Breastfeeding persuasive speech

Series

www.thelancet.com Vol 387 January 30, 2016 475

Breastfeeding 1

Breastfeeding in the 21st century: epidemiology, mechanisms, and lifelong eff ect Cesar G Victora, Rajiv Bahl, Aluísio J D Barros, Giovanny V A França, Susan Horton, Julia Krasevec, Simon Murch, Mari Jeeva Sankar, Neff Walker, Nigel C Rollins, for The Lancet Breastfeeding Series Group*

The importance of breastfeeding in low-income and middle-income countries is well recognised, but less consensus exists about its importance in high-income countries. In low-income and middle-income countries, only 37% of children younger than 6 months of age are exclusively breastfed. With few exceptions, breastfeeding duration is shorter in high-income countries than in those that are resource-poor. Our meta-analyses indicate protection against child infections and malocclusion, increases in intelligence, and probable reductions in overweight and diabetes. We did not fi nd associations with allergic disorders such as asthma or with blood pressure or cholesterol, and we noted an increase in tooth decay with longer periods of breastfeeding. For nursing women, breastfeeding gave protection against breast cancer and it improved birth spacing, and it might also protect against ovarian cancer and type 2 diabetes. The scaling up of breastfeeding to a near universal level could prevent 823 000 annual deaths in children younger than 5 years and 20 000 annual deaths from breast cancer. Recent epidemiological and biological fi ndings from during the past decade expand on the known benefi ts of breastfeeding for women and children, whether they are rich or poor.

Introduction “In all mammalian species the reproductive cycle comprises both pregnancy and breast-feeding: in the absence of latter, none of these species, man included, could have survived”, wrote paediatrician Bo Vahlquist in 1981.1 3 years earlier, Derek and Patrice Jelliff e in their classic book Breast Milk in the Modern World2 stated that “breast-feeding is a matter of concern in both industrialised and developing countries because it has such a wide range of often underappreciated consequences”.3 The Jelliff es anticipated that breastfeeding would be relevant to “present-day interest in the consequences of infant nutrition on subsequent adult health”.3 These statements were challenged by the American Academy of Pediatrics, which in its 1984 report on the scientifi c evidence for breastfeeding stated that “if there are benefi ts associated with breast-feeding in populations with good sanitation, nutrition and medical care, the benefi ts are apparently modest”.4

In the past three decades, the evidence behind breastfeeding recommendations has evolved markedly (appendix p 3). Results from epidemiological studies and growing knowledge of the roles of epigenetics, stem cells, and the developmental origins of health and disease lend strong support to the ideas proposed by Vahlquist and the Jelliff es. Never before in the history of science has so much been known about the complex importance of breastfeeding for both mothers and children.

Here, in the fi rst of two Series papers, we describe present patterns and past trends in breastfeeding throughout the world, review the short-term and long- term health consequences of breastfeeding for the child and mother, estimate potential lives saved by scaling up breastfeeding, and summarise insights into how

breastfeeding might permanently shape individuals’ life course. The second paper in the Series5 covers the determinants of breastfeeding and the eff ectiveness of promotion interventions. It discusses the role of breast- feeding in HIV transmission and how knowledge about this issue has evolved in the past two decades, and examines the lucrative market of breastmilk substitutes, the environmental role of breastfeeding, and its economic implications. In the context of the post-2015 development agenda, the two articles document how essential breastfeeding is for building a better world for future generations in all countries, rich and poor alike.

Lancet 2016; 387: 475–90

See Editorial page 404

See Comment pages 413 and 416

This is the first in a Series of two papers about breastfeeding

*Members listed at the end of the paper

International Center for Equity in Health, Post-Graduate Programme in Epidemiology, Federal University of Pelotas, Pelotas, Brazil (Prof C G Victora MD, Prof A J D Barros MD, G V A França PhD); Department of Maternal, Newborn, Child and Adolescent Health (MCA), WHO, Geneva, Switzerland (R Bahl MD, N C Rollins); Department of Economics, University of Waterloo, ON, Canada (Prof S Horton PhD); Data and Analytics Section, Division of Data, Research, and Policy, UNICEF, New York, NY, USA (J Krasevec MSc); University Hospital Coventry and Warwickshire, Coventry, UK (Prof S Murch PhD); WHO Collaborating Centre for Training and Research in Newborn Care, All India Institute of Medical Sciences (AIIMS), New Delhi, India (M J Sankar DM); and Institute for International Programs, Bloomberg School of Public Health, Baltimore, MD, USA (N Walker PhD)

Correspondence to: Prof Cesar G Victora, International Center for Equity in Health, Post-Graduate Programme in Epidemiology, Federal University of Pelotas, Pelotas, RS, 96020, Brazil cvictora@equidade.org

See Online for appendix

Search strategy and selection criteria

We obtained information about the associations between breastfeeding and outcomes in children or mothers from 28 systematic reviews and meta-analyses, of which 22 were commissioned for this review. See appendix pp 23–30 for the databases searched and search terms used. We reviewed the following disorders for young children: child mortality; diarrhoea incidence and admission to hospital; lower respiratory tract infections incidence, prevalence, and admission to hospital; acute otitis media; eczema; food allergies; allergic rhinitis; asthma or wheezing; infant growth (length, weight, body-mass index); dental caries; and malocclusion. For older children, adolescents, and adults, we did systematic reviews for systolic and diastolic blood pressure; overweight and obesity; total cholesterol; type 2 diabetes; and intelligence. For mothers, we did systematic reviews covering the following outcomes: lactational amenorrhoea; breast and ovarian cancer; type 2 diabetes; post-partum weight change; and osteoporosis.

http://crossmark.crossref.org/dialog/?doi=10.1016/S0140-6736(15)01024-7&domain=pdf

Series

476 www.thelancet.com Vol 387 January 30, 2016

Breastfeeding indicators and data sources for this review WHO has defi ned the following indicators for the study of feeding practices of infants and young children:6 early initiation of breastfeeding (proportion of children born in the past 24 months who were put to the breast within an hour of birth); exclusive breastfeeding under 6 months (proportion of infants aged 0–5 months who are fed exclusively with breastmilk. This indicator is based on the diets of infants younger than 6 months during the 24 h before the survey [to avoid recall bias], not on the proportion who are exclusively breastfed for the full 6-month period); continued breastfeeding at 1 year (proportion of children aged 12–15 months who are fed breastmilk); and continued breastfeeding at 2 years (proportion of children aged 20–23 months who are fed breastmilk).

Because few high-income countries report on the aforementioned indicators, we calculated additional indicators to allow global comparisons: ever breastfed (infants reported to have been breastfed, even if for a short period); breastfed at 6 months (in high-income countries, the proportion of infants who were breastfed from birth to 6 months or older; in low-income and middle-income countries [LMICs] with standardised surveys, the proportion of infants aged 4–7 months [median age of 6 months] who are breastfed); and breastfed at 12 months (in high-income countries, the proportion of children breastfed for 12 months or longer;

in LMICs, the proportion of children aged 10–13 months [median age of 12 months] who are breastfed).

For this review, we used the last three, additional indicators for comparisons between high-income countries and LMICs only. Otherwise, we reported on the standard international indicators (appendix p 4).

For LMICs, we reanalysed national surveys done since 1993, including Demographic and Health Surveys, Multiple Indicator Cluster Surveys, and others (appendix pp 5–12). Nearly all surveys had response rates higher than 90% and used standardised questionnaires and indicators.

For all high-income countries with 50 000 or more annual births, we did systematic reviews of published studies and the grey literature and contacted local researchers or public health practitioners when data from a particular country were not available or when there was ambiguity (appendix pp 13–17). Information about breastfeeding from national samples was not available from many countries. Although 27 out of 35 countries had some information about breastfeeding at a national level, response rates were often in the 50–70% range, indicators were rarely standardised, and recall periods tended to be long. We used administrative or other data when surveys were not available. If necessary, we estimated the proportion of infants breastfed at 12 months on the basis of information available for breastfeeding at 6 months and vice versa. We calculated time trends using multilevel linear regression models (hierarchical mixed models) that take into account that two or more surveys were included in the analyses for each country. We explored departures from linearity with fractional polynomial regression models.7 In all analyses, we weighted country data by their populations of children younger than 2 years of age (see appendix pp 18–22 for statistical methods).

We did systematic searches of the published literature, and, when possible, meta-analyses for outcomes postulated to be associated with breastfeeding (appendix pp 23–30). These systematic reviews and meta-analyses were specially commissioned by WHO to provide background information for this Series.

We used the Lives Saved Tool8 to predict how many deaths of children younger than 5 years would be prevented if breastfeeding patterns as of 2013 were scaled up in the 75 countries that are part of the Countdown to 2015 eff ort,9 which account for more than 95% of all such deaths worldwide. We assumed that 95% of children younger than 1 month and 90% of those younger than 6 months would be exclusively breastfed, and that 90% of those aged 6–23 months would be partly breastfed. We applied the relative risks for the protection against all infectious causes of death obtained from our new meta-analyses10 to all infectious causes of death in children younger than 2 years, and also to the 15% of deaths caused by complications of prematurity that occur after the fi rst week of life

Key messages

• Children who are breastfed for longer periods have lower infectious morbidity and mortality, fewer dental malocclusions, and higher intelligence than do those who are breastfed for shorter periods, or not breastfed. This inequality persists until later in life. Growing evidence also suggests that breastfeeding might protect against overweight and diabetes later in life.

• Breastfeeding benefi ts mothers. It can prevent breast cancer, improve birth spacing, and might reduce a woman’s risk of diabetes and ovarian cancer.

• High-income countries have shorter breastfeeding duration than do low-income and middle-income countries. However, even in low-income and middle-income countries, only 37% of infants younger than 6 months are exclusively breastfed.

• The scaling up of breastfeeding can prevent an estimated 823 000 child deaths and 20 000 breast cancer deaths every year.

• Findings from studies done with modern biological techniques suggest novel mechanisms that characterise breastmilk as a personalised medicine for infants.

• Breastfeeding promotion is important in both rich and poor countries alike, and might contribute to achievement of the forthcoming Sustainable Development Goals.

For the Demographic and Health Surveys see http://www. measuredhs.com/aboutsurveys/

dhs/start.cfm

For the Multiple Indicator Cluster Surveys see http://mics.

unicef.org/surveys

Series

www.thelancet.com Vol 387 January 30, 2016 477

(appendix pp 31–36). We also estimated the potential number of deaths from breast cancer that could have been prevented by extending the duration of breastfeeding (appendix pp 37–38).

Epidemiology: levels and trends We obtained complete information about 127 of the 139 LMICs (appendix pp 5–12), accounting for 99% of children from such countries. For high-income countries, we obtained data for 37 of 75 countries, but for several countries, only a subset of the indicators were available (appendix pp 13–17): these data should, therefore, be interpreted with caution.

Globally, the prevalence of breastfeeding at 12 months is highest in sub-Saharan Africa, south Asia, and parts of Latin America (fi gure 1). In most high-income countries, the prevalence is lower than 20% (appendix pp 13–17). We noted important diff erences—eg, between the UK (<1%) and the USA (27%), and between Norway (35%) and Sweden (16%).

We assessed breastfeeding indicators according to country income groups (fi gure 2). Information about early initiation or exclusive or continued breastfeeding at 2 years was not available for most high-income countries. We noted a strong inverse correlation (Pearson’s r=–0·84; p<0·0001; appendix p 39) between breastfeeding at 6 months and log gross domestic product per person; our regression analyses showed that for each doubling in the gross domestic product per head, breastfeeding prevalence at 12 months decreased by ten percentage points.

Most mothers in all country groups started breastfeeding; only three countries (France, Spain, and the USA) had rates below 80% for ever breastfeeding. However, early

initiation was low in all settings, as was exclusive breastfeeding (fi gure 2). Breastfeeding at 12 months was widespread in low-income and lower-middle-income settings, but uncommon elsewhere.

Except for early initiation, prevalence of all indicators decreased with increasing national wealth. Low-income countries had a high prevalence of breastfeeding at all ages, but the rates of initiation and exclusive breastfeeding are unsatisfactory even in these countries.

Surprisingly, most national level breastfeeding indicators were not strongly correlated (appendix p 39). We found only a moderate correlation (Pearson’s r=0·54) between exclusive and continued breastfeeding at 1 year in LMICs. Although the prevalence of

Figure 1: Global distribution of breastfeeding at 12 months Data are from 153 countries between 1995 and 2013.

0

Percentage of children who receive any breastmilk at 12 months of age (%)

10 20 30 40 50 60 70 80 90 100 No data

Figure 2: Breastfeeding indicators by country income group in 2010 Data are from national surveys that used standard indicators, and were weighted by national populations of children under 2 years. Data for up to 153 countries.

Early initiation of breastfeeding

Ever breastfed

Exclusive breastfeeding at 0–5 months

Breastfeeding at 6 months

Breastfeeding at 12 months

Continued breastfeeding 20–23 months

0 20 40 60 80 100 Percentage of children

Low income Lower-middle income Upper-middle income High income

Series

478 www.thelancet.com Vol 387 January 30, 2016

continued breastfeeding was high throughout west and central Africa, rates of exclusive breastfeeding varied widely (fi gure 3). Countries from eastern and southern Africa tended to have on average lower rates of continued breastfeeding but higher rates of exclusive breastfeeding than did those in west Africa. In Latin America and the Caribbean, and in central and eastern Europe and the Commonwealth of Independent States, both indicators tended to be lower than in Africa. South Asian countries had high rates of both indicators whereas countries in the Middle East and north Africa had lower rates. Countries from east Asia and the Pacifi c region had moderate to high prevalence of both indicators.

In children younger than 6 months in LMICs, 36·3 million (63%) were not exclusively breastfed at the time of the most recent national survey. The corresponding percentages were 53% in low-income countries, 61% in lower-middle-income countries, and 63% in upper-middle-income countries. In children aged 6–23 months in LMICs, 64·8 million (37%) were not receiving any breastmilk at the time of the most recent national survey, with corresponding rates of 18% in low-income, 34% in lower-middle-income, and 55% in upper-middle-income countries. 101·1 million children in LMICs were not breastfed according to international recommendations.

In most LMICs, data were available from several surveys over time, making it possible to explore time trends both at the national level and for children in the poorest and richest 20% of families. Our analyses were restricted to surveys for which breakdown of

breastfeeding indicators by wealth quintiles was possible (214 surveys for exclusive and 217 for continued breastfeeding; appendix pp 18–22), accounting for 83% of the total 2010 population of children younger than 2 years of age in LMICs. We reported linear trends because there was no evidence of departures from linearity. Exclusive breastfeeding rates increased slightly from 24·9% in 1993 to 35·7% in 2013 (fi gure 4). In the richest 20% of families, the increase was much steeper, whereas the poorest 20% followed the general trend. Continued breastfeeding at 1 year (12–15 months) dropped slightly at the global level (from 76·0% to 73·3%), partly due to a decline among the poorest 20% in each country (fi gure 4).

Epidemiology: within-country inequalities We analysed 98 surveys from LMICs to investigate within-country inequalities according to wealth quintile (appendix p 40). Wealth-related inequalities in exclusive breastfeeding were small but disparities in continued breastfeeding rates were consistent: poorer people tend to breastfeed for longer than their richer counterparts in all country groupings, but especially in middle-income countries. Similar results based on 33 countries have been reported elsewhere.11

Our review of studies from high-income countries showed that high-income, better-educated women breast- feed more commonly than do those in low-income groups with fewer years of formal education.12–20 Breastfeeding initiation in the USA was more common in mothers with lower education up until the 1960s, but the social gradient has since reversed.4

Figure 3: The relation between exclusive breastfeeding at 0–5 months and continued breastfeeding at 12–15 months, by region Datapoints are countries (values from the most recent survey from 117 countries, 2000–13) and are coloured according to their region. The shaded ellipses include at least 80% of the points in each region.

0 20 40 60 80 100 0

20

40

60

100

80

Co nt

in ue

d br

ea st

fe ed

in g

at 1

2– 15

m on

th s (

% )

Exclusive breastfeeding at 0–5 months (%)

Central and eastern Europe and the Commonwealth of Independent States East Asia and Pacific Eastern and southern Africa Latin America and the Caribbean Middle East and north Africa South Asia West and central Africa

Series

www.thelancet.com Vol 387 January 30, 2016 479

Breastfeeding is one of few positive health-related behaviours in LMICs that is less frequent in rich people, both between and within countries. The low rates of continued breastfeeding in richer families raises the possibility that poorer mothers will move towards breastmilk substitutes as their income increases, a concern that is reinforced by decreasing rates in poor populations.

Short-term eff ects in children: mortality and morbidity The results of 28 meta-analyses on the associations between breastfeeding and outcomes in the children and mothers, of which 22 were commissioned for this Series, are summarised in the table. Because studies varied with regard to their feeding classifi cations, for several outcomes we compared longer versus shorter breastfeeding dur- ations (eg, never vs ever breastfed, breastfed for less or more than a given number of months, and for a few outcomes longer vs shorter durations of exclusive breastfeeding). We tested for heterogeneity due to the type of breastfeeding categorisation, and in its absence we pooled the diff erent studies. We described the results of randomised trials on how breastfeeding promotion aff ects health, nutrition, or developmental outcomes, but not of trials in which the endpoint was restricted to breastfeeding indicators; these are reviewed in the second article in the Series.5

Only three studies in LMICs provide information about mortality according to exclusive, predominant, partial, or no breastfeeding in the fi rst 6 months of life (table). A strong protective eff ect was evident, with exclusively breastfed infants having only 12% of the risk of death compared with those who were not breastfed.10 Another three studies in LMICs showed that infants younger than 6 months who were not breastfed had 3·5-times (boys) and 4·1-times (girls) increases in mortality compared with those who received any breastmilk, and that that protection decreased with age.33 These results are lent support by studies of children aged 6–23 months, in whom any breastfeeding was associated with a 50% reduction in deaths (table).

Breastfeeding might also protect against deaths in high-income countries. A meta-analysis of six high-quality studies showed that ever breastfeeding was associated with a 36% (95% CI 19–49) reduction in sudden infant deaths.34 Another meta-analysis of four randomised controlled trials showed a 58% (4–82) decrease in necrotising enterocolitis,34 a disorder with high case-fatality in all settings.35

In terms of child morbidity, overwhelming evidence exists from 66 diff erent analyses, mostly from LMICs and including three randomised controlled trials, that breastfeeding protects against diarrhoea and respiratory infections (table).21 About half of all diarrhoea episodes and a third of respiratory infections would be avoided by breastfeeding. Protection against hospital admissions

due to these disorders is even greater: breastfeeding could prevent 72% of admissions for diarrhoea and 57% of those for respiratory infections. We discuss the risks associated with breastmilk substitutes in terms of biological and chemical contamination in appendix p 41.

Our reviews suggest important protection against otitis media in children younger than 2 years of age, mostly from high-income settings, but inconclusive fi ndings for older children (table).22 We saw no clear evidence of protection against allergic disorders: no association with eczema or food allergies and some evidence of protection against allergic rhinitis in children younger than 5 years.23 When we analysed the 29 studies of asthma, we noted statistically signifi cant evidence of a 9% (95% CI 2–15) reduction in asthma with breastfeeding, but eff ects were smaller and non-signifi cant when we restricted analyses to the 16 studies with tighter control of confounding (a reduction of 5% [−6 to 15]) or to the 13 cohort studies (6% reduction [−11 to 20]).

On the basis of 49 studies done mostly in LMICs, our analyses of oral health outcomes (table) showed that breastfeeding was associated with a 68% reduction (95% CI 60–75) in malocclusions.26 Most studies were restricted to young children with deciduous teeth, but malocclusion in this age group is a risk factor for malocclusion in permanent (adult) teeth.36,37 However, breastfeeding for longer than 12 months and nocturnal feeding were associated with 2–3-times increases in dental caries in deciduous teeth, possibly due to inadequate oral hygiene after feeding.25

Figure 4: National and wealth quintile-specifi c time trends in exclusive and continued breastfeeding, 1993–2013 Data are weighted by national populations of children younger than 2 years at the time of the survey. Analyses restricted to 66 countries with information about household wealth.

1990 1995 2000 2005 2010 2015 0

20

40

60

80

100

Pr ev

al en

ce (%

)

Year Exclusive breastfeeding 0–5 months (national) Exclusive breastfeeding 0–5 months (poorest quintiles) Exclusive breastfeeding 0–5 months (richest quintiles)

Continued breastfeeding 12–15 months (national) Continued breastfeeding 12–15 months (poorest quintiles) Continued breastfeeding 12–15 months (richest quintiles)

Series

480 www.thelancet.com Vol 387 January 30, 2016

Information about breastfeeding and child growth was derived from 17 studies, including 15 randomised controlled trials, mostly from middle-income countries.24 Attained weight and length at about 6 months did not diff er, but there was a small reduction (Z score −0·06 [95% CI –0·12 to 0·00]) in body-mass index (BMI) or bodyweight for length in children whose mothers received the breastfeeding promotion intervention compared with those whose mothers did not receive the promotion intervention (table).

Long-term eff ects in children: obesity, non-communicable diseases, and intelligence We updated existing meta-analyses38 on the associations between breastfeeding and outcomes related to non- communicable diseases (table). Most studies are from high-income settings. Based on all 113 studies identifi ed, longer periods of breastfeeding were associated with a 26% reduction (95% CI 22–30) in the odds of overweight or obesity.27 The eff ect was consistent across income classifi cations. The only breastfeeding promotion trial

Outcome Types of comparison (breastfeeding categories)

Studies (n)

Age range of outcome

Pooled eff ect (95% CI)

Confounding and eff ect modifi cation

Other biases Conclusions

Eff ects on children, adolescents, or adults according to breastfeeding pattern

Sankar et al (2015)10

Mortality due to infectious diseases

Exclusive versus predominant

3 <6 months OR 0·59 (0·41–0·85)

All studies from LMICs, where confounding by SEP would probably underestimate the eff ect of breastfeeding. Confounder-adjusted studies showed similar eff ects

Studies that avoided reverse causation (breastfeeding stopped because of illness) showed similar eff ects. No evidence of publication bias but very few studies available

Consistent evidence of major protection. Few studies used the four breastfeeding categories in young infants, but evidence from other studies comparing any versus no breastfeeding is very consistent

Sankar et al (2015)10

Mortality due to infectious diseases

Exclusive versus partial

3 <6 months OR 0·22 (0·14–0·34)

See above See above See above

Sankar et al (2015)10

Mortality due to infectious diseases

Exclusive versus none

2 <6 months OR 0·12 (0·04–0·31)

See above See above See above

Sankar et al (2015)10

Mortality due to infectious diseases

Any versus none 9 6–23 months OR 0·48 (0·38–0·60)

See above See above See above

Horta et al (2013)21

Diarrhoea incidence

More versus less breastfeeding (eg, exclusive vs non-exclusive; predominant vs partial; partial vs none; any breastfeeding vs no breastfeeding)

15 <5 years RR 0·69 (0·58–0·82)

Most studies were from LMICs, where confounding would probably underestimate an eff ect. Confounder-adjusted studies showed similar eff ects. Three RCTs of breastfeeding promotion (not included in the meta- analysis) showed protection against diarrhoea morbidity (pooled OR 0·69 [0·49–0·96])

Few studies that allowed for reverse causation also showed protection. Publication bias is unlikely to explain the fi ndings because results from large and small studies were similar

Strong evidence of major protection against diarrhoea morbidity and admissions to hospital, particularly in young infants, based on a large number of studies

Horta et al (2013)21

Diarrhoea incidence

See above 23 <6 months RR 0·37 (0·27–0·50)

See above See above See above

Horta et al (2013)21

Diarrhoea incidence

See above 11 6 months to 5 years

RR 0·46 (0·28–0·78)

See above See above See above

Horta et al (2013)21

Admission to hospital for diarrhoea

See above 9 <5 years RR 0·28 (0·16–0·50)

See above See above See above

Horta et al (2013)21

Lower respiratory infections (incidence or prevalence)

See above 16 <2 years RR 0·68 (0·60–0·77)

Most studies were from LMICs, where confounding would probably underestimate the eff ect of breastfeeding. Confounder- adjusted studies showed similar eff ects

Studies that avoided reverse causation showed similar eff ects. No evidence of publication bias

Strong evidence of a reduction in severe respiratory infections in breastfed children, based on a large number of studies

Horta et al (2013)21

Admissions to hospitals for respiratory infections

See above 17 <2 years RR 0·43 (0·33–0·55)

The only available RCT showed an RR of 0·85 (0·57–1·27), a non-signifi cant reduction in admissions to hospital

See above See above

(Table continues on next page)

Series

www.thelancet.com Vol 387 January 30, 2016 481

that reported on this outcome did not detect an association; in this trial, the investigators reported important early diff erences between inter vention and comparison groups in terms of exclusive breastfeeding, but at 12 months of age only 19% of children in the intervention group and 11% of children in the comparison

group were breastfed.39,40 A 2005 meta-analysis41 of breastfeeding and mean BMI included 36 articles of which 11 included adjustment for socioeconomic status, maternal smoking, and maternal BMI; their pooled eff ect did not suggest an association with breastfeeding. In our review,27 23 high-quality studies with sample sizes of

Outcome Types of comparison (breastfeeding categories)

Studies (n)

Age range of outcome

Pooled eff ect (95% CI)

Confounding and eff ect modifi cation

Other biases Conclusions

(Continued from previous page)

Bowatte et al (2015)22

Acute otitis media More versus less breastfeeding (ever vs never; exclusive breastfeeding at 6 months vs not exclusive breastfeeding at 6 months; any breastfeeding for ≥3–4 months vs <3–4 months)

11 ≤2 years OR 0·67 (0·62–0·72)

Egger’s test for small study eff ects showed weak evidence for publication bias (p=0·360)

Consistent evidence of reduction in acute otitis media during the fi rst 2 years of life associated with longer durations of breastfeeding, based on 11 studies. No evidence of protection after 2 years

Bowatte et al (2015)22

Acute otitis media See above 5 >2 years OR 1·21 (0·60–2·45)

Most studies were done in HICs. Several studies did not adjust for important confounders

High heterogeneity (I²=84%) among the fi ve studies of children older than 2 years

See above

Lodge et al (2015)23

Eczema More versus less breastfeeding (ever vs never; exclusive breastfeeding at 6 months vs not exclusive breastfeeding at 6 months; any breastfeeding for ≥3–4 months vs <3–4 months)

17 ≤2 years OR 0·95 (0·85–1·07)

About a third of the studies were from LMICs, and results are similar to those from HICs. Few studies in young children account for reverse causation. Several studies did not adjust for essential confounders

Some evidence of publication bias, with smaller pooled eff ect sizes in larger studies

No evidence of an association between breastfeeding and eczema

Lodge et al (2015)23

Ezcema See above 20 >2 years OR 1·09 (0·99–1·20)

See above See above See above

Lodge et al (2015)23

Food allergies See above 10 ≤5 years OR 1·07 (0·90–1·26)

See above The ten studies on food allergy in children ≤5 years were highly heterogeneous (I²=88%)

No evidence of an association between breastfeeding and food allergies

Lodge et al (2015)23

Food allergies See above 4 >5 years OR 1·08 (0·73–1·26)

See above See above See above

Lodge et al (2015)23

Allergic rhinitis See above 5 ≤5 years OR 0·79 (0·63–0·98)

See above See above Possible protection against allergic rhinitis in children <5 years, based on only fi ve studies

Lodge et al (2015)23

Allergic rhinitis See above 9 >5 years OR 1·05 (0·99–1·12)

See above See above No evidence for those older than 5 years

Lodge et al (2015)23

Asthma or wheezing

See above 29 5–18 years OR 0·91 (0·85–0·98)

The protective eff ect of asthma was smaller and not signifi cant in 16 studies with thorough control for confounders (OR 0·95 [0·85–1·06]) and in the 13 cohort studies (OR 0·94 [0·80–1·11]). There were too few studies to estimate association with asthma in adults

See above Inconclusive evidence for the association between breastfeeding and the risk of asthma or wheezing

(Table continues on next page)

Series

482 www.thelancet.com Vol 387 January 30, 2016

Outcome Types of comparison (breastfeeding categories)

Studies (n)

Age range of outcome

Pooled eff ect (95% CI)

Confounding and eff ect modifi cation

Other biases Conclusions

(Continued from previous page)

Giugliani et al (2015)24

Length Randomised trials or quasi- experiments comparing children receiving breastfeeding promotion interventions with control children

17 About 6 months (range 3–24)

Z score 0·03 (−0·02 to 0·08)

Most studies are from middle-income countries. Confounding is unlikely because 15 of the 17 studies were randomised trials. Analyses were by intent to treat, so that low compliance with breastfeeding promotion might underestimate the magnitude of the eff ect

Evidence of publication bias for BMI, with small studies showing larger reductions

No evidence of an eff ect on breastfeeding promotion on length at 6 months of age

Giugliani et al (2015)24

Weight See above 16 See above Z scores 0·03 (−0·06 to 0·12)

See above See above No evidence of an eff ect on breastfeeding promotion on weight at 6 months of age

Giugliani et al (2015)24

BMI or weight for length

See above 11 See above Z scores −0·06 (−0·12 to 0·00)

See above See above Some evidence supporting a reduction in BMI or weight for length

Tham et al (2015)25

Dental caries Breastfeeding >12 months versus ≤12 months

4 <6 years OR 2·69 (1·28–5·64)

Most studies did not control for the introduction of sugary foods and drinks. Most studies were from HICs, where high SEP would be expected to negatively confound the association

Publication biases veer toward studies that show an association between breastfeeding beyond 12 months or on demand and dental caries

Consistent evidence that breastfeeding >12 months has detrimental eff ects on deciduous teeth

Tham et al (2015)25

Dental caries Breastfeeding on demand or nocturnal feeding versus not (in breastfed children)

6 <6 years OR 2·90 (2·33–3·60)

See above See above Consistent evidence that breastfeeding on demand has detrimental eff ects on deciduous teeth

Peres et al (2015)26

Malocclusion Never versus ever breastfeeding; longer versus shorter duration of exclusive breastfeeding; or longer versus shorter duration of any breastfeeding

41 Childhood, adolescence, and adulthood

OR 0·32 (0·25–0·40)

80% of the studies were from LMICs. Because malocclusions are not associated with SEP or any other known determinant of breastfeeding patterns, it is unlikely that these results are aff ected by confounding

Some evidence of publication bias but the association was also present in the larger and better designed studies

Consistent evidence of a major, two-thirds reduction in malocclusions in deciduous teeth in breastfed individuals

Horta et al (2015)27

Systolic blood pressure

Never versus ever breastfed; or longer versus shorter breastfed duration

43 Childhood, adolescence and adulthood

−0·80 mm Hg (−1·17 to −0·43)

Three-quarters of the studies were from LMICs. Evidence of residual confounding as eff ect in studies from HIC but not in those from LMICs

Evidence of publication bias in systolic blood pressure studies

No evidence of a reduction in blood pressure associated with breastfeeding

Horta et al (2015)27

Diastolic blood pressure

Never versus ever breastfed; or longer versus shorter breastfeeding duration

38 Childhood, adolescence, and adulthood

−0·24 mm Hg (−0·50 to 0·02)

See above Evidence of publication bias in diastolic blood pressure studies

See above

Horta et al (2015)27

Overweight or obesity

Never versus ever breastfed; longer versus shorter duration of exclusive breastfeeding; or longer versus shorter duration of any breastfeeding

113 Childhood, adolescence, and adulthood

OR 0·74 (0·70–0·78)

In HICs, residual confounding by SEP is a possibility; however, the eff ect size was similar in studies from LMICs (a third of all studies). 23 high-quality studies showed a smaller pooled reduction of 13% (95% CI 6–19)

Some evidence of publication bias with larger eff ects in small studies, but even large and well controlled studies showed a 20% reduction in prevalence

Suggestive evidence of protection, including high-quality studies and those from low-income or middle-income settings

(Table continues on next page)

Series

www.thelancet.com Vol 387 January 30, 2016 483

Outcome Types of comparison (breastfeeding categories)

Studies (n)

Age range of outcome

Pooled eff ect (95% CI)

Confounding and eff ect modifi cation

Other biases Conclusions

(Continued from previous page)

Horta et al (2015)27

Total cholesterol Never versus ever breastfed; or longer versus shorter breastfeeding duration

46 Childhood, adolescence, and adulthood

−0·01 mmol/L (−0·05 to 0·02)

No evidence of heterogeneity with nearly all studies showing small eff ects. Three-quarters of the studies were from HICs

No evidence of an association

No evidence of an association

Horta et al (2015)27

Type 2 diabetes Never versus ever breastfed; longer versus shorter duration of exclusive breastfeeding; or longer versus shorter duration of any breastfeeding

11 Childhood, adolescence, and adulthood

OR 0·65 (0·49–0·86)

Only two of 11 studies were from LMICs; these studies showed 14% reduction; residual confounding might have aff ected HIC studies

Few available studies; no evidence of publication bias

Restricted evidence of protection, based on 11 studies

Horta et al (2015)28

Intelligence Never versus ever breastfed; or longer versus shorter breastfeeding duration

16 Childhood, adolescence, and adulthood

IQ points: 3·44 (2·30–4·58)

In HICs (14 of the 16 studies), residual confounding by SEP was a possibility; however, the eff ect was also present in two studies from LMICs. One high-quality RCT showed a statistically signifi cant increase in IQ of more than 7 points

Some evidence of publication bias with larger eff ects in small studies, but even large studies showed an eff ect. Nine studies with adjustment for maternal IQ showed diff erence of 2·62 points (1·25–3·98)

Consistent eff ect of about 3 IQ points in observational studies; also present a large RCT on this topic

Eff ects on women who breastfed

Chowdhury et al (2015)29

Lactational amenorrhoea

Highest versus lowest duration of breastfeeding

13 Women (<1 year post partum)

RR 1·17 (1·04–1·32)

Most studies were from LMICs. Residual confounding unlikely. Strongest eff ects when exclusive or predominant breastfeeding are compared with partial (RR 1·21) or no breastfeeding (RR 1·23)

No evidence of publication bias

Consistent eff ect on prolonging lactational amenorrhoea, especially for exclusive or predominant breastfeeding

Chowdhury et al (2015)29

Breast cancer Highest versus lowest duration of breastfeeding

76 Adult women OR 0·81 (0·77–0·86)

Three-quarters of the studies were from HICs. Parity reduces the risk of breast cancer and is also associated with greater lifetime breastfeeding duration. Most studies did not adjust appropriately for parity and therefore tended to exaggerate eff ect size. A thoroughly adjusted pooled analysis of 47 studies shows an OR of 0·96 for each 12 months of breastfeeding30

Some evidence of publication bias but the association was also present in the larger and better designed studies

Consistent protective eff ect of breastfeeding against breast cancer in 47 well designed studies, of a 4·3% reduction per 12 months of breastfeeding in the better controlled studies

Chowdhury et al (2015)29

Ovarian cancer Highest versus lowest duration of breastfeeding

41 Adult women OR 0·70 (0·64–0·75)

Only six studies from LMICs. Confounding by parity might aff ect the results but socioeconomic confounding is unlikely. Studies with fi ne adjustment for parity and exclusion of nulliparous women showed less protection with an OR of 0·82 (0·75–0·89)

Some evidence of publication bias, with smaller pooled eff ect sizes in the 22 studies with samples larger than 1500 women (OR 0·76 [0·69–0·84])

Suggestive evidence of a protective eff ect of breastfeeding

Chowdhury et al (2015)29

Osteoporosis (distal radius)

Highest versus lowest duration of breastfeeding

4 Adult women SDS −0·132 (−0·260 to –0·003)

All studies from HICs. High heterogeneity in the distal radius analyses with the largest study showing no association and smaller studies showing protection

Not assessed because of small number of studies

Insuffi cient evidence

(Table continues on next page)

Series

484 www.thelancet.com Vol 387 January 30, 2016

more than 1500 participants and adjustment for socioeconomic status, maternal BMI and perinatal morbidity showed a pooled reduction in the prevalence of overweight or obesity of 13% (95% CI 6–19).

For the incidence of type 2 diabetes, the pooled results from 11 studies indicate a 35% reduction (95% CI 14–51). We deemed only three studies to be of high quality, which indicated a potentially important, but not statistically signifi cant, reduction of 24% (95% CI ranging from a 60% reduction to a 47% increase).27 The direction and magnitude of the association with diabetes are consistent with fi ndings for overweight. An earlier review of six studies indicated a possible protective eff ect against type 1 diabetes.34 The meta- analyses for systolic (43 studies) and diastolic (38 studies) blood pressure, and total cholesterol (46 studies) showed no evidence of protective eff ects of breastfeeding.27

Breastfeeding was consistently associated with higher performance in intelligence tests in children and adolescents, with a pooled increase of 3·4 intelligence quotient (IQ) points (95% CI 2·3–4·6) based on the fi ndings of 16 observational studies that controlled for several confounding factors including home stimulation (table).28 Nine studies also adjusted for maternal intelligence, showing a pooled eff ect of 2·6 points (1·3–4·0). A large randomised trial reported an increase of more than 7 IQ points at 6·5 years of age,42 and a similar eff ect was reported in a non-randomised trial in

which preterm infants were fed formula or breastmilk.43 Positive associations with attained schooling were reported from the UK,44,45 New Zealand,46 and Brazil,47 but a joint analysis of four cohorts in LMICs showed mixed results.48 A study in Brazil including 30 years of follow-up suggested an eff ect of breastfeeding on intelligence, attained schooling, and adult earnings, with 72% of the eff ect of breastfeeding on income explained by the increase in IQ.49 A review of 18 studies suggested that breastfeeding is associated with a 19% reduction (95% CI 11–27) in the incidence of childhood leukaemia.50

Eff ects on the mother The table shows the results of new reviews (published in July, 2015) on lactational amenorrhoea, breast and ovarian cancer, type 2 diabetes, and osteoporosis.29 We also cite existing reviews on diabetes, weight retention, and maternal depression. Most studies were from high-income countries, except for those on lactational amenorrhoea.

The role of breastfeeding in birth spacing is well recognised. In 2003, it was estimated that in countries where continued breastfeeding is prevalent, eg, Uganda and Burkina Faso, 50% more births would be expected in the absence of breastfeeding.51 Our review confi rms that increased breastfeeding, and especially exclusive or predominant breastfeeding, were associated with longer periods of amenorrhoea.29 Findings from randomised controlled trials of breastfeeding promotion interventions also confi rm this eff ect.52

Outcome Types of comparison (breastfeeding categories)

Studies (n)

Age range of outcome

Pooled eff ect (95% CI)

Confounding and eff ect modifi cation

Other biases Conclusions

(Continued from previous page)

Chowdhury et al (2015)29

Osteoporosis (femoral neck)

Highest versus lowest duration of breastfeeding

4 Adult women SDS −0·142 (−0·426 to 0·142)

All studies from HICs. None of the studies showed an association

Not assessed because of small number of studies

Insuffi cient evidence

Aune et al (2013)31

Type 2 diabetes Highest versus lowest duration of breastfeeding

6 Adult women RR 0·68 (0·57–0·82)

Several confounding factors were adjusted for. Signifi cant protection also seen for 3-month and 12-month increases in breastfeeding duration. Five of the six studies were from HICs. All six studies showed protection

Few available studies; no evidence of publication bias

Restricted evidence of protection against type 2 diabetes in women who breastfed for longer periods

Neville et al (2014)32

Post-partum weight change

Qualitative review 45 Women (<2 years post partum)

Not estimated because of diff erent outcome measures at variable post- partum ages

Studies were highly variable. Most studies saw no association. Of the fi ve studies with high methodological quality, four reported benefi cial eff ects. Nearly all studies from HICs

Not assessed in the published review

The role of breastfeeding on post-partum weight change is uncertain

Data are odds ratio (95% CI), risk ratio (95% CI), Z score (95% CI), mm Hg (95% CI), mmol/L (95% CI), intelligence quotient (95% CI), or standard deviation scores (95% CI). In 22 sets of analyses, the summary eff ect sizes are the pooled results of studies comparing longer versus shorter breastfeeding durations (either never vs ever breastfed; exclusive breastfeeding for more than a specifi c number of months vs less than that number of months; or any breastfeeding for more than a specifi c number of months vs less than that number of months). Separate results for each type of categorisation are available in the appendix. OR=odds ratio. LMICs=low-income and middle-income countries. SEP=socioeconomic position. RR=risk ratio. RCT=randomised controlled trial. HICs=high-income countries. BMI=body-mass index. IQ=intelligence quotient. SDS=SD score.

Table: Results of meta-analyses on the associations between breastfeeding and outcomes in children and mothers

Series

www.thelancet.com Vol 387 January 30, 2016 485

Evidence exists of a robust inverse association between breastfeeding and breast cancer (table). The largest individual-level analysis on this topic included about 50 000 patients with cancer from 47 studies,30 which is about half those included in our meta-analysis. Each 12-month increase in lifetime breast feeding was associated with a reduction of 4·3% (95% CI 2·9–6·8) in the incidence of invasive breast cancer. This analysis included thorough adjustment parity and other confounders; nulliparous women were excluded. The results did not vary substantially according to menopausal status. Our meta-analysis suggests a higher magnitude of protection, but when restricted to the 14 studies with fi ne adjustment for parity and exclusion of nulliparous women, the reduction comparing longer versus shorter breastfeeding durations was 7% (95% CI 3–11).29

The meta-analysis of 41 studies on breastfeeding and ovarian cancer shows a 30% reduction associated with longer periods of breastfeeding (95% CI 25–36). Confounding by parity might aff ect the results but socio- economic confounding is unlikely because socioeconomic status is only weakly associated with ovarian cancer incidence. The pooled reduction, based on studies with fi ne adjustment for parity and exclusion of nulliparous women, was 18% (14–42).29 We also reviewed the evidence on osteoporosis, fi nding no evidence of an association between breastfeeding and bone mineral density in the four studies available (table).29

A meta-analysis of six cohort studies on type 2 diabetes showed an odds ratio of 0·68 (95% CI 0·57–0·82).31 In view of this fi nding, an association could be predicted with overweight, but a review of 54 articles on the possible role of breastfeeding on post- partum weight change was inconclusive.32 Few studies are available for the long-term association between nursing and adiposity. After the review of studies on overweight and breastfeeding was published, an analysis of 740 000 British women with long-term follow-up showed that mean BMI was 1% lower for every 6 months that the woman had breastfed.53 A qualitative review of 48 studies showed clear associations between breastfeeding and reduced mat- ernal depression,54 but it is more likely that depression aff ects breastfeeding than the opposite.

Estimating lives saved for children and mothers The Lives Saved Tool8 estimates that 823 000 annual deaths would be saved in 75 high-mortality LMICs in 2015 if breastfeeding was scaled up to near universal levels. This corresponds to 13·8% of the deaths of children under 2 years of age. For preventable deaths, 87% would have occurred in infants younger than 6 months due to a combination of high death rates and low prevalence of exclusive breastfeeding.

We also estimated the potential eff ect of breastfeeding on breast cancer mortality (appendix pp 31–37). Using the estimates of protection from the pooled study, we estimate

that existing global rates of breastfeeding avert 19 464 annual breast cancer deaths compared with a scenario in which no women breastfed (table).30 The low-income regions with long breastfeeding durations (Africa and south Asia) account for 58% of currently prevented deaths, despite only accounting for 36% of the global population included in this analysis. We also estimate that an additional 22 216 lives per year would be saved by increasing breastfeeding duration from present levels to 12 months per child in high-income countries and 2 years per child in LMICs. We cannot model the same eff ect in all countries given the diff erences in data availability and the fact that very few children in high-income countries are breastfed for longer than 12 months. Latin America, central and eastern Europe, the Commonwealth of Independent States, and high- income countries would benefi t most because of their higher incidence of breast cancer and also shorter breastfeeding durations at present.