 1 and subsequent figures are wider
in women than men because of the smaller number of
observations in women. The cross-sectional results from the
screened and unscreened data sets are nearly identical for
diastolic pressure. However, systolic pressure is approximately 3-5 mm Hg higher in the unscreened men and
women after age 60. Systolic pressure increases with crosssectional age in both the screened and unscreened groups.
Cross-sectionally, diastolic pressure appears to plateau be-
Ml 79
tween ages 50 and 60 in men and at approximately age 60 in
women. Although gender differences in systolic and diastolic pressure narrow with age, there is no evidence of a gender
cross-over in either the screened or unscreened data sets.
Cross-sectional vs longitudinal age effects. — Figure 2
and Table 3 show the systolic and diastolic pressure levels
and longitudinal changes over 10 years of follow-up as
estimated by the linear mixed-effects models. The blood
pressure levels shown at the beginning of each 10-year
longitudinal curve represent the cross-sectional estimates for
different ages at entry into the BLSA. The cross-sectional
and longitudinal results for both systolic and diastolic pressure are not significantly different in either sex, except for
systolic pressure in men at ages 40 and 70 (Figure 2, Table
3). Forty-year-old men have higher systolic pressures at their
initial visit than 30-year-old men who are followed longitudinally to age 40. However, blood pressures in these same
men measured in the supine position at first awakening do
not reveal a longitudinal decline in systolic pressure in 30year-olds (data not shown). The lower initial systolic and
diastolic pressures in the 70-year-old group of men compared to 60-year-old men who were followed to age 70
probably represent recruiting bias or selective survival. Men
who join the BLSA at older ages tend to be healthier than
men who joined at younger ages and who reach a similar age
while in the study (23). Because of the shorter follow-up in
women, this recruiting bias is less evident in female BLSA
participants.
Prior to age 45-50, systolic pressure increases with age
cross-sectionally but is relatively stable longitudinally in
both sexes (Figure 2, Table 3). The longitudinal increase in
systolic pressure is approximately 5 mm Hg per decade in
both men and women at age 50. After age 50, the longitudinal rate of increase in systolic pressure accelerates to over 8
mm Hg per decade in men, whereas in women the rate of
Table 2. Observed Distribution of Hypertension at First Examination and Changes
in Blood Pressure Levels (mm Hg/decade) for BLSA Men and Women
Men
Percent hypertensive at first
exam (systolic > 140 or
diastolic > 90 mm Hg)
40-59.9
Age 2= 60
Age < 40
40-59.9
Age & 60
14%
(58/400)
26%
(96/370)
41%
(109/267)
1%
(1/116)
8%
(8/100)
32%
(37/117)
n = 89
18%
8
20
6
48
n = 24
14%
12
34
18
22
n = 42
12%
16
37
15
20
n = 30
19%
8
33
11
29
33%
13
10
10
33
4%
10%
10
36
24
21
20%
3
40
13
23
Percent exhibiting change in blood pressure level (mm Hg/decade)
Systolic
n = 231
n = 215
> 10 decline
28%
21%
5.1 to 10 decline
9
8
5 decline to 5 increase
30
23
5.1 to 10 increase
12
13
>10 increase
22
35
Diastolic
> 10 decline
5.1 to 10 decline
5 decline to 5 increase
5.1 to 10 increase
>10 increase
Women
Age < 40
21%
8
29
12
29
12%
7
38
7
36
8
50
4
33
PEARSON ETAL.
M180
(a) Unscreened
(b) Screened
140
Men
Men
Women
o
I I
120 •
I I
l
l
.'ill
1
. ISystolic
9
^ * ^ ^
Women
E
ES
^zgg&P^&SZ^r*
Oiastolic
80
I
? s ? f "
fin
20 30 40 50 60 70 80
Age
I
3
^ ^ ^
Systolic Blood Pressure
100
I
f ^ ^ V s ^ ^ ^ ? ^ ^ ^
Oiastolic Blood Pressure
B0
100 •
)d Pressi
1
o
,i
20 30 40 50 60 70 80
Age
Figure 1. Observed cross-sectional average (± 95% confidence interval)
systolic and diastolic blood perssures for (a) the unscreened BLSA data set
and (b) the BLSA data set screened for health problems or medications that
affect blood pressure. Circles represent the mean blood prressure for
women, and triangles represent the mean blood pressure for men.
CQ
o
to
30
40
50
60
70
80
Age
Figure 2. Average (with 95% confidence intervals) systolic and diastolic
blood pressure levels estimated from the mixed-effects model for 10-year
longitudinal follow-up in healthy BLSA men and women.
Table 3. Blood Pressure Levels (mm Hg) and 10-yr Change in Blood Pressure (mm Hg/decade)
for BLSA Men and Women as Estimated from the Most Parsimonious Mixed-Effects Models
Systolic
Diastolic
Follow-up
Duration (yr)
Men
Women
p-value
Men
Women
p-value
30
Level at baseline
Level at 5
Level at 10
10-yrchange
118.3
116.4
115.8
-2.5*
105.9
105.4
105.6
-0.4
.000
.000
.004
.549
76.3
76.9
77.4
1.1*
70.5
69.8
71.9
1.4
.000
.000
.122
.855
40
Level at baseline
Level at 5
Level at 10
10-yr change
120.5
120.1
120.9
0.4
109.5
108.5
113.7
4.1
.000
.000
.003
.102
78.6
79.2
79.7
1.1*
72.4
71.6
76.2
3.8*
.000
.000
.114
.061
50
Level at base-line
Level at 5
Level at 10
10-yr change
123.8
126.3
128.7
4.9*
115.3
114.6
121.1
5.7*
.000
.000
.002
.742
79.9
80.5
81.0
1.1*
74.0
73.4
76.7
2.8*
.000
.000
.011
.167
60
Level at baseline
Level at 5
Level at 10
10-yrchange
127.9
132.6
136.5
8.5*
121.7
122.0
126.1
4.4*
.000
.000
.000
.091
80.3
80.9
81.4
1.1*
75.1
74.5
75.3
0.2
.000
.000
.005
.538
70
Level at baseline
Level at 5
Level at 10
10-yrchange
132.7
137.0
141.3
8.6*
127.1
129.3
127.1
0.1
.001
.000
.000
.029
79.8
80.3
80.9
1.1*
75.5
74.2
73.5
-2.0
.000
.000
.068
.105
Initial
Age
•Longitudinal rate of change is significantly different from zero (p < .05).
increase slows to 4 mm Hg per decade at age 60 and is
virtually flat after age 70.
Diastolic pressure increases longitudinally at a constant
rate of 1 mm Hg per decade in men aged 30-80 (Table 3).
However, in women, diastolic pressure changes minimally
before age 40, increases at a rate of 3-4 mm Hg per decade
during middle age, then plateaus at age 60, followed by
evidence of a decline after age 70.
Gender differences. — Both cross-sectionally and longitudinally, women have significantly lower systolic and dias-
tolic pressures than men at all ages (Figure 2, Table 3). The
gender difference in systolic pressure narrows from approximately 12 mm Hg at age 30 to approximately 6 mm Hg at age
70, but there is no gender cross-over. The gender difference
in diastolic pressure narrows only slightly from 5.8 mm Hg
at age 30 to 4.3 mm Hg at age 70. Use of estrogen-containing
medications (i.e., estrogen replacement therapy, contraceptives) does not significantly affect systolic or diastolic pressure in BLSA women overall (p = .59 for systolic and p =
.73 for diastolic), or in women younger than 45 years (p =
.87 for systolic and/? = .58 for diastolic), women aged 4 5 -
AGE-ASSOCIATED CHANGES IN BLOOD PRESSURE
54.9 years (p = .45 for systolic and/? = .43 for diastolic),
or women 55 years or older (p = .96 for systolic and p =
.40 for diastolic).
Baseline blood pressure and subsequent rate of change.
— In these healthy individuals, baseline blood pressure
levels accounted for less than 5% of the variance in 10-year
rate of change in blood pressure, except for systolic pressure
in men. Figure 3 shows that high baseline systolic pressures
tend to be associated with more rapid increases in systolic
100
CD
~O CO CO
o
CD
!
120
140
160
Beginning Systolic Blood Pressure (mm Hg)
40- 60 Year Old Men
•
•
«
•
#
m
CO
• 1
_c
CD
c
•
»*
*r*»0*'
•
•
•
: :
CO
6
100
120
140
160
Beginning Systolic Blood Pressure (mm Hg)
60 - 80 Year Old Men
100
120
140
160
Beginning Systolic Blood Pressure (mm Hg)
Figure 3. Relationship between initial systolic blood pressure and 10-year
rate of change in systolic blood pressure in healthy BLS A men stratified by
age (total R2 = .53, partial R2 = . 14 for initial systolic blood pressure).
M181
pressure in men of all ages. Baseline systolic pressure
accounts for 14% of the variance in rate of change in systolic
pressure. In men, each 10 mm Hg elevation in baseline
systolic pressure is associated with approximately a 2 mm
Hg per decade increase in the 10-year rate of change in
systolic pressure, regardless of age.
DISCUSSION
Previous unscreened population studies have been extremely useful in understanding the medical and public
health ramifications of the associations between chronologic
age and blood pressure. However, the findings from unscreened samples may not be generalizable to healthy individuals. The present study has the advantages of careful
screening to exclude individuals with diseases or medication
usage that may affect the age-associated trends in blood
pressure and a long-term longitudinal design across a broad
age range in both sexes. Thus, these findings provide a
clearer description of age-associated changes in blood pressure in these healthy BLSA men and women without the
confounding influences of diseases or medications known to
affect blood pressure. Additional findings include: (a) BLSA
cross-sectional and longitudinal findings are more similar
than has been observed in studies of unscreened samples; (b)
there is no evidence of a gender cross-over in this group of
healthy men and women; and (c) compared to previous
studies of unscreened samples, healthy BLSA men and
women show a weaker association between baseline blood
pressure and subsequent rate of blood pressure change.
Age effects. — The BLSA cross-sectional and longitudinal findings are generally quite similar. On average, systolic
blood pressure is relatively stable in healthy men and women
until middle age (approximately age 50 in men and age 40 in
women), then increases by 5-8 mm Hg per decade. Women
may show some longitudinal evidence of a plateau or decline
in systolic blood pressure after age 70, but men do not.
Diastolic pressure increases steadily in healthy BLSA men at
a rate of 1 mm Hg per decade with no sign of a plateau or
decline. Diastolic pressure in women is relatively stable
between ages 30 and 40, increases significantly between
ages 40 and 60, but then stabilizes and may decline slightly
after age 70. Overall, systolic pressure increases about 15
mm Hg in men and 21 mm Hg in women between ages 30
and 70 years, whereas diastolic pressure increases only 3.5
mm Hg in men and 5 mm Hg in women over this interval.
Thus, these healthy BLSA men do not show any sign of a
plateau or decline in systolic or diastolic pressure, whereas
women exhibit a plateau for both systolic and diastolic
pressure. This contrasts with other longitudinal results in
unscreened samples in which systolic pressure did not plateau over time in either men or women, whereas diastolic
pressure exhibited a plateau followed by a decline in both
men and women (3); both systolic and diastolic pressure
declined in men and women aged 65-98 years (5), or
systolic pressure declined after age 65 in both men and
women primarily due to secular trends in treatment of
hypertension (24).
The similarity between the BLSA cross-sectional and
longitudinal findings probably reflects the health and medi-
Ml 82
PEARSON ETAL.
cation screening in our study. The screening minimizes the
confounding effects of age-associated differences in the
incidence and prevalence of hypertension and its treatment.
Furthermore, our analyses do not reveal any significant
selective attrition with respect to blood pressure in these
healthy men and women. In contrast, the Framingham study
(3) shows a more marked difference between cross-sectional
and longitudinal results, perhaps because their results are
based on an unscreened sample in whom selective mortality
associated with hypertension would be expected to have a
greater impact. Other studies have observed significant cohort differences in systolic and diastolic pressure in men and
women over age 65 (5).
Gender difference. — BLSA women have lower blood
pressures than men across the adult age span and do not
exhibit the gender cross-over in systolic pressure which has
been observed in most cross-sectional studies. In this study,
longitudinal rates of change in blood pressure are similar
between sexes. These findings are not affected significantly
by use of estrogen-containing medications, although we
could not control for dosage or individual medications.
Diastolic pressure is approximately 4-6 mm Hg lower in
women than in men at all ages, whereas the gender difference in systolic pressure shrinks from approximately 11 mm
Hg lower for women at age 40 to only 5.5 mm Hg lower at
age 70. This is similar to longitudinal results reported from
the Framingham study, where the mean systolic pressures of
men and women converge but do not cross over, and the ageassociated trends for diastolic pressure are roughly parallel
in men and women (3). The gender cross-over observed by
others in unscreened samples may be related to a combination of selective mortality in men with elevated blood pressure, more aggressive treatment of hypertension in men, and
perhaps age and cohort differences in the treatment of
hypertension.
Baseline blood pressure and subsequent rate of change.
— A number of studies in unscreened samples have reported
a significant positive association between baseline blood
pressure levels and subsequent rates of blood pressure
change (9-12). In healthy BLSA men and women, baseline
blood pressure level explains little of the variance in 10-year
rate of change in blood pressure, except for a modest
relationship for systolic pressure in men. The weaker association in these healthy volunteers suggests that the accelerated rise in blood pressure in men with higher initial pressures may reflect the progression of hypertension rather than
an aging effect per se.
Between-subject variability. — Although average patterns
can be detected, there is a significant degree of betweenperson variability in both blood pressure level and longitudinal pattern of change that makes it difficult to extrapolate the
average trends to individual subjects. At any age, many
individuals show little or no longitudinal increase in systolic
or diastolic pressure. Thus, age-associated increases in blood
pressure are neither universal nor inevitable.
Study limitations. — Although longitudinal studies allow
a direct estimation of changes in blood pressure, they are still
subject to possible confounding due to cohort effects, secular trends, selective attrition, and other factors. However,
examination of the BLSA data revealed no evidence of these
biases. Although our efforts to screen for medical problems
that affect blood pressure might have missed some occult
disease, this is unlikely to seriously bias the results because
similar findings are obtained in the unscreened data. Since
the BLSA study group is predominantly White and middle to
upper class, these results may not be generalizable to minority or lower socioeconomic groups. Lastly, BLSA participants are health-conscious volunteers who may be more
likely to take an active role in prevention and intervention
efforts than the general population.
Conclusions
These findings demonstrate that in a group of male and
female volunteers screened to exclude diseases and medication use that affect blood pressure, cross-sectional and longitudinal blood pressure patterns are more similar than is
usually observed in unscreened population studies. In further contrast to previous studies in unscreened populations,
these healthy volunteers have no gender cross-over in blood
pressure and have only a weak association between baseline
blood pressure and subsequent rate of pressure change.
These findings suggest that several of the previously described age-associated patterns of blood pressure change
partially reflect the effects of hypertension and its treatment,
rather than intrinsic blood pressure changes in healthy individuals as they age.
ACKNOWLEDGMENTS
We wish to acknowledge the contributions of Dr. James L. Fozard,
associate scientific director for the BLSA; Dr. E. Jeffrey Metter, medical
officer for the BLSA; and all the BLSA clinical staff who collected the data
used in this study.
Address correspondence to Dr. Jay D. Pearson, Department of Epidemiology, Merck Research Laboratories, BL2-3, P.O. Box 4, West Point, PA
19486-0004. E-mail: [email protected]
REFERENCES
1. Johnson BC, Epstein FH, Kjelsberg, MO. Distributions and familial
studies of blood pressure and serum cholesterol levels in a total
community — Tecumseh, Michigan. J Chronic Dis 1965; 18:147-60.
2. Eilertsen E, Humerfelt S. The blood pressure in a representative
population sample. Act Med Scand 1968; 183:293-305.
3. Kannel WB, Gordon T. Evaluation of cardiovascular risk in the
elderly: the Framingham study. Bull NY Acad Med 1978;54:573-91.
4. Harlan WR, Hull AL, Schmouder RL, Landis JR, Thompson FE,
Larkin FA. Blood pressure and nutrition in adults: the National Health
and Nutrition Examination Survey. Am J Epidemiol 1984; 120:17-28.
5. Bush TL, Linkens R, Maggi S, Hale WE. Blood pressure changes with
aging: evidence for a cohort effect. Aging 1989; 1:39-45.
6. Harlan WR, Osborne RK, Graybiel A. A longitudinal study of blood
pressure. Circulation 1962;26:530-43.
7. Bjerkedal T, Natvig H. Changes in blood pressure with age. Acta Med
Scand 1966; 180:257-72.
8. Oberman A, Lane NE, Harlan WR, Graybiel A, Mitchell RE. Trends
in systolic blood pressure in the thousand aviator cohorts over a twentyfour-year period. Circulation 1967;36:812-22.
9. Miall WE, Chinn S. Blood pressure and ageing: results of a 15-17 year
follow-up study in South Wales. Clin Sci Mol Med 1973;45:23s-33s.
AGE-ASSOCIATED CHANGES IN BLOOD PRESSURE
10. Hsu PH, Mathewson FAL, Rabkin SW. Blood pressure and body mass
index patterns — a longitudinal study. J Chronic Dis 1977;30:93—113.
11. Svardsudd K, Tibblin G. A longitudinal blood pressure study: change
of blood pressure during 10 yr in relation to initial values: the study of
men born in 1913. J Chronic Dis 1980;33:627-33.
12. Sparrow D, Garvey AJ, Rosner B, Thomas HE Jr. Factors in predicting
blood pressure change. Circulation 1982;65:789-94.
13. Sigurdsson JA. High blood pressure in women: a cross-sectional and
longitudinal follow-up study. Acta Med Scand Suppl 1983;669:1-39.
14. Landahl S, Bengtsson C, Sigurdsson JA, Svanborg A, Svardsudd K.
Age-related changes in blood pressure. Hypertension 1986;8:1044-9.
15. Stamler J, Lindberg HA, Benson DM, Shaffer A, Miller W, Poindexter A. Epidemiological analysis of hypertension and hypertensive
disease in the labor force of a Chicago utility company. Hypertension
1958;7:23-50.
16. Prior I AM, Evans JG, Harvey HPB, Davidson F, Lindsey M. Sodium
intake and blood pressure in two Polynesian populations. N Engl J Med
1968;279:515-20.
17. Truswell AS, Kennelly BM, Hansen JDL, Lee RB. Blood pressures of
the !Kung Bushmen in Northern Botswana. Am Heart J 1972,84:5-12.
18. Oliver WJ, Cohen EL, Neel JV. Blood pressure, sodium intake, and
sodium-related hormones in the Yanomamo Indians, a "no-salt"
culture. Circulation 1975;52:146-51.
19. Friedlaender JS, Page LB. Population biology and aging: the example
of blood pressure. Am J Hum Biol 1989; 1:355-65.
20. Shock NW, Greulich RC, Andres R, Lakatta EG, Arenberg D, Tobin
JD. Normal human aging: The Baltimore Longitudinal Study of
Aging. NTH pub. no. 84-2450. Washington, DC: U.S. Government
Printing Office, 1984.
21. Laird NM, Ware JH. Random effects models for longitudinal data.
Biometrics 1982;38:963-74.
22. Lindstrom MJ, Bates DM. Newton-Raphson and EM algorithms for
linear mixed-effects models for repeated measures data. J Am Stat
Assoc 1988;83:1014-22.
23. Metier EJ, Walega D, Metter EL, Pearson JD, Brant LJ, Hiscock BS,
Fozard JL. How comparable are healthy 60- and 80-year-old men? J
Gerontol Med Sci 1992;47:M73-8.
24. Glynn RJ, Field TS, Satterfield S, Hebert PR, Buring JE, Taylor
JO, Hennekens CH. Modification of increasing systolic blood pressure in the elderly during the 1980's. Am J Epidemiol 1993; 138:
365-79.
25. Draper NR, Smith H. Applied regression analysis. New York: John
Wiley and Sons, 1981:257-62.
26. Peixoto JL. A property of well-formulated polynomial regression
models. Am Stat 1990;44:26-30.
Received February 23, 1996
Accepted October 2, 1996
Appendix
Linear mixed-effects regression models are used to analyze these
longitudinal data (21,22). Mixed-effects models allow estimation
of the group-average blood pressure curve and also allow each
subject's estimated longitudinal change to deviate from the group
average.
In the mixed-effects model, the fixed-effects estimate the average intercept and rates of change for the independent variables,
while the random effects represent the deviation for each individual
from the average intercept and slope terms. Thus, the random
effects account for natural heterogeneity in initial level and patterns
of change among the individuals in the study. Mixed-effects
models account for correlation among repeated measurements
within individuals and allow for the analysis of unbalanced data
M183
where individuals have differing numbers of observations taken at
varying intervals between the observations. In fact, individuals
with even a single observation can be included, although they
contribute only to the cross-sectional estimates and not the longitudinal estimates.
Arranging the terms in the mixed-effects model to see how the
longitudinal change depends upon first age, the full model for
systolic and diastolic pressures on the i* subject at timey is:
= C3o + bm) + (Pi + bn) timeu + fp2 + bn) tinted +
+ frfage2, + %fage\ + [fafage, + frfage2, + & fage*A timev
+ [&fage, + frofage2, + frjage*,] time\
+ p i 6 BMI4U + p 17 BMI1V x Fcig, + p,, BMB« X Fcig,
+ 3, 9 BMI4y X Fcig, + Pa, Status, + p 2I Status, X timeu
+ P22 Sexhornty + £<,-,
where longitudinal change is represented by follow-up time (time
and time2), and cross-sectional differences are represented by
polynomial terms in age at first visit (fage, fage2, and fage3).
Interaction terms are included to allow the longitudinal patterns of
change to differ with age at entry (interactions between powers of
fage x powers of time). Several variables are included to adjust
statistically for the effects of body mass index, cigarette smoking,
and the use of estrogen-containing sex hormones. BMIJ represents
lean people (BMI < 20 kg/m2), BMI3 are overweight people (25 <
BMI < 30 kg/m2) and BMI4 is a variable for obese people (BMI >
30 kg/m2). Fcig is a variable for cigarette smoking at the first visit
whereas Cigcur indicates whether or not they are currently smoking at each visit. Status shows whether the person had withdrawn
from the study which tests for selective attrition. Sexhorm indicates
whether the woman was taking sex hormones. The full model for
men excludes the Sexhorm term.
Three random effects (bm, bn, bn) are inclueed in the full model
to account for natural heterogeneity among individuals with respect
to blood pressure level (intercept), and longitudinal change (time
and time2). Thus, each person's longitudinal systolic and diastolic
blood pressure patterns have level and shape that deviate from the
overall average.
In order to reduce the multicollinearity among the polynomial
and interaction terms, the follow-up time and first age variables are
centered on the mean follow-up time and age at first visit by
subtracting 4 and 52 years respectively from time and fage for men
and 1.5 and 55 years respectively for women (25). The most
parsimonious well-formulated models (26) are obtained by backward elimination of the highest-order nonsignificant polynomial
and cross-product terms. Terms are retained in the model if they
reach a significance level of 0.05. We use FORTRAN software
developed by Mary Lindstrom which uses a Newton-Raphson
algorithm to obtain restricted maximum likelihood estimators for
the mixed-effects models and assumed an unstructured variancecovariance matrix for the random effects (22).
The final models for men included fixed-effects terms 0-4, 1416 and random-effects terms 0-2 for diastolic pressure and fixedeffects terms 0-11, 14-16 and random-effects terms 0-2 for
systolic pressure. The final models for women included fixedeffects terms 0-11, 15-16 and random-effects terms 0-2 for
diastolic pressure and fixed-effects terms 0-10, 12, 15-16, 19 and
random-effects terms 0-1 for systolic pressure. Parameter estimates and the estimated standard errors for the final models are
available from the authors upon request.