Professional Documents
Culture Documents
4
Copyright © 1990 by The Johns Hopkins University School of Hygiene and Public Health Printed in U.S A.
All rights reserved
Evidence for an independent role of in- reviews (1-8) the authors have concluded
creased physical activity in the primary that physically active people are at lower
prevention of coronary heart disease has risk for coronary heart disease than those
grown in recent years. In most recent major who are inactive. Recently, Powell et al. (9)
produced a thorough review of this topic,
combining approaches that had been used
Received for publication February 17, 1989, and in in earlier reviews. In their review, they
final form March 7, 1990
1
Technology Assessment Group, Harvard School
present extensive tables that document
of Public Health, Boston, MA. study characteristics, including an overall
2
University of Pennsylvania School of Medicine, rating of study quality, and summarize the
Section of General Internal Medicine, Clinical Epi- relative risk information reported for each
demiology Unit, Philadelphia, PA.
3
Channing Laboratory, Harvard Medical School, of the studies. The authors stop short, how-
Boston, MA. ever, of using formal techniques of meta-
Reprint requests to Dr. Jesse A. Berlin, University analysis to make quantitative summary
of Pennsylvania School of Medicine, Clinical Epide-
miology Unit, 331 R Nursing Education Building, statements about the relation between
Philadelphia, PA 19104-6095. physical activity and the incidence of cor-
This work was supported by research grant onary heart disease. A subsequent work
HS05936 from the National Center for Health Ser-
vices Research and Health Care Technology Assess- (10), concluding that exercise is an econom-
ment and by HL 35464 from the National Institutes ically favorable risk-reduction strategy
of Health. when compared with other preventive or
The authors wish to thank Drs. Kenneth Powell,
Sander Greenland, and Moyses Szklo for helpful com- therapeutic interventions for coronary
ments on earlier versions of this paper. Unpublished heart disease, used a median relative risk
data were kindly provided by Ralph D'Agostino and presented by Powell et al. (9) in making
Albert Belanger for the Framingham Study and by calculations. In this paper, we apply the
Paul Sorlie for the Puerto Rico Heart Health Program.
612
META-ANALYSIS OF PHYSICAL ACTIVITY AND CORONARY HEART DISEASE 613
techniques of meta-analysis to the data col- Such differences could have quite marked
lected by Powell et al., as well as more physiologic effects. We therefore separate
recent studies addressing this relation, in these two general types of activity. We ad-
order to make more formal quantitative dress this issue in our primary analysis by
statements and to explore features of study pooling results separately from studies of
design that may influence the observed re- work-related activity and studies of leisure-
lation between physical activity and coro- time activity. The few studies that use both
nary heart disease.
An additional obstacle to pooling the re- arate sub-components for each of the three
sults from the individual studies is the va- major components of quality. Powell et al.
riety of coronary heart disease outcomes describe seven separate desirable features
reported from the previous studies. These of a physical activity measure, four desir-
include coronary heart disease incidence, able components of a coronary heart dis-
coronary heart disease death, myocardial ease measure, and five desirable aspects of
infarction, myocardial infarction plus sud- the epidemiologic methods, for a total of 16
den death combined, angina pectoris, and components. The authors rated each com-
fication on type of activity (work vs. lei- scores: assessment of disease, measurement
syre), outcome (coronary heart disease, cor- of activity, and epidemiologic methods. The
onary heart disease death, etc.), degree of weight assigned to each study was the in-
activity (moderate vs. sedentary), and qual- verse of the variance of the log relative risk.
ity score. That is, it does not assume ho- For each study, we chose only one out-
mogeneity of relative risks across studies. cqme to include in the regression model (so
Heterogeneity may result, for example, as to maintain independence among obser-
In addition to these general rules, we also of nonoccupational activity that have ap-
adopted some technical rules. To convert peared since the publication of Powell et
confidence intervals into estimates of the al. (9) and that met their inclusion criteria.
variance of the log relative risk, we first In a separate analysis, these studies were
transformed the interval to the log scale. combined for the appropriate outcomes.
The lower endpoint was subtracted from Because we did not feel we could exactly
the upper endpoint to compute the length replicate the scoring system used by Powell
O O So O O O
cupational activity categories (see tables
zz < zzz
3-5). In these tables, we report the chi-
square statistic from the test for heteroge-
neity of relative risks for each pooled rela-
tive risk.
•2
Several patterns emerge from the data. -S
o
Many of the pooled relative risks are sta- c
o
tistically significant and accordingly their
confidence intervals exclude 1.0. Statistical a) o> a>
2 §
significance is not generally apparent using 2 2 2
U
Q.
risks, although the confidence intervals g
spo
0)
tend to be narrower when more studies are 11 c to
c
o
pooled. When unpublished data from the « a
aj
North Karelia, Finland male 3,978 CHD death High leisure Low leisure 1.4 (0.9-2.2) Age, BP, CH, SM, 5 20
residents (26) other
a
MI High work and High/low or low/ 1.3 (1.0-1.7) Age 1
high leisure high
High work and Low/low 2.5 (1.8-3.7) Age
high leisure
North Karelia, Finland fe- 3,688 MI High work and High/low or low/ 1.8 (0.9-3.5) Age 5 20
male residents (26) high leisure high
Low/low 4.0 (2.1-7.8) Age
Western Finland residents 15,088 CHD Work and leisure Low activity 1.4 (1.1-1.7) Age, BP, SM, CH,
(52) high activity other
Fable continues I
Downloaded from https://academic.oup.com/aje/article/132/4/612/102184 by University of Kent user on 10 October 2022
TABLE 2—Continued
Nonoccupational cohort studies used in the meta-analysis*
Subjective Objective
Cohort Cohort High activity Relative risk quality
Outcomet Low activity group Adjustments? quality
(reference no.) size group (95% CI+) score || scorell
Lipid Research Clinics Prev- 4,276 CHD High physical Low fitness 2.7 (1.4-5.1) Age, BP, CH, SM,
alence Study Participants fitness based other
(53) on exercise
test
MRFIT trial participants 12,138 CHD death Upper tertile Lowest tertile 1.5 (1.09-2.04) Age, BP, CH, SM,
(54) MI physical fit- 1.2 (1.04-1.43) other
ness
Oslo, Norway residents (55) 2,014 CHD death Upper quartile of Lowest quartile 4.8 (2.2-10.7) Age
physical fit-
ness
Middle two quar- 1.5 (0.6-3.8) Age
tiles
US railroad workers (56) 3,043 CHD death Upper (fourth) Lowest quartile 1.28 (0.99-1.63) Age, BP, CH, SM
quartile of Third quartile 1.05 (1.00-1.11)
kcal/expendi-
ture w
Belgian physical fitness 2,363 CHD Upper quartile of Lowest quartile 0.69 (0.25-1.93) None
study participants (57) leisure activity
Middle two quar- 0.90 (0.40-2.04)
tiles
Finnish men (58) 636 CHD death High total Light and sedentary 1.3 (0.89-1.89) Age, BP, CH, SM, o
other o
Honolulu heart study partic- CHD Upper tertile of Lowest tertile Age, SM, other s
ipants (59) total activity Middle tertile
Ages 45-64 7,221 1.45 (1.14-1.89)
Ages 65+ 423 2.33 (1.01-5.26)
Framingham, Massachusetts 4,121 CHD Work and lei- Low activity 1.27 (1.01-1.61) Age, sex 6 26
residents (30, 60)** sure, high ac-
tivity
Moderate activity 1.20 (1.00-1.44)
Puerto Rico residents (31)tt 8,838 CHD Framingham ac- Lowest tertile 1 41 (1.04-1.91) Age 6 26
tivity indexes
upper tertile
Middle tertile 0.95 (0.67-1.37)
* Adapted from Powell et al. (9).
t Abbreviations: AP, angina pectoris; CHD, coronary heart disease; CHF, congestive heart failure; MI, myocardial infarction; SD, sudden death.
$ CI, confidence interval.
§ Abbreviations: BP, blood pressure; CH, cholesterol; SM, smoking.
|| Subjective quality scores were based on assessments of quality by Powell et al. (9) of three components of study design: measurement of activity, measurement of disease
status, and epidemiologic methods. Each component was rated as "unsatisfactory," "satisfactory," or "good" by these authors. We assigned numerical values of 0, 1, or 2 to
the three levels of quality, respectively. Thus, the subjective scores could range from values 0 to 6.
11 Objective quality scores were based on ratings by Powell et al. (9) of separate subcomponents for each of the three major components of study design. Possible values of
the objective score ranged from 0 to 32.
** Unpublished data provided by R. D'Agostino and A. Belanger and analyzed by the authors using logistic regression.
t t Unpublished data provided by P. Sorlie and analyzed by the authors using logistic regression.
Downloaded from https://academic.oup.com/aje/article/132/4/612/102184 by University of Kent user on 10 October 2022
META-ANALYSIS OF PHYSICAL ACTIVITY AND CORONARY HEART DISEASE 621
TABLE 3 risk of 1.6 for the five occupational activity
Pooled relative risks from studies of occupational studies with mixed comparison groups is
activity and risk of heart disease not very close to the null. If we compute
Outcome*
No. of Relative risk x'for the simple, inverse-variance weighted av-
studies (95% CI) heterogeneity} erages of the same five angina studies, ig-
A. Relative risks for high activity compared with noring heterogeneity, the pooled relative
moderate activity groups from studies that reported risk is 1.9 (95 percent CI 1.5-2.3). Because
TABLE 6
Summary relative risks from studies of occupational activity and risk of coronary heart disease
incidence and death
Satisfactory! Unsatisfactoryt
Outcome* N o of Relative risk x 2 for No. of Relative risk 2
x for
studies (95% CI$) heterogeneity? studies (95% CI$) heterogeneity?
A. Relative risks for high activity compared with moderate activity groups from studies that reported both
B. Relative risks for high activity compared with low activity groups from studies that did not separate
moderate and sedentary comparison groups
CHD 1 1.8 (0.9-3.3) - 1 1.1 (0.6-2.2)
CHD death 5 1.7 (1.3-2.2) 2.64 1 1.1 (0.9-1.2)
AP 3 2.2 (1.3-3.9) 9.46 2 0.9 (0.3-2.2) 3.39
C. Relative risks for high activity compared with sedentary groups from studies that reported both moderate
and sedentary comparison groups
CHD 2 1.9 (1.0-3.6) 0.00 2 1.0 (0.5-2.3) 1.77
CHD death 2 1.8(1.5-2.3) 1.66 3 1.9(1.4-2.6) 1.22
* Abbreviation: CHD, coronary heart disease; AP, angina pectoris.
t Satisfactory: subjective quality score >2. Unsatisfactory: subjective quality score 0-2. Subjective quality
scores were based on assessments of quality by Powell et al. (9) of three components of study design:
measurement of activity, measurement of disease status, and epidemiologic methods. Each component was
rated as "unsatisfactory," "satisfactory," or "good" by these authors. We assigned numerical values of 0, 1, or 2
to the three levels of quality, respectively. Thus the subjective scores could range from 0 to 6. Objective scores
were based on ratings by Powell et al. (9) of separate subcomponents for each of the three major components
of study design. Possible values of the objective score ranged from 0 to 32.
t CI, confidence interval.
§ Degrees of freedom for x 2 are one less than the number of studies.
ables in the model compared with the esti- The association also held for the objective
mate from the univariate model for quality quality score (slope = 0.0457, SE = 0.0123,
score. In addition, the coefficients for the p = 0.001). The relative risk predicted using
variables other than quality score were the objective quality score would be 0.7 for
small relative to the coefficient for quality a study with an objective score of zero and
score. The regression equation, based on 3.1 for a study with a perfect objective score
27 studies, for the most parsimonious of 32. It is of note, in terms of the degree
model is: of extrapolation involved in these predicted
risks that, among the published results, the
=
°° 2 9 2 + °' 1341 * (subjective highest observed subjective score was 5 out
of a possible 6, and the highest observed
The standard error of the slope for qual- objective score was only 23 out of a pos-
ity score is 0.0273 (p = 0.0001), suggesting sible 32.
a highly significant linear relation between In multiple regressions involving the in-
quality score and log relative risk. Using dividual components of the quality scores
this equation, the predicted relative risk for and the adjustment and activity type vari-
a study with a quality score of zero is 1.0 ables, only the quality score for epidemio-
(essentially no association between physi- logic methods was a statistically significant
cal activity and risk of coronary heart dis- predictor of the log relative risk. The cri-
ease), and for a study with a quality score teria for the evaluation of epidemiologic
of 6 the relative risk is predicted to be 2.3. methods used by Powell et al. (9) included
624 BERLIN AND COLDITZ
whether physical activity status clearly pre- between lack of activity and coronary heart
ceded the period observed for outcome disease risk.
events, whether the analysis was adjusted The data we combined from previous co-
for covariates, and whether loss to follow- hort studies indicate that the protective
up was less than 20 percent. The fact that effect of physical activity lies in prevention
the quality score contained information on of the occurrence of major cardiovascular
the adjustment for covariates may partially events, rather than in the reduction of the