American Journal of Epidemiology
Copyright © 1998 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
Vol. 148, No. 6
Printed In U.SA.
ORIGINAL CONTRIBUTIONS
Systolic Blood Pressure Trends in US Adults between 1960 and 1980
Influence of Antihypertensive Drug Therapy
Shiriki K. Kumanyika, 12 J. Richard Landis,1'3 Yvonne L Matthews-Cook,1 Susan L. Aimy,1-4 and
Susan J. Shirk Boehmer1
Recent blood pressure trends reflect progress in hypertension control, but prevalent drug therapy precludes
direct estimation of the component due to primary prevention. In data gathered on persons aged 35-74 years
in three successive US health examination surveys (1960-1980), systolic blood pressure levels assuming no
drug therapy were imputed by reassigning blood pressure to the upper end of the distribution for respondents
reporting use of antihypertensive medication. Blood pressure was partitioned into four ordinal categories
based on weighted percentiles of the 1960-1962 distributions for 35- to 44-year-old males and females who
reported no use of antihypertensive medication. Cumulative loglt models (a = 0.01) were used to estimate ageand sex-specific trends for blacks and whites within two strata (<25 or ^25) of body mass index (BMI) (weight
(kg)/height (m)2). Before imputation, systolic blood pressure decreased between 1960 and 1980; after imputation, significant decreases remained only in 35- to 44-year-olds. Strong associations of black race and BMI
a25 with higher blood pressures were present in models with and without drug therapy. Thus, according to
the models, there has been little progress in decreasing racial or BMI-related blood pressure differentials.
Above the age of 44 years, blood pressure trends were largely attributable to medication use. In contrast, data
for 35- to 44-year-olds suggest progress in primary prevention. Am J Epidemiol 1998; 148:528-38.
antihypertensive agents; blacks; blood pressure; body weight; ethnic groups; hypertension; obesity
from antihypertensive therapy is obscuring a lack of
progress in the primary prevention of hypertension.
Decreases due to drug therapy could be occurring
while underlying blood pressure trends are stable or
even rising.
In a previous analysis of secular trends in the blood
pressures of 18- to 34-year-olds (2), we observed a
decline in systolic blood pressure among males and
among females with a body mass index (BMI) (weight
(kg)/height (m)2) less than 25. This finding indicated
success in primary prevention, because the small number of persons (<1.5 percent) reporting antihypertensive medication use were excluded from the analysis
sample. The prevalence of antihypertensive drug treatment at ages 35 and over is higher, however, and
excluding persons on medication risks introducing a
substantial bias towards lower blood pressures as the
prevalence of drug therapy increases. That is, individuals taking antihypertensive medications presumably
had higher blood pressure levels initially than those
not taking such drugs. Furthermore, the bias would
be greatest for those most likely to be treated for
hypertension—i.e., persons in high risk groups such as
blacks or the overweight.
Due in large part to the success of efforts to identify
persons with high blood pressure and refer them to
physicians for appropriate treatment, the majority of
persons with hypertension are aware of this condition,
and at least half of such individuals are being treated
with antihypertensive medication (1). This widespread
treatment for hypertension may be enhancing a secular
decline in blood pressure due to primary prevention of
blood pressure elevation. However, it is also possible
that any decrease in blood pressure levels resulting
Received for publication June 17, 1996, and accepted for publication March 4, 1998.
Abbreviations: BMI, body mass index; NHES I, National Health
Examination Survey, cycle I; NHANES I, First National Health and
Nutrition Examination Survey; NHANES II. Second National Hearth
and Nutrition Examination Survey.
1
Center for Biostatistics and Epidemiology, The Pennsylvania
State University College of Medicine, Herahey, PA.
2
Current affiliation: Department of Human Nutrition and Dietetics, College of Health and Health Development Sciences, University
of Illinois at Chicago, Chicago, IL
3
Center for Clinical Epidemiology and Biostatistics, University of
Pennsylvania School of Medicine, Philadelphia, PA.
4
Current affiliation: Statistics Unlimited, Inc., Westford, MA.
Reprint requests to Dr. Shiriki K. Kumanyika, MC 517, University
of Illinois at Chicago, 1919 West Taylor Street, Room 650, Chicago,
IL 60612-7256.
528
Drug Therapy and Systolic Blood Pressure Trends
As an alternative approach to evaluating blood pressure trends without considering drug therapy, we used
an imputation method that involved reassigning the
blood pressure of anyone reporting the use of antihypertensive medication to the upper end of the distribution. We applied this approach to statistical models
of blood pressure trends for 35- to 74-year-old males
and females in three successive US national health
examination surveys conducted between 1960 and
1980. Substantial increases in hypertension awareness
and in the prescription of antihypertensive medications
occurred in the United States during this period (3).
We estimated the effects of drug treatment indirectly
by examining the statistical significance of time trends
before and after the imputation.
MATERIALS AND METHODS
Population and sample
Data used for these analyses were obtained from the
National Health Examination Survey, cycle I (NHES
I), conducted from 1960 to 1962, and the first and
second National Health and Nutrition Examination
Surveys (NHANES I and NHANES II), conducted,
respectively, from 1971 to 1975 and 1976 to 1980.
Different cross-sectional probability samples were
drawn from each survey using multistage, complex
sample designs that included provision for oversampling to improve the reliability of estimates for certain
population subgroups (4-6). Nationally representative
estimates for the US civilian, noninstitutionalized population can be obtained from these data by incorporating sample weights reflecting different probabilities of
selection and differential nonresponse patterns for various subgroups.
These analyses included all respondents with race
coded as black or white who were between the ages of
35 and 74 years at the time of interview, with the
exception of persons with missing data on blood pressure, medication status, weight, or height and females
who were pregnant at the time of examination. The
numbers excluded were minimal: for males, 29 respondents from the NHES I sample, 19 from the
NHANES I sample, and 25 from the NHANES H
sample (0.78 percent of a total sample of 9,328); for
females, 78 respondents from the NHES I sample, 52
from the NHANES I sample, and 112 from the
NHANES H sample (2.1 percent of a total of 11,428).
The final unweighted analysis sample sizes were 1,874
and 2,128 for males and females, respectively, in
NHES I; 3,612 and 4,886 for males and females in
NHANES I; and 3,769 and 4,172 for males and females in NHANES II. Across the three surveys, the
sample included 8,053 white males, 9,565 white females, 1,202 black males, and 1,619 black females.
Am J Epidemiol
Vol. 148, No. 6, 1998
529
Measurements
Blood pressure was measured with mercury sphygmomanometers by trained observers using standardized protocols that were similar, although not identical, across surveys (7). More than one blood pressure
measurement was taken for respondents in NHES I
and NHANES n, but only data from the initial, seated
blood pressure measurement were comparable across
all three surveys. The initial measurement was thus
selected for these analyses. Height and weight were
measured in a consistent manner across surveys and
were recorded automatically (8-10). The weight of the
standardized clothing worn in NHES I (2.0 pounds
(0.91 kg) (8)) differed from that of the other two
surveys (0.5 pounds (0.23 kg) (9, 10)). The difference
(1.5 pounds (0.68 kg)) was subtracted from each
NHES I respondent's weight. A cutpoint of BMI <25,
the upper limit of recommended weight in several
published weight classifications (11), was used to differentiate overweight and non-overweight strata.
Each survey included a question that could be used
to classify persons according to antihypertensive medication use. NHES I respondents were considered to be
taking anithypertensive medication if they reported
having been told that they had high blood pressure and
answered "yes" to a question on whether they were
taking high blood pressure medication (4). NHANES I
respondents were asked whether they had used "any
medicine, drugs, or pills for high blood pressure"
during the past 6 months (5). Persons who responded
that they used such substances "regularly" or "occasionally" were classified as taking medication.
NHANES II respondents were considered to be taking
antihypertensive medication if they indicated that they
had ever been told they had high blood pressure or
hypertension, a doctor had ever prescribed medicine
for their high blood pressure or hypertension, and they
were currently taking such medication (6).
Statistical analyses
Both systolic and diastolic blood pressures were
analyzed. However, because of measurement protocol
differences across surveys that complicate the interpretation of diastolic blood pressure trends (12), only
the systolic blood pressure results are presented in
detail. Analyses incorporated adjustments for sample
weights and for the complex sample design, unless
otherwise stated.
Blood pressure values were partitioned into four
categories based on thresholds derived from the NHES
I data for 35- to 44-year-old adults who were not
taking antihypertensive medication. The systolic blood
pressure cutpoints, which corresponded to the 50th,
530
Kumanyika et al.
75th, and 90th percentiles of the weighted distributions within-sex, were 127.0, 139.9, and 154.7 mmHg
for males and 120.6, 131.5, and 148.0 mmHg for
females. These same cutpoints were applied to categorize the distributions for all age groups in each
survey. We developed predictive models for the
weighted proportions of persons in each of the four
blood pressure categories defined by these three
thresholds by fitting modified proportional odds models to the weighted cumulative logits associated with
each of the three fixed thresholds.
Conceptually, proportional odds modeling is directly analogous to fitting three parallel logistic regression models with a common odds ratio for the race,
BMI, and secular trend parameters at each blood pressure threshold while still incorporating the ordinal
nature of the blood pressure thresholds. The cumulative logit is the logarithm of the ratio of the probability
of being above a specified threshold to the corresponding probability of being below that threshold. Thus,
the variation in the distribution of systolic blood pressure values over time was modeled in terms of the
three cumulative logits corresponding to the three
threshold cutpoints. The resulting data can be interpreted in terms of the odds of a randomly selected
person's exceeding versus not exceeding a particular
cutpoint in comparison with a randomly selected reference person. Within the model for each subgroup,
the reference group was the group presumed to be at
lowest risk: whites of a given sex and age with
BMI less than 25 in NHES I. There is an inherent
assumption that model effects are identical at each of
the selected thresholds, so that the interpretation is the
same for all cutpoints of the distribution in question.
The weighted proportions and complex samplebased covariance matrices were calculated within each
age group using PC CARP (13). Predictive models
were fitted to the weighted cumulative logits, incorporating the complex sample-based covariance matrix,
using SAS PROC CATMOD (14). Fitted proportions,
odds ratios, and 95 percent confidence intervals were
subsequently estimated with SAS IML (15).
Separate models were generated for each 10-year
age group within the 35- to 74-year age range, withinsex. Within each age group, the data structure of the
analysis was a 12 X 4 contingency table of observed
frequencies and weighted proportions. The 12 rows
corresponded to two levels of race (black or white)
within two levels of BMI (<25 or >25) in each of the
three surveys, and the four columns corresponded to
the four ordinal blood pressure categories noted above.
Each model proceeded from the same baseline
model containing one BMI effect, one race effect
nested within each of the two BMI strata, separate
linear secular trend effects across the three surveys for
each of the four BMI X race subgroups, and separate
nonlinear (quadratic) secular trend effects for each of
the four BMI X race subgroups. Only significant trend
parameters were retained in the model. A significant
linear trend, for example, implied that the model-based
odds ratio (on the logarithmic scale) of a randomly
selected person's exceeding a particular cutpoint increased (decreased) over time at the estimated annual
rate in comparison with a randomly selected reference
person. Similarly, a significant quadratic effect (negative) implied that the slope of this trend in the odds
(on a logarithmic scale) over time increased between
1960-1962 and 1971-1975 and then decreased between 1971-1975 and 1976-1980.
This overall modeling approach, described elsewhere in detail (16, 17), requires that all predictive
factors be classified in a few discrete categories to
avoid problems with small sample sizes. Instances of
small sample sizes were handled with modified procedures to ensure stable results. However, in order to
further investigate potential BMI effects within the
two broad binary categories, especially above a BMI
of 25, we refitted all final models within SAS PROC
LOGISTIC (18), incorporating an effect for BMI as a
continuous variable. This methodology permitted the
inclusion of sampling weights but could not accommodate adjustments for the complex sample design
effects across the three surveys.
To investigate the effects of antihypertensive drug
therapy, in the final models we first estimated the odds
ratios ignoring blood pressure treatment (observed
data) and then recomputed them after executing a
procedure in which all respondents who reported drug
therapy and whose observed blood pressure was below
the 90th percentile reference cutpoint (154.7 mmHg
for males and 148.0 mmHg for females) were reassigned to the highest category of the distribution. This
imputation approach not only adjusted for the use of
antihypertensive medication but also reflected the extent of blood pressure lowering among those taking
medication. That is, only those taking medication
whose blood pressure was below the reference 90th
percentile cutpoint were affected by the imputation
procedure. To ensure comparability when investigating the effects of drug therapy, we included the same
set of independent variables in the predictive models
for the observed data and the imputed data, provided
that a given term was statistically significant in either
of the two models.
Goodness-of-fit statistics indicated that the final cumulative logit models accounted for at least 80 percent
of the total variation in seven of 16 analysis groups
(i.e., observed and imputed models for four age groups
Am J Epidemiol
Vol. 148, No. 6, 1998
Drug Therapy and Systolic Blood Pressure Trends
within each sex) and at least 69 percent of the total
variation in all but one age-sex stratum. The models
for males aged 65-74 years had an unacceptably poor
fit (less than 40 percent of the total variation was
explained), and those data are therefore not presented.
531
having a systolic blood pressure above or below the
sex-specific 90th percentile reference threshold. The
results of this classification procedure are shown in
table 2, expressed as the proportion below the cutpoint.
If all persons taking antihypertensive medication had
well-controlled systolic blood pressures, then few or
none of the observations would have been in the top
category, and all percentages in table 2 would be near
100 percent. However, as is shown, the extent of blood
pressure lowering in persons reporting drug therapy
was variable. In a few subgroups of black respondents
in NHES I, the percentage below the cutpoint was
zero; i.e., everyone in that subgroup who reported
using antihypertensive medication nevertheless had a
relatively high blood pressure reading. There was no
consistent pattern by age, race, or sex; this may have
been influenced by the small numbers of persons reporting antihypertensive drug therapy in some subgroups.
The blood pressure category (i.e., above or below
the reference 90th percentile) of persons taking antihypertensive medication was also examined according
to various BMI parameters (e.g., proportion with
BMI ^ 25, mean BMI, and median BMI). Persons on
drug therapy whose blood pressures were above the
reference 90th percentile threshold generally had
higher BMI levels than those on drug therapy who had
lower blood pressures. BMI differences between persons whose hypertension was better controlled versus
RESULTS
Secular trends in systolic blood pressure
distributions and antihypertensive medication use
Mean systolic blood pressures and selected percentiles for systolic blood pressure across the three surveys are shown in table 1 for each of the four age
groups, within-sex. A downward trend in systolic
blood pressure between 1960-1962 (NHES I) and
1976-1980 (NHANES H) is evident, particularly at
older ages. Trends in antihypertensive medication use
are shown for males and females in figures 1 and 2,
respectively. The percentage of persons taking antihypertensive medication generally increased across age
groups and surveys, was higher in blacks than in
whites within each BMI category, and was higher
among persons with BMI ^ 25 within each race and
age group.
Blood pressure category by medication status
and BMI status
As was noted above, persons who reported taking
antihypertensive medication were first classified as
TABLE 1. Weighted mean systolic blood pressure* and selected weighted percentile i of systolic blood pressure in three US
national health examination surveys, by age group and sex, 1960-1980
Age group
(years)
and
survey*
Males
Unweighted
s&rnplo
nt
Females
Unweighted
Sflrnpfo
Weighted percentle
Weighted
mean
50th
75th
90th
nt
Weighted
mean
Weighted percentile
50th
75th
90th
35-44
NHESI
NHANES1
NHANES II
678
653
720
130.8
127.9
125.7
127.4
126.4
123.5
140.6
136.4
135.7
156.2
149.2
146.3
739
1,616
803
123.5
120.0
120.4
119.7
117.9
133.6
130.8
130.0
149.3
148.8
140.5
45-54
NHESI
NHANES1
NHANES II
529
755
672
136.2
135.4
131.9
133.8
130.7
129.0
149.2
144.0
141.6
161.9
161.5
158.4
677
822
728
135.5
133.0
129.8
131.2
129.5
127.0
145.4
143.5
140.6
169.0
161.0
158.3
55-64
NHESI
NHANES1
NHANES II
405
588
1,201
144.1
140.0
139.0
140.6
138.7
137.6
159.2
151.8
150.1
173.8
166.4
168.3
433
659
1,275
149.1
143.7
138.3
142.9
139.5
135.1
163.2
159.5
149.2
190.3
176.4
170.1
65-74
NHESI
NHANES1
NHANES II
262
1,616
1,176
152.3
146.9
144.0
149.8
144.2
139.8
168.8
160.6
158.3
189.1
178.8
177.0
279
1,789
1,366
163.2
152.3
146.7
160.0
149.0
141.9
183.9
166.4
160.6
206.2
185.2
180.9
124.7
* NHES I, National Health Examination Survey, cycle I (1960-1962); NHANES I, First National Health and Nutrition Examination Survey
(1971-1975); NHANES II, Second National Health and Nutrition Examination Survey (1976-1980).
t Analyses were restricted to respondents with nonmissing information on systolic blood pressure, diastolic blood pressure, and high blood
pressure treatment status.
Am J Epidemiol
Vol. 148, No. 6, 1998
532
Kumanyika et al.
Race-BMI category and survey
FIGURE 1. Percentages of black males and white males taking antihypertensive medication, by 10-year age group and body mass index
(BMI) category, as reported in three US national hearth examination surveys conducted between 1960 and 1980.
p
70 -
pertensn/emed
1 60 60 -
• 36-44y
I46-S4y
White females
BMI < 25
BMI ;> 25
IM-76y
65-64y
Black females
BMI < 25
I .
I BMI ^ 25
40 -
1 20 CO
O)
12
B 10 -
Race-BMI category and survey
FIGURE 2. Percentages of black females and white females taking antihypertensive medication, by 10-year age group and body mass
index (BMI) category, as reported in three US national health examination surveys conducted between 1960 and 1980.
Am J Epidemiol
Vol. 148, No. 6, 1998
Drug Therapy and Systolic Blood Pressure Trends
533
TABLE 2. Unweighted number of persons and percentage of persons in three US national health examination surveys with
systolic blood pressure below the 90th percentile Imputation threshold among persons who reported taking antihypertensive
medication, by race, body mass index category, age group, and sex, 1960-1980
Females
Males
Race, D M I *
category, and
age group
(years)
NHES I*
NHANES I*
NHANES II*
NHANESI
NHANESD
No.t
%
<154.7
mmHgt
Whfles wtth BMI < 25
35-44
45-54
55-64
65-74
2
2
12
15
100
50
25
40
4
12
19
90
75
67
63
51
8
13
58
89
100
54
71
57
3
15
14
22
67
40
50
14
20
29
33
151
75
59
42
38
6
37
86
145
100
62
55
55
Writes wttti BMI ^25
35-44
45-54
55-64
65-74
8
15
19
17
13
47
32
35
12
36
47
149
67
53
57
50
17
54
131
150
76
61
59
60
12
26
45
45
67
39
22
18
39
41
72
295
54
39
42
30
26
63
196
296
65
57
53
38
Blacks with BMI < 25
35-44
45-54
55-64
65-74
2
3
3
5
100
0
33
0
1
10
8
34
100
70
63
47
6
2
10
19
33
100
80
68
2
8
1
4
50
50
0
25
12
4
6
35
42
50
33
26
2
5
13
19
50
60
46
32
Blacks wtth BMI z 25
35-44
45-54
55-64
65-74
2
1
3
2
100
100
0
0
7
10
9
37
71
60
33
35
4
5
24
16
50
80
63
44
12
15
12
9
25
33
50
11
35
23
32
88
63
44
34
27
14
31
40
68
57
52
53
49
No.
%
<154.7
mmHg
No.
%
<154.7
mmHg
NHESI
No.
%
<148.0
mmHg
No.
%
<148.0
mmHg
No.
%
<148.0
mmHg
• BMI. body mass Index (weight (kgVnelgnt (m)*); NHES I, National Health Examination Survey, cycle I (1960-1962); NHANES I, First National Healh and
Nutrition Examination Survey (1971-1975); NHANES II, Second National Health and Nutrition Examination Survey (1976-1980).
t Number o< respondents reporting treatment (mediation use) lor high blood pressure.
t Percentage o) respondents wtth a systoBc blood pressure below the sex-epecSlc 90% percentBe cutpolnt (see text).
less well controlled on drug therapy by this criterion
were significantly different at p < 0.05 (age groupadjusted parametric tests) for white males and black
females in 1960-1962, for all sex-race groups in
1971-1975, and for black males and white females in
1976-1980.
Model-based estimates of secular trends, race,
and BMI effects on systolic blood pressure
assuming drug therapy versus no drug therapy
The cumulative logit models generated to evaluate
the effects of time, race, and BMI on systolic blood
pressure are shown in table 3 for males and in table 4
for females. Within each race, sex, and age group,
these odds ratios compared persons in each survey
with whites in the lower of the two BMI strata (BMI <
25) in the earliest survey (NHES I). Decreasing odds
ratios across rows reflect a downward secular trend,
and the linear trend in the odds was statistically significant (a = 0.01) for the rows highlighted in bold
type. Few significant quadratic terms were noted, but
where present they are mentioned below.
Males. In the models with no adjustment for effects of drug therapy, significant decreases in systolic
blood pressure were observed across surveys in seven
of the 12 strata presented for males (table 3). For
example, among 35- to 44-year-old white males with
Am J Epidemiol
Vol. 148, No. 6, 1998
BMI < 25, a randomly selected NHANES I and
NHANES II respondent was, respectively, 60 percent
and 50 percent less likely than a randomly selected
white male in the same BMI stratum in NHES I to
have a blood pressure exceeding a given threshold.
Decreasing systolic blood pressure over time was also
observed in 35- to 44-year-old white and black males
with BMI 2: 25, in 45- to 54-year-old white males
with BMI < 25, in 45- to 54-year-old black
males regardless of BMI category, and in 55- to 64year-old black males with BMI £: 25. The quadratic
term was significant for the trend in odds ratios among
45- to 54-year-old black males with BMI > 25. For
the latter group, there was a significant decrease in
systolic blood pressure between 1971-1975 and
1976-1980 but an increase between 1960-1962 and
1971-1975.
The imputed models assuming that no drug therapy
was in effect can be interpreted in the same manner as
above for the data with no imputations, except that all
blood pressure distributions, including those for the
reference group (NHES I white males with BMI <
25), have been altered so that all persons reporting use
of antihypertensive medication are in the uppermost
category of the distribution. As noted previously, the
extent to which the distributions in a given subgroup
534
Kumanyika et al.
with a body mass
TABLE 3. Model-based odds ratios for systolic blood pressure among1 males, relattve to NHES 1*respondents
i
index less than 25, in three US national health examination i•urveys, by age group and race,1960-1980
Age group
(years), rawi,
and BMI*
category
35-44
White
<25
Z25
Black
<25
i25
45-54
White
<25
i25
Black
<2S
£25
55-64
White
<2S
S25
Black
<25
i25
Model Inducing drug therapy (no imputation)
NHES I
NHANES I*
Model with data Imputed to assume no drug therapy
NHANES II*
NHES I
NHANES I
NHANES II
95% Cl*
OR
95% Cl
OR
95% CM
OR
95% Cl
OR
95% Cl
OR
95% Cl
1.6-2.8
0.6
1.5
0.5-0.7
1.1-2.0
0.5*
2.1
1.3
a 4-0.6
0.9-1.9
1.0
Z2
1.7-2.9
0.7
1.6
0.6-0.9
1.2-Z2
0.6
1.5
0.4-0.8
1.1-2.0
1.7
4.9
1.0-3.0
3.6-6.7
1.7
3.6
1.0-3.0
2.6-4.8
1.7
3.1
1.0-3.0
2.2-4.3
25
7&
1.7-4.7
5.5-10.4
25
6.6
1.7-4.7
4.1-7.7
25
5.0
1.7-4.7
3.6-7.0
1.0
1.1
0.7
1.3
O5-O.9
05-15
0.6
1.4
0.4-0.8
0.8-1.7
1.0-15
1.0
1.2
03-1.7
0.8
1.6
0.6-1.0
1.1-2.1
0.7
1.7
05-0.9
1.3-2.3
3.0
6.1
1.7-6.6
3.1-12.0
1.6
8.3
05-2.6
4.8-14.2
1.2
15
17-2.1
1.0-3J
2.7
32
13-4.7
1.8-5.5
2.1
6.2
1.3-3.3
3.6-10.7
15
1.7
1.1-3.2
05-3.1
1.0
1.8
1.5-2.3
1.0
1.8
15-23
1.3-25
1.0
1.8
1.5-2.3
1.0
1.6
13-1.9
1.0
1.6
1.3-15
1.0
1.6
1.3-15
1.8
9.6
1.3-25
6.6-16.7
1.9
4.4
15
3.2
1.3-2.8
2.1-4.9
1.1
4.0
0.8-1.6
2^-7.3
1.1
4.8
0.8-1.6
35-6.4
1.1
5.1
05-1.6
3.4-7.7
OR*
1.0t
• NHES I, National Heath Examination Survey, cycle I (1960-1962); BMI, body mass Index (weight (kg>tielght (m)»); NHANES I, First National Health and
Nutrtion Examination Survey (1971-1975); NHANES II, Second National Health and Nutrition Examination Survey (1976-1980); OR, odds ratio; Cl, confidence
Interval.
t Referent.
t Bold type Indteates that the trend trom 1960-1962 to 1976-1980 was statlstlcaly stgrtffcant at p < 0.01.
change is a function of the proportion taking medication and of whether their blood pressures were below
the 90th percentile reference threshold before the imputation. The finding of a decrease in systolic blood
pressure over time in three strata of 35- to 44-year-old
males was unchanged under the assumption that no
drug therapy was in effect. However, among 45- to
54-year-old males, the pattern in the imputed data was
less favorable than the pattern seen prior to imputation: 1) a significant linear decrease in systolic blood
pressure was no longer observed in low-BMI white
males or in black males in this age group (although the
significant quadratic term in the 45- to 54-year-old
black males with BMI ^ 25 remained); and 2) in white
males with BMI ^ 25, an increase in systolic blood
pressure over time, suggested in the observed data,
became statistically significant in the model assuming
no drug therapy. The significant decrease in systolic
blood pressure in 55- to 64-year-old black males with
BMI 5: 25 also became nonsignificant under the assumption that no drug therapy was in effect
Differences in odds ratios by race and BMI were
evident in both the observed and the imputed models
in table 3; these differences were statistically significant in both the observed and the imputed models,
except for one case (there was no race effect in males
aged 55—64 years under the assumption of no drug
therapy (p = 0.43)). That is, within-race, odds ratios
for males with BMI ^ 25 were greater than those for
males with BMI < 25, reflecting a consistent association of higher BMI with an overall higher level of
blood pressure. Within BMI strata, odds ratios for
black males were higher than those for white males,
reflecting a consistent association of black race with
higher systolic blood pressure. The odds ratios for
black males with BMI ^ 25 then reflected the combined effects of race and BMI on systolic blood pressure in each time period. In spite of the decline in
blood pressure across surveys, an NHANES II black
male with BMI ^ 25 was still 2-3 times as likely as an
NHES I white male with BMI < 25 to have a systolic
blood pressure exceeding a given threshold (i.e., odds
ratios were 3.1, 1.8, and 3.2, respectively, for
NHANES n black males aged 35-44, 45-54, and
55-64 years who had BMI levels > 25).
Females. Model results for females are presented
in table 4. Before imputation, significant decreases in
systolic blood pressure over time were observed in 10
of the 16 strata: in 35- to 44-year-olds, except white
females with BMI ^ 25; in 45- to 54-year-old white
females; in 55- to 64-year-old white females with
BMI > 25; and in all 65- to 74-year-old females. A
significant quadratic term in the model for 35- to
44-year-old females with BMI < 25 suggested a
steeper decline between 1971-1975 and 1976-1980
than between 1960-1962 and 1971-1975. In the imAm J Epidemiol
Vol. 148, No. 6, 1998
Drug Therapy and Systolic Blood Pressure Trends
535
TABLE 4. Model-based odds ratios for systolic blood prassure among females, relative to NHES I* respondents with a body
mass index less than 25, in three US national healthi examination surveys, byage group and race, 196O-1980
Age group
(years), race,
andBMI*
category
Model with data Imputed to assume no drug therapy
Model Indudng drug therapy (no imputation;I
NHES I
NHANES I*
NHES I
NHANES II*
NHANES II
NHANESI
OR*
95%CI»
OR
95% Cl
OR
95% Cl
OR
95% Cl
OR
95% Cl
OR
95% Cl
1.0t
25
1.8-3.4
0.8
2.5
0.6-1.0
1.8-3.4
0.4*
25
a 3-0.5
1.8-3.4
1.0
22
1.7-2.9
0.8
22
0.6-1.0
1.7-2.9
0.4
22
0.3-0.5
1.7-2.9
4.3
5.1
Z8-6.6
3.7-6.9
3.4
3.2
2.0-6.7
2.5-4.3
1.6
2.7
1.0-2.7
1.9-3.8
3.6
6.0
2.5-4.9
4.4-8.1
Z7
3.8
1.8-4.2
2.8-5.0
1.5
3.1
1.0-Z2
2J-4.4
1.0
2.8
2.1-3.6
0.6
1.8
0.5-0.8
1.4-2.3
0.6
1.5
0.4-0.7
1.1-2.0
1.0
2.4
1.8-3.3
0.7
2.0
0.6-0.8
1.6-2.6
0.6
15
0.4-0.8
1.4-2.4
4.0
4.1
2.8-5.8
3.0-5.6
4.0
3.8
2.8-5.8
2.8-5.2
4.0
4.1
Z8-5.8
3.0-5.6
45
5.0
3.1-6.4
3.6-6.9
45
5.0
3.1-6.4
3.6-5.9
45
5.0
3.1-6.4
3.6-6.9
1.0
3.9
3.1-6.1
1.0
2.3
1.9-2.7
1.0
1.8
1.6-2.3
1.0
3.8
3.0-4.9
1.0
2.9
2.5-3.4
1.0
2.6
22-32
2.0
2.6
1.3-3.1
ZO-3.5
2.0
2.6
1.3-3.1
2.0-3.5
2.0
2.6
1.3-3.1
£0-3.5
22
3.8
1.5-3.5
25-4.9
22
3.8
15-35
2.9-4.9
22
3.8
A 5-35
25-4.9
1.0
1.7
1.1-2.6
0.7
1.1
0.5-1.0
0.8-1.6
0.4
0.9
a 3-0.6
0.6-1.2
1.0
1.8
1.3-2.5
0.9
1.5
0.6-1.3
1.1-2.1
0.6
1.4
0.4-05
1.0-2.1
1.5
2.3
1.0-Z2
1.5-3.5
1.0
1.4
0.6-1.6
1.0-2.0
0.6
1.2
0.4-1.0
0.8-1.7
1.7
3.3
1.2-2.4
2.1-5.3
15
2.8
0.9-Z4
1.8-4.6
1.0
2.6
0.6-1.7
1.6-4.5
35-44
White
<25
£25
Black
<25
£25
45-54
White
<25
£25
Black
<2S
£25
55-64
White
<25
£25
Black
<25
£25
65-74
Wttte
<25
£25
Black
<25
£25
• NHES I, National Health Examination Survey, cycle I (1960-1962); BMI, body mass Index (weight (kg)Aielght (m)*); NHANES I, First National Health and
NutrHon Examination Survey (1971-1975); NHANES II, Second National Health and Nutrition Examination Survey (1976-1980); OR, odds ratio; Cl, confidence
Interval.
t Reterert.
t Bold type Indicates that the trend from 1960-1962 to 1976-1980 was statistically significant at p < 0.01.
puted models, findings of significant declines in systolic blood pressure were unchanged for females below the age of 65, except for attenuation to
nonsignificance of the decline for 45- to 54-year-old
females with BMI a 25. Declines in systolic blood
pressure across surveys among females aged 65-74
years, observed consistently before the imputation,
were no longer present under the assumption of no
drug therapy.
Black females and females with BMI a 25 had
significantly higher systolic blood pressures than
white females and females with BMI < 25, respectively, with some exceptions for the race effect. Before
the imputation, race effects were suggested but were
not statistically significant at a = 0.01 in 35- to
44-year-old females with BMI > 25 (p = 0.014) and
65- to 74-year-old females with BMI < 25 (p = 0.04);
in both cases, these effects met the criterion of significance under the assumption of no drug therapy. No
significant race effect was evident among 55- to 64year-old females either before (p = 0.79) or after
\p = 0.14) the imputation. BMI effects were highly
significant throughout.
Am J Epidemiol
Vol. 148, No. 6, 1998
Additional analyses of BMI effects. A more refined
examination of the effects of BMI on systolic blood
pressure was carried out using logistic regression models in which BMI could be included as a continuous
variable. This was of particular interest with respect to
clarification of racial differences that were independent of BMI, since the overall distribution of BMI
values in black females was substantially higher than
that for white females. For example, in 1976-1980
(NHANES II), the 85th percentiles of BMI for black
females in the 10-year age groups included here
ranged from 33.6 to 35.2, as compared with a range of
30.4-31.9 for white females (10). In addition to controlling for residual confounding due to BMI as such,
this analysis also reduced the potential impact of measurement artifacts associated with differences in the
sphygmomanometer cuff sizes available across the
three surveys (12), because BMI and arm circumference are highly correlated (19). Most of the race
effects observed in the cumulative logit models were
also significant in the logistic regression models, although in some strata (45- to 54-year-old males with
BMI < 25, 55- to 64-year-old males with BMI > 25,
536
Kumanyika et al.
and 35- to 44-year-old females with BMI S: 25) there
was no longer an impression of a significant race
effect in either the observed data or the imputed data,
or both, with this better adjustment for BMI. In one
case—the model for 55- to 64-year-old males with
BMI < 25 assuming no drug therapy—a highly significant race effect emerged where none had been seen
in the cumulative logits.
Diastolic blood pressure
Examination of means and percentiles for the diastolic blood pressure distributions suggested marked
increases (about 4 mmHg) between the 1960-1962
and 1971-1975 surveys, followed by a modest decrease (about 1 mmHg) between 1971-1975 and
1976-1980. In the cumulative logit models for males,
there were some significant quadratic terms, but there
appeared to be a significant increase in diastolic blood
pressure over time in about half of the age-race-BMI
subgroups, with little change under the assumption
that no drug therapy was in effect. Means and percentiles of diastolic blood pressure for females gave a
mixed picture of the changes across surveys. No clear
pattern of change was evident in the cumulative logit
models for diastolic blood pressure models in females,
before or after the imputation, despite the appearance
in several subgroups of significant quadratic terms
suggestive of a substantial rise in diastolic blood pressure between 1960-1962 and 1971-1975 that was
subsequently reversed between 1971-1975 and 1976—
1980. Effects of race and BMI on diastolic blood
pressure were similar to those noted for systolic blood
pressure; that is, being black and having a BMI ^ 25
were consistently associated with higher diastolic
blood pressures in both males and females. These
within-survey effects would presumably be less influenced by the potential lack of comparability in diastolic blood pressure measurements across surveys.
DISCUSSION
This analysis was motivated by an interest in identifying blood pressure trends relevant to primary prevention of hypertension. We attempted to estimate
blood pressure trends that would have occurred in the
absence of antihypertensive drug therapy, which increased in prevalence during this period. The categorical (i.e., cumulative logit) approach to modeling of
blood pressure distributions is informative if one is
willing to assume that the blood pressures of persons
treated with medication would have been in the upper
part of the range if no treatment had occurred. This
approach may be more appropriate than one in which
blood pressure is handled as a continuous variable.
Imputing a specific value for each individual who
reported drug therapy would require several assumptions—e.g., about blood pressure levels at the time the
medicine was prescribed, the efficacy of the medication^) prescribed, and adherence to the medication
regimen (20). No algorithm is available for calculating
"correction factors" with which to derive, retrospectively, pretreatment blood pressures from measured
blood pressures.
Our cutpoints (154.7 mmHg for males and 148.0
mmHg for females) for defining the upper category of
the blood pressure distribution were derived statistically rather than based on presumed treatment thresholds. Since the definition of elevated systolic blood
pressure that was relevant between 1960 and 1980 was
>160 mmHg, it is highly likely that persons taking
antihypertensive medication would have had blood
pressures at least as high as our cutpoints. If anything,
physicians who did not follow guidelines to the letter
would have withheld treatment until a blood pressure
level higher than the recommended treatment cutpoint
had been observed (21). To base the cutpoints on
specific treatment thresholds would also have required
knowledge or assumptions about the prescribing behavior of clinicians with respect to persons of a given
race, sex, or age. Data on physician practices in prescribing antihypertensive medication do not provide a
clear basis for making such assumptions (21).
The observed decrease in mean systolic blood pressure and increase in the percentage of persons receiving antihypertensive drug therapy across time are consistent with prior reports from the National Center for
Health Statistics (3, 7, 22). The model-based estimates
before imputation also supported the finding of an
overall secular decline in systolic blood pressure, although not consistently in all age-race-BMI subgroups. The finding on the key question of interest—
that is, how much of the decline was attributable to
drug therapy—differed according to age. Above age
45, particularly for males aged 45-54 years and females aged 65-74 years, the declines in blood pressure
appeared to be attributable to the increasing prevalence of antihypertensive therapy over time. This is a
positive finding from the perspective of population
blood pressure control, but not with respect to progress
in primary prevention in these age groups during the
time period studied. The apparent failure of drug therapy to impact substantially on the black-white disparity in blood pressure levels during this time period is
also disappointing, given the special focus on blacks in
screening programs. In addition, the data showing
minimal or no differences in systolic blood pressure
Am J Epidemiol
Vol. 148, No. 6, 1998
Drug Therapy and Systolic Blood Pressure Trends
after imputation in several substrata in which antihypertensive therapy was prevalent underscore the fact
that having a prescription for antihypertensive medication was not synonymous with adequate blood pressure control. A high proportion of persons who reported undergoing drug therapy had blood pressures
that were already at the upper end of the distribution;
this finding was more common among blacks and,
particularly, among persons with BMI > 25. An association between being overweight and having poorer
blood pressure control with antihypertensive medication was also reported by Schmieder et al. (23) based
on screening of German civil servants. Unless this
problem is addressed through improved pharmacologic or nonpharmacologic treatment, a lesser ability
to control blood pressure with medication among overweight persons in the face of an increasing prevalence
of obesity in the United States (24) may continue to
limit gains in hypertension control.
The goal of antihypertensive drug therapy is to
lower blood pressure sufficiently to limit damage to
target organs and reduce morbidity and mortality (25,
26). Although more numerous and better drugs for
lowering blood pressure have become available since
the 1970s, the ability to achieve this goal with medication alone can still be questioned. In the Third
National Health and Nutrition Examination Survey
(1988-1991), only 21 percent of persons being treated
with antihypertensive medication had blood pressures
that had been controlled to < 140/90 mmHg. While
this is an improvement over the 11 percent of such
persons whose blood pressure was controlled below
this threshold in NHANES n, it indicates that most
hypertensive persons being treated with medication
may be bearing the financial costs and medical risks
associated with chronic drug therapy without deriving
the full benefit of treatment. The potential benefits of
medication (as observed in efficacy trials, for example
(25)) will only be observed in the general population if
treatment is sufficient to lower blood pressure as effectively as in these trials, and this may be difficult to
achieve (27).
Considering this, our findings in the 35- to 44-yearolds are encouraging, since they suggest that factors
favorable to primary prevention were reaching
younger cohorts of US adults in the 1960s and 1970s.
The observation of a significant secular decline in
systolic blood pressure in this age group suggests a
lower incidence of blood pressure elevation, possibly
due to nonpharmacologic control of blood pressure.
Our finding that the significance of the decline before
imputation held up in the models assuming no drug
therapy is consistent with the relatively low prevalence
Am J Epidemiol
Vol. 148, No. 6, 1998
537
of antihypertensive drug therapy in this age group.
This finding is also consistent with our prior observation of a decrease in systolic blood pressure among 18to 34-year-olds (2), for whom drug treatment was not
an issue.
Factors responsible for the decrease in systolic
blood pressure cannot be directly identified with the
data available. Candidates would include reductions in
sodium intake, alcohol intake, or body weight or increases in physical activity among younger cohorts in
the US population (28). Of these factors, body weight
is the only one for which national survey data can
support sound inferences about time trends that might
be parallel to the blood pressure trends. No declines in
body weight were reported between 1960 and 1980
(24), which indirectly suggests that changes in body
weight were not responsible for the blood pressure
trends. Since the 1970s, BMI levels have increased
dramatically in the US population among children,
adolescents, and adults (24, 29). Thus, the strong association between higher BMI and higher blood pressure throughout these data and the finding that it was
largely independent of drug therapy raises concern
with respect to current and future blood pressure
trends.
Our use of a single blood pressure reading leaves
open the possibility of misclassification in the assignment of people to blood pressure categories, but such
misclassification was probably similar across surveys,
particularly for systolic blood pressure. However, as
with any trend analysis, the possibility that increases
or decreases over time may be due to artifacts introduced by protocol changes across surveys must be
considered. Based on our own evaluation of the methods used in these national surveys and on a discussion
of the relevant methodological issues by those directly
involved in designing and conducting the surveys (12),
we are confident that the systolic blood pressure trends
found here reflect actual trends in the blood pressure
distributions in the population. However, as we noted
in "Materials and Methods," this is less certain for the
observed trends in diastolic blood pressure, which are
more subject to measurement bias (12). Thus, the
suggestion, in our model-based estimates for diastolic
blood pressure, of increases over time does not necessarily contradict the more favorable findings for
systolic blood pressure. Taken together, both sets of
findings underscore the critical importance of tracking
progress in primary prevention and providing for trend
analyses in which underlying risk factor distributions
can be evaluated. Observed trends that fail to exclude
the effects of antihypertensive drug therapy may foster
a false sense of security.
538
Kumanyika et al.
ACKNOWLEDGMENTS
This research was supported by grant HL33407 from the
National Heart, Lung, and Blood Institute.
The authors gratefully acknowledge the technical support
provided by Tonya J. Sharp in the conduct of these analyses.
13.
14.
15.
REFERENCES
1. Burt VL, Whelton P, Roccella EJ, et al. Prevalence of hypertension in the US adult population: results from the Third
National Health and Nutrition Examination Survey,
1988-1991. Hypertension 1995;25:305-13.
2. Kumanyika SK, Landis JR, Matthews YL, et al. Secular trends
in blood pressure among adult blacks and whites aged 18-34
years in two body mass index strata, United States,
1960-1980. Am J Epidemiol 1994;139:141-54.
3. Rowland M, Roberts J. Blood pressure levels and hypertension in persons ages 6-74 years, United States 1976-80.
Hyattsville, MD: National Center for Health Statistics, 1982.
(Advance data from vital and health statistics, no. 84) (DHHS
publication no. (PHS) 82-1250).
4. National Center for Health Statistics. Plan and initial program
of the Health Examination Survey. Rockville, MD: National
Center for Health Statistics, 1965. (Vital and health statistics,
series 1, no. 4) (DHEW publication no. (PHS) 1000).
5. National Center for Health Statistics. Plan and operation of
the Health and Nutrition Examination Survey: United
States—1971-1973. Hyattsville, MD: National Center for
Health Statistics, 1973. (Vital and health statistics, series 1,
no. 10a) (DHEW publication no. (PHS) 79-1310).
6. McDowell A, Engel A, Massey JT, et al. Plan and operation of
the Second National Health and Nutrition Examination Survey, 1976-1980. Hyattsville, MD: National Center for Health
Statistics, 1981. (Vital and health statistics, series 1, no. 15)
(DHHS publication no. (PHS) 81-1317).
7. Dridz T, Dannenberg AL, Engel A. Blood pressure levels in
persons 18-74 years of age in 1976-80, and trends in blood
pressure from 1960-1980 in the United States. Hyattsville,
MD: National Center for Health Statistics, 1986. (Vital and
health statistics, series 11, no. 234) (DHHS publication no.
(PHS) 86-1684).
8. National Center for Health Statistics. Weight, height, and
selected body dimensions of adults: United States—1960-62.
Washington, DC: National Center for Health Statistics, 1965.
(Vital and health statistics, series 11, no. 8) (DHEW publication no. (PHS) 1000).
9. National Center for Health Statistics. Obese and overweight
adults in the United States. Hyattsville, MD: National Center
for Health Statistics, 1983. (Vital and health statistics, series
11, no. 230) (DHHS publication no. (PHS) 83-1680).
10. National Center for Health Statistics. Anthropometric reference data and prevalence of overweight: United States—
1976-1980. Hyattsville, MD: National Center for Health Statistics, 1987. (Vital and health statistics, series 11, no. 238)
(DHHS publication no. (PHS) 87-1688).
11. Sichieri R, Everhart JE, Hubbard VS. Relative weight classifications in the assessment of underweight and overweight in
the United States. Int J Obes Relat Metab Disord 1992; 16:
303-12.
12. Burt VL, Cutler JA, Higgins M, et al. Trends in the preva-
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
lence, awareness, treatment, and control of hypertension in the
adult US population: data from the health examination surveys, 1960 to 1991. Hypertension 1995;26:60-9.
Fuller WA. PC CARP. Ames, IA: Statistical Laboratory, Iowa
State University, 1986.
SAS Institute, Inc. SAS language guide, release 6.03 ed. Cary,
NC: SAS Institute, Inc, 1988.
SAS Institute, Inc. SAS/IML user's guide, release 6.03 ed.
Cary, NC: SAS Institute, Inc, 1988.
Landis JR, Lepkowski JM, Davis CS, et al. Cumulative logit
models for weighted data from complex sample surveys, hi:
Proceedings of the social statistics section of the American
Statistical Association. Washington, DC: American Statistical
Association, 1987:165-70.
Freeman JL, Freeman DH Jr, Ostfeld AM. Analysis of data
from successive complex sample surveys, with an example of
hypertension prevalence from the United States Health Examination Survey, hit J Epidemiol 1983; 12:230-7.
SAS Institute, Inc. SAS/STAT user's guide, version 6, 4th ed.
Vol 2. Cary, NC: SAS Institute, Inc, 1989.
Micozzi MS, Albanes D, Jones DY, et al. Correlations of body
mass indices with weight, stature, and body composition in
men and women in NHANES I and n. Am J Clin Nutr
1986;44:725-31.
Heitjan DF, Landis JR. Assessing secular trends in blood
pressure: a multiple imputation approach. J Am Stat Assoc
1994;89:750-9.
National Heart, Lung, and Blood Institute. Diagnosis and
management of hypertension—1987. A nationwide survey of
physicians' knowledge, attitudes and reported behavior. (Final
report). Washington, DC: US GPO, 1989. (NHLBI publication no. 89-2968).
Dannenberg AL, Drizd T, Horan MJ, et al. Progress in the
battle against hypertension: changes in blood pressure levels
in the United States from 1960 to 1980. Hypertension 1987;
10:226-33.
Schmieder RE, Gatzka C, Klingbeil A, et al. Obese hypertensive patients are less effectively treated than lean hypertensives. J Hypertens Suppl 1993;1 l(suppl 5):S348-9.
Kuczmarski RJ, Flegal KM, Campbell SM, et al. Increasing
prevalence of overweight among US adults: The National
Health and Nutrition Examination Surveys, 1960 to 1991.
JAMA 1994;272:205-ll.
National High Blood Pressure Education Program. The fifth
report of the Joint National Committee on Detection, Evaluation, and Treatment of High Blood Pressure (JNC V). Arch
Intern Med 1993;153:154-83.
Furberg CD, Berglund G, Manolio TA, et al. Overtreatment
and undertreatment of hypertension. J Intern Med 1994;235:
387-97.
Thurmer HL, Lund-Larsen PG, Tverdal A. Is blood pressure
treatment as effective in a population setting as in controlled
trials? Results from a prospective study. J Hypertens 1994;12:
481-90.
Working Group on Primary Prevention of Hypertension. National High Blood Pressure Education Program Working
Group report on primary prevention of hypertension. Arch
Intern Med 1993; 153:186-208.
Troiano RP, Flegal KM, Kuczmarski RJ, et al. Overweight
prevalence and trends for children and adolescents: the National Health and Nutrition Examination Surveys, 1963 to
1991. Arch Pediatr Adolesc Med 1995,149:1085-91.
Am J Epidemiol
Vol. 148, No. 6, 1998