Practice of Epidemiology Meta-Analysis: Richard M. Martin, David Gunnell, and George Davey Smith
Practice of Epidemiology Meta-Analysis: Richard M. Martin, David Gunnell, and George Davey Smith
Practice of Epidemiology Meta-Analysis: Richard M. Martin, David Gunnell, and George Davey Smith
1
Copyright 2005 by the Johns Hopkins Bloomberg School of Public Health Printed in U.S.A.
All rights reserved DOI: 10.1093/aje/kwh338
PRACTICE OF EPIDEMIOLOGY
META-ANALYSIS
Breastfeeding in Infancy and Blood Pressure in Later Life: Systematic Review and
Meta-Analysis
From the Department of Social Medicine, University of Bristol, Bristol, United Kingdom.
Received for publication January 29, 2004; accepted for publication June 25, 2004.
blood pressure; bottle feeding; breast feeding; cardiovascular system; hypertension; infant nutrition; milk,
human; review literature
Evidence is growing that blood pressure levels in both with the lowest breastfeeding rates (10) and the highest risks
childhood and young adulthood are influenced by factors of premature cardiovascular disease (11), and it may
operating early in life (14) and are associated with later increase understanding of cardiovascular disease mecha-
cardiovascular disease (5). Specifically, several cohort nisms operating through early life exposures.
studies suggest that blood pressure may be determined by Interpreting individual studies of the association between
early nutritional exposures, including sodium intake in breastfeeding and blood pressure in isolation is complicated.
infancy (6), consumption of formula feed (7), and breast- Firstly, cohort studies include infants born in different
feeding (8). Detection, treatment, and control of hyperten- decades during the 20th century (8, 12, 13). The composition
sion in adulthood does not reduce cardiovascular disease risk of bottle (artificial) feeds has changed during this time, and
to normotensive levels (9), supporting efforts to identify associations with particular components of these feeds may
primary prevention interventions that could be started in explain differences in results. Secondly, different definitions
early life. Any long-term effect of breastfeeding on blood of breastfeeding have been used (13, 14). Thirdly, the
pressure levels may have implications for policies promoting strength of the relation may depend on the age at outcome
breastfeeding, particularly among the least affluent families measurement (15, 16). Finally, control for confounding
Correspondence to Dr. Richard M. Martin, Department of Social Medicine, University of Bristol, Canynge Hall, Whiteladies Road, Bristol,
United Kingdom, BS8 2PR (e-mail: richard.martin@bristol.ac.uk).
15 Am J Epidemiol 2005;161:1526
16 Martin et al.
factors may have been inadequate (17). We conducted a study to determine its impact on the overall pooled mean
systematic review and meta-analysis of studies reporting on difference.
blood pressure levels in breast- and bottle-fed subjects and Selected study characteristics, chosen a priori, were
explored possible sources of heterogeneity using meta- entered as indicator variables in separate meta-regression
regression (18). analyses (18) to assess their impact on between-study varia-
tion (heterogeneity), as follows: study size (<1,000/ 1,000);
MATERIALS AND METHODS reliance on maternal recall of breastfeeding beyond infancy
(yes/no); whether breastfeeding occurred for at least 2
Included studies months (yes/no); whether breastfeeding was exclusive for at
least 2 months (yes/no); age at measurement of blood pres-
Articles were included if they fulfilled the following sure (<10 years/1145 years/>45 years); decade of birth
criteria: 1) having been breastfed in infancy was compared (before 1980/after 1980); proportion of target population
with bottle (artificial) feeding, 2) systolic or diastolic blood included in the main analysis (<30 percent/3160 percent,
pressure had been measured as an outcome, and 3) an esti- >61 percent); method of blood pressure measurement (auto-
mate of the mean difference in blood pressure between mated/manual); and whether effect estimates in the final
breast- and bottle-fed groups could be extracted from the models controlled for social factors in childhood or adult-
article. Our review was restricted to human subjects. hood (yes/no), maternal factors in pregnancy (yes/no), or
current weight (yes/no). Papers that assessed blood pressure
in infancy only (age <1 year) were investigated separately
Am J Epidemiol 2005;161:1526
Breastfeeding and Blood Pressure: A Systematic Review 17
TABLE 1. Studies reporting on associations between method of infant feeding and blood pressure beyond 12 months of age that
were not included in the current meta-analysis
Age at which
First author, source No. breastfed*;
Infant feeding Infant year of Age at which infant outcome Description
(year of publication) no. bottle fed
comparison birth feeding was assessed measurement of results
(reference no.) (sex)
occurred
Baranowski, families 245 total (M + F) Duration of any Not stated Interviewer 34 years No significant correlations
from an ethnically breastfeeding administered between duration of
diverse population in questions to mother breastfeeding and SBP
Texas (1992) (22) 34 years after or DBP observed;
infants birth quantitative estimates not
reported
Cobaleda Rodrigo, 1,893 total (M + F) Ever vs. never 19651983 018 years; method 018 years No significant differences
Madrid, Spain (1989) breastfed unclear between duration of
(23) breastfeeding and SBP
or DBP observed; no
quantitative estimates
given
Simpson, births in 692 total (M + F) Ever vs. never 19721973 3 years; method 7 years No significant difference in
Dunedin maternity breastfed unclear breastfeeding rates or
hospital, New Zealand duration of breastfeeding
(1981) (37) when comparing children
involved 10,062 subjects) (19, 20, 43), 15 studies with From these 15 studies, 17 estimates of systolic blood pres-
17,503 participants were included in the meta-analysis sure differences were derived, of which 12 included males
relating breastfeeding with blood pressure beyond 12 and females combined and five were sex specific. Eleven
months (Web table 1). systolic blood pressure observations (nine studies) were of
Two of these 15 studies were based on a follow-up of a children (aged 116 years), and six observations (five
randomized controlled trial in preterm infants (15, 16), eight studies) occurred in later adulthood (age 17 years). One
study reported results for diastolic blood pressure only (25).
were prospective cohorts (8, 14, 20, 2527, 36, 43), and one
From the 15 studies, 13 estimates of diastolic blood pressure
was a historical cohort (13); in four cross-sectional surveys
differences were derived, 12 of which included males and
of blood pressure, infant feeding history was based on retro- females combined and one of which was for males only.
spective recall by the mother (12, 19, 34, 35). These studies Nine diastolic blood pressure observations (eight studies)
included populations from the United Kingdom, Finland, were of children aged 116 years, and four observations
Holland, Belgium, Italy, Czech Republic, Croatia, South (four studies) occurred in adulthood (age 17 years).
Africa, and Australia. Individual studies were relatively
homogeneous with respect to ethnicity. The year of birth of
Definitions of breastfeeding
the subjects ranged from 1918 to 1994. The proportion of the
target population included in the main analysis was unstated The 15 studies used different definitions of breastfeeding.
in one paper (35), less than 30 percent in four studies (12, 13, In a randomized controlled trial with follow-up at ages 7.5
15, 36), 3060 percent in four studies (8, 20, 27, 43), and 8 years (16) and ages 1316 years (15), preterm infants were
more than 60 percent in six studies (14, 16, 19, 25, 26, 34). randomly assigned to donated, banked breast milk or
Am J Epidemiol 2005;161:1526
18 Martin et al.
preterm formula (either as the sole diet or a supplement to Breastfeeding and systolic blood pressure
mothers milk) until they weighed 2,000 g or were
The results for systolic blood pressure, shown in figure 2,
discharged to home. In the other studies, the exposure was
are based on 14 studies with 17 observations. Mean systolic
defined as 1) any breastfeeding in five studies (12, 19, 25,
blood pressure was lower in breastfed infants compared with
26, 35); 2) exclusive breastfeeding in five studies (exclusive bottle-fed infants according to 10 observations from eight
for the first 10 days only (13), for at least 3 months (27, 34), studies (8, 14, 15, 20, 26, 35, 36, 43). Seven observations
for at least 15 weeks (8), or for at least 12 months (36)); 3) (from six studies) showed no or little difference in systolic
both any breastfeeding and exclusive breastfeeding for at blood pressure among breastfed versus formula-fed infants
least 2 months in one study (43); and 4) any breastfeeding for (12, 13, 16, 19, 27, 34). Two of these seven observations
at least 3 months in one study (14) and at least 6 months in were from the randomized controlled trial in preterm infants
another (20). In all studies except the randomized controlled with follow-up at ages 78 years (16). When the original
trial (15, 16), the comparator group was exclusive bottle study was followed up into adolescence (ages 1316 years),
feeding. Five of the studies (providing six observations) having received breast milk was associated with a 2.7-
relied on maternal recall beyond infancy, ranging from 318 mmHg reduction in blood pressure (15).
years (14), to 3 years (27), to 57 years (34), to 2028 years In a random-effects model, mean systolic blood pressure
(12), and to 4460 years (19). was lower among breastfed infants (mean difference: 1.4
Am J Epidemiol 2005;161:1526
Breastfeeding and Blood Pressure: A Systematic Review 19
mmHg, 95 percent confidence interval (CI): 2.2, 0.6; p = observed heterogeneity was explained by study size. In a
0.001) (figure 2). There was also evidence of marked hetero- stratified meta-analysis, a smaller effect of breastfeeding on
geneity between studies (216 = 42.0, p < 0.001). Exclusion later systolic blood pressure was observed in the larger
of the study by Singhal et al. (15) (because of lack of inde- studies (n 1,000) (difference: 0.6 mmHg, 95 percent CI:
pendence from Lucas et al.s study (16)) had little impact on 1.2, 0.02; p = 0.06) compared with the smaller studies (n <
the pooled difference (1.3, 95 percent CI: 2.2, 0.5). 1,000) (difference: 2.3 mmHg, 95 percent CI: 3.7, 0.9;
Controlling for study size in a meta-regression analysis p = 0.001). This difference was unlikely to be due to chance
lowered the 2 estimate of between-study variation from 1.69 (p = 0.02). There was evidence of heterogeneity in models
when study size was not included in the model to 0.47 when restricted to small studies (212 = 27.1, p = 0.007) but less
study size was included, suggesting that some of the evidence among the four larger studies (23 = 6.1, p = 0.1).
Am J Epidemiol 2005;161:1526
20 Martin et al.
In studies where the duration of breastfeeding was at least breastfeeding on later diastolic blood pressure was similar in
2 months, the pooled blood pressure difference between the four larger studies (n 1,000) (difference: 0.4 mmHg,
breast- and bottle-fed groups (2.0 mmHg) was on average 95 percent CI: 0.9, 0.1; p = 0.10) compared with the seven
1.6 mmHg larger (95 percent CI: 0.4, 3.5; p = 0.1) than in smaller studies (n < 1,000) (difference: 0.6 mmHg, 95
studies with a shorter duration of breastfeeding (pooled percent CI: 1.5, 0.2; p = 0.15). Studies that relied on
difference: 0.6 mmHg). Similarly, the difference in blood maternal recall of breastfeeding beyond infancy showed
pressure between breast- and bottle-fed groups was 1.4 pooled differences in mean diastolic blood pressure (0.0
mmHg greater (95 percent CI: 0.4, 3.2; p = 0.1) in those mmHg) that were 0.6 mmHg smaller (95 percent CI: 0.2,
born up to 1980 (pooled difference: 2.7 mmHg) compared 1.1; p = 0.004) than in studies that did not rely on recall
with those born after 1980 (pooled difference: 0.8 mmHg). (pooled difference: 0.7 mmHg).
Only four of the 17 observations on systolic blood pressure We found little evidence that between-study heterogeneity
controlled for potential socioeconomic (19, 20, 43) or in estimates was explained by age at measurement of blood
maternal antenatal factors (such as body mass index, pressure (p = 0.5), decade of birth (p = 0.2), stipulation of a
smoking in pregnancy, education, parity, marital status) (8, minimum duration of breastfeeding (p = 0.5), proportion of
20, 43) or current body size (8, 20, 43). Controlling for the target population in the main analysis (p = 0.2), whether
confounding produced a greater than 30 percent reduction in breastfeeding was exclusive for at least 2 months (p = 0.2),
crude effect estimates in two (19, 43) of three studies in method of blood pressure measurement (p = 0.4), or whether
which comparison with crude estimates was possible. In effect estimates controlled for socioeconomic factors (p =
Am J Epidemiol 2005;161:1526
Breastfeeding and Blood Pressure: A Systematic Review 21
Blood pressure in infancy mmHg (95 percent CI: 4.0, 0.6; p = 0.15), although there
was some evidence of heterogeneity (25 = 11.8; p = 0.04).
Overall, six studies were identified that examined the rela- The pooled diastolic blood pressure difference in infancy
tion between infant feeding mode and blood pressure associated with breastfeeding was 1.1 (95 percent CI: 4.0,
measured before 12 months of age (32, 39, 40, 4547) (table 1.8; p = 0.4; 23= 8.2, p = 0.04).
2). The mean difference in blood pressure by feeding mode,
and the associated standard error, could be estimated from
DISCUSSION
four of these studies (six observations) (32, 40, 45, 46). In
random-effects models, the pooled systolic blood pressure Breastfeeding was associated with a 1.4- and 0.5-mmHg
difference in infancy associated with breastfeeding was 1.7 reduction in systolic and diastolic blood pressure, respec-
Am J Epidemiol 2005;161:1526
22 Martin et al.
tively, although differences in systolic blood pressure association (49). When all the studies were considered, we
between feeding groups were reduced in large (difference: found similar effect estimates in studies with more than 60
0.6 mmHg) compared with smaller (difference: 2.3 mmHg) percent follow-up and in those with less than 30 percent
studies. These pooled estimates are similar to those found by follow-up, suggesting that the association between breast-
Owen et al. (48) in a recent review, even though the current feeding and blood pressure did not systematically vary
report includes recently published data on an extra 10,062 between studies according to follow-up rates.
subjects from three studies that included more than 1,500 Although reporting of ever having been breastfed after up
participants each. to 20 years is highly correlated with obstetric records (50),
breastfeeding duration may be remembered less accurately
Chance, bias, and confounding (51). Three cross-sectional studies relied on retrospective
reporting of exclusive (34) or any breastfeeding 7 years (34),
A number of studies reported inverse associations between 28 years (12), and 60 years (19) after birth, and these studies
breastfeeding and blood pressure, including two (of three) showed little evidence of an association between breast-
with more than 3,500 subjects each (20, 43), suggesting that feeding and blood pressure. In meta-regression analysis, reli-
these findings are unlikely to be due to type 1 error alone. ance on maternal recall was associated with an attenuation of
Selection bias would arise if excluded subjects had a the difference in diastolic (but not systolic) blood pressure
different breastfeedingblood pressure association between breast- and bottle-fed groups. Publication bias is a
compared with those who were included. In one study, a concern because most studies in this review were small, and
protective effect of breast milk on blood pressure was mean blood pressure differences were greater in the smaller
observed when 26 percent of the original cohort were compared with the larger studies.
followed up at ages 1316 years (15), but not when 81 Relatively few studies controlled for potential
percent were examined at ages 7.58 years (16), suggesting confounding factors, although adjusted effect estimates were
either the possibility of selection bias in the later follow-up attenuated by at least 30 percent in two studies (19, 43). In
or an amplification of the breastfeedingblood pressure the meta-regression analyses, studies controlling for socio-
Am J Epidemiol 2005;161:1526
Breastfeeding and Blood Pressure: A Systematic Review 23
TABLE 2. Studies relating breastfeeding to blood pressure levels in infancy (before 12 months of age), by year of publication
* M, male; F, female; SBP, systolic blood pressure; DBP, diastolic blood pressure.
economic factors showed smaller systolic blood pressure pressure (54) and may influence infant feeding practices
differences between breast- and bottle-fed subjects. The (55). In the only study to examine this issue (43), the associ-
distribution of breastfeeding was less socially determined ation of breastfeeding with blood pressure was not altered by
before World War II (52) compared with now (10), and postnatal growth.
results from prewar cohorts may be free from confounding
by social class (53). The two prewar studies reviewed Relevance to contemporary cohorts
showed little evidence of any association between breast-
feeding and blood pressure (13, 19), although nondifferential Modern formula feeds, which more closely resemble the
misclassification is a possibility in the Caerphilly cohort that nutrient content of breast milk, were not developed until the
relied on recall of breastfeeding status 4460 years after mid-1970s (56). Previously, bottle-fed infants were given
infancy (19), and the Dutch famine cohort may not be gener- unmodified cows milk preparations and other alternatives
alizable (13). Accelerated postnatal weight gain is a potential such as condensed milk (52, 57). Several studies of infants
confounding factor because it is associated with raised blood born since 1980, however, show a blood-pressure-lowering
Am J Epidemiol 2005;161:1526
24 Martin et al.
effect of breastfeeding (8, 15, 20, 25, 26, 43), suggesting that draft of the paper, and coordinated its completion under the
if the results are causal, they are relevant to modern cohorts. supervision of G. D. S. and D. G. The first draft was signifi-
cantly revised after comments from these two authors. All
authors contributed to and approved the final version.
Population health implications
Help in developing the electronic search of the MEDLINE
Reductions in population mean blood pressure levels of as and EMBASE databases was provided by Margaret Burke,
little as 2 mmHg could reduce the prevalence of hyperten- Cochrane Heart Group Trials Search Coordinator.
sion by up to 17 percent, the number of coronary heart
disease events by 6 percent, and strokes and transient
ischemic attacks by 15 percent (9, 58). This reduction
equates to preventing 3,000 coronary heart disease events REFERENCES
and 2,000 strokes annually among those under age 75 years 1. Stary HC. Lipid and macrophage accumulations in arteries of
in the United Kingdom (59). The effect estimates from our children and the development of atherosclerosis. Am J Clin
meta-analysis could therefore translate into the prevention of Nutr 2000;72:1297S306S.
a substantial number of deaths annually. 2. Berenson GS, Srinivasan SR, Bao W, et al. Association
between multiple cardiovascular risk factors and atherosclero-
sis in children and young adults. N Engl J Med 1998;338:1650
Mechanisms 6.
3. Zinner SH, Martin LF, Sacks F, et al. A longitudinal study of
Am J Epidemiol 2005;161:1526
Breastfeeding and Blood Pressure: A Systematic Review 25
17. Simmons D. NIDDM and breastfeeding. Lancet 1997;350:157 Onesti G, Kim KE, eds. Hypertension in the young and the old.
8. New York, NY: Grune and Stratton, 1981:15563.
18. Deeks JJ, Altman DG, Bradburn MJ. Statistical methods for 38. Lucas A, Morley R, Hudson GJ, et al. Early sodium intake and
examining heterogeneity and combining results from several later blood pressure in preterm infants. Arch Dis Child 1988;63:
studies in meta-analysis. In: Egger M, Davey Smith G, Altman 6567.
DG, eds. Systematic reviews in health care: meta-analysis in 39. de Swiet M, Shinebourne EA. Blood pressure in infancy. Am
context. London, United Kingdom: BMJ Publishing Group, Heart J 1977;94:399401.
2003:285312. 40. Schachter J, Kuller LH, Perkins JM, et al. Infant blood pressure
19. Martin RM, Ben-Shlomo Y, Gunnell D, et al. Breastfeeding and heart rate: relation to ethnic group (black or white), nutri-
and cardiovascular disease risk factors, incidence and mortal- tion and electrolyte intake. Am J Epidemiol 1979;110:20518.
ity: the Caerphilly Study. J Epidemiol Community Health (in 41. Fall CHD, Osmond C, Barker DJP, et al. Fetal and infant
press). growth and cardiovascular risk factors in women. BMJ 1995;
20. Lawlor DA, Najman JM, Sterne J, et al. Associations of paren- 310:42832.
tal, birth, and early life characteristics with systolic blood pres- 42. Viikari J, Akerblom HK, Rasanen L, et al. Cardiovascular risk
sure at 5 years of age: findings from the Mater-University study in young Finns. Experiences from the Finnish Multicentre
of pregnancy and its outcomes. Circulation 2004;110:241723. Study regarding the prevention of coronary heart disease. Acta
21. Sterne JAC, Egger M, Davey Smith G. Investigating and deal- Paediatr Scand Suppl 1990;365:1319.
ing with publication and other biases. In: Egger M, Davey 43. Martin RM, Ness AR, Gunnell D, et al. Does breastfeeding in
Smith G, Altman DG, eds. Systematic reviews in health care: infancy protect against obesity in childhood? The Avon Longi-
meta-analysis in context. London, United Kingdom: BMJ Pub- tudinal Study of Parents and Children. Circulation 2004;109:
lishing Group, 2001:189208. 125966.
Am J Epidemiol 2005;161:1526
26 Martin et al.
Hypertension 1991;17:I16I20. 65. Singhal A, Fewtrell M, Cole TJ, et al. Low nutrient intake and
59. Petersen S, Rayner M, Press V. Coronary heart disease statis- early growth for later insulin resistance in adolescents born pre-
tics: 2000 edition. London, United Kingdom: British Heart term. Lancet 2003;361:108997.
Foundation, 2001. 66. Reaven G, Hoffman BB. A role for insulin in the aetiology and
60. Hofman A, Hazebroek A, Valkenburg HA. A randomized trial course of hypertension? Lancet 1987;2:4357.
of sodium intake and blood pressure in newborn infants. JAMA 67. Martin RM, Davey Smith G, Mangtani P, et al. Association
1983;250:3703. between breast feeding and growth: The Boyd-Orr cohort
61. Lucas A, Sarson DL, Blackburn AM, et al. Breast vs. bottle: study. Arch Dis Child Fetal Neonatal Ed 2002;87:F193F201.
endocrine responses are different with formula feeding. Lancet
68. Wadsworth ME, Hardy RJ, Paul AA, et al. Leg and trunk length
1980;1:12679.
62. Salmenpera L, Perheentupa J, Siimes MA. Exclusively breast- at 43 years in relation to childhood health, diet and family cir-
fed healthy infants grow slower than reference infants. Pediatr cumstances; evidence from the 1946 national birth cohort. Int J
Res 1985;19:30712. Epidemiol 2002;31:38390.
63. Axelsson IE, Ivarsson SA, Raiha NC. Protein intake in early 69. Gunnell D, Whitley E, Upton MN, et al. Associations of height,
infancy: effects on plasma amino acid concentrations, insulin leg length, and lung function with cardiovascular risk factors in
metabolism, and growth. Pediatr Res 1989;26:61417. the Midspan Family Study. J Epidemiol Community Health
64. Langenberg C, Hardy R, Kuh D, et al. Influence of height, leg 2003;57:1416.
and trunk length on pulse pressure, systolic and diastolic blood 70. Singhal A, Lucas A. Early origins of cardiovascular disease: is
pressure. J Hypertens 2003;21:53743. there a unifying hypothesis? Lancet 2004;363:16425.
Am J Epidemiol 2005;161:1526