School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
ALSPAC AND CROSS-COHORT STUDIES
Causal Effects of Breastfeeding onChild Health Outcomes
Marie-Jo Brion
Sir Henry Wellcome Postdoctoral Fellow
MRC Centre for Causal Analyses in Translational Epidemiology
School of Social and Community Medicine
University of Bristol
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Cause and Effect in Observational Studies:
Getting the Right Answer Examples where observational studies and RCTs fail to arrive at
the same answer
Protective effects on CVD from observational studies : Beta carotene Vitamin E supplements Vitamin C supplements Hormone replacement therapy
Large RCTs showed no protective effect
Likely explanation: confounding in particular, confounding by socioeconomic position
** NEED BETTER CAUSAL APPROACHES **
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
What are the causal effects of breastfeeding
on IQ, obesity and blood pressure?
Evidence from comparing high-income
with middle-income cohortsMarie-Jo Brion, Debbie Lawlor, Alicia Matijasevich, Bernardo Horta, Luciana Anselmi,
Ana Maria B Menezes, Cesar Victora, George Davey Smith
Brion et al (2011) IJE 40(3): 670
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Objectives
Explore novel approach for improving causal inference in observational studies
Comparing associations from HIC to LMIC cohorts where associations of confounders with health outcomes are likely to differ between cohorts
Assessing causal effects of breastfeeding on child BP, BMI and IQ
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
higher socioeconomic position
healthy diet
physical activity
healthier families
education
Causal effects ??
less smoking
better living conditions
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Methods ALSPAC, UK (N~5000) and Pelotas 1993, Brazil (N~1000)
Meta-analyses of 5 LMIC (COHORTS consortium; N~11,000)Fall et al., IJE 2011
Randomised trial of breastfeeding promotion (PROBIT Belarus trial)Kramer et al., Arch Gen Psychiatry 2008; Kramer et al., AJCN 2007
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Analyses Associations of indicators of SEP with breastfeeding
Maternal education paternal education family income occupational social class
Associations of breastfeeding duration with child outcomes:
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Prevalence (%)
Breastfeeding duration Pelotas ALSPAC
0 to <1 month 15.6% 36.8 %
1 to < 3 months 25.4% 15.6 %
3 to < 6 months 23.6% 13.7 %
6 months or more 35.3% 33.9 %
Ever breastfed 96.8% 84.3%
Exclusive breastfeeding at 2 months 65.7% 38.3%
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Breastfeeding at 3 months according to family income
2030
4050
6070
8090
100
1 2 3 4 5
Income (Low to High)
Bre
as
tfe
ed
ing
at
3 m
on
ths
(%
)
Pelotas
ALSPAC
Ever breastfed according to family income
3040
50607080
90100
1 2 3 4 5
Income (Low to High)P
erc
en
t e
ve
r b
rea
stf
ed
(%
) Pelotas
ALSPAC
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Relative/Slope Index of Inequality
Logistic regression: SEP indicator with binary outcomes Odds of outcome in highest SEP level (1) versus the lowest (0)
Linear regression: SEP indicator with continuous outcomes Difference in outcome between highest SEP level (1) and the lowest (0)
3040
5060
70
Bre
astfe
edin
g pr
eval
ence
0 .2 .4 .6 .8 1
Cumulative proportion of the population according to income
ALSPAC Pelotas
Breastfeeding by income
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
01
23
45
67
89
10O
R
ALSPAC PelotasCohort
ORs for breastfeeding by social class
02
46
810
1214
16O
R
ALSPAC PelotasCohort
ORs for breastfeeding by maternal education
01
23
45
67
89
10O
R
ALSPAC PelotasCohort
ORs for breastfeeding by paternal education
01
23
45
67
89
10O
R
ALSPAC PelotasCohort
ORs for breastfeeding by income
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
-6-4
-20
-6-4
-20
ALSPAC Pelotas ALSPAC Pelotas
class income
maternal education paternal education
diffe
renc
e in
SB
P (
mm
Hg
)Systolic BP
-4-2
02
-4-2
02
ALSPAC Pelotas ALSPAC Pelotas
class income
maternal education paternal education
diffe
renc
e in
DB
P (
mm
Hg)
cohort
Diastolic BP-1
01
2-1
01
2
ALSPAC Pelotas ALSPAC Pelotas
class income
maternal education paternal education
diffe
renc
e in
BM
I (kg
/m2
)
510
1520
255
1015
2025
ALSPAC Pelotas ALSPAC Pelotas
class income
maternal education paternal educationdi
ffere
nce
in IQ
IQBMI
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
higher socioeconomic position
healthy diet
physical activity
healthier families
education
Causal effects ??
less smoking
better living conditions
Discordant ALSPAC-Pelotas
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Mean difference in outcome by breastfeeding category
OutcomeALSPAC
β (95% CI) P trend
Pelotas
β (95% CI) P trend
SBP, -0.35 (-0.55, -0.14) 0.001 -0.13 (-0.83, 0.57) 0.7(mmHg)
DBP -0.16 (-0.31, -0.01) 0.04 0.05 (-0.50, 0.60) 0.9(mmHg)
BMI -0.16 (-0.22, -0.09) <0.001 0.14 (-0.07, 0.36) 0.2(kg/m2)
IQ 0.97 (0.62, 1.32) <0.001 1.97 (0.88, 3.05) <0.001
Breastfeeding categories: 0 to <1m; 1 to <3m; 3 to <6m; 6m or more
Fully adjusted models
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
SBP (mmHg)ALSPACPelotas
p hetero = 0.6
DBP (mmHg)ALSPACPelotas
p hetero = 0.5
BMI (kg/m2)ALSPACPelotas
p hetero = 0.009
IQALSPACPelotas
p hetero = 0.09
CHILD OUTCOME
-0.35 (-0.55, -0.14)-0.13 (-0.83, 0.57)
-0.16 (-0.31, -0.01) 0.05 (-0.50, 0.60)
-0.16 (-0.22, -0.09) 0.14 (-0.07, 0.36)
0.97 (0.62, 1.32)1.97 (0.88, 3.05)
EFFECT SIZE (95% CI)
0-1 -.5 0 .5 1 1.5 2 2.5 3 3.5
ALSPAC-PELOTAS BREASTFEEDING ASSOCIATIONS
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
SBP (mmHg)ALSPACPelotas
p hetero = 0.6
DBP (mmHg)ALSPACPelotas
p hetero = 0.5
BMI (kg/m2)ALSPACPelotas
p hetero = 0.009
IQALSPACPelotas
p hetero = 0.09
CHILD OUTCOME
-0.35 (-0.55, -0.14)-0.13 (-0.83, 0.57)
-0.16 (-0.31, -0.01) 0.05 (-0.50, 0.60)
-0.16 (-0.22, -0.09) 0.14 (-0.07, 0.36)
0.97 (0.62, 1.32)1.97 (0.88, 3.05)
EFFECT SIZE (95% CI)
0-1 -.5 0 .5 1 1.5 2 2.5 3 3.5
ALSPAC-PELOTAS BREASTFEEDING ASSOCIATIONS
SBP (mmHg)ALSPACCOHORT
p hetero <0.001
DBP (mmHg)ALSPACCOHORT
p hetero = 0.005
BMI (kg/m2)ALSPACCOHORT
p hetero <0.001
CHILD OUTCOME
-0.35 (-0.55, -0.14) 0.12 (-0.01, 0.24)
-0.16 (-0.31, -0.01) 0.10 (-0.01, 0.20)
-0.16 (-0.22, -0.09) 0.04 (0.00, 0.08)
EFFECT SIZE (95% CI)
0-1 -.5 0 .5 1 1.5 2 2.5 3 3.5
ALSPAC-LMIC COHORT CONSORTIUM BREASTFEEDING ASSOCIATIONS
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Comparison Method Validation
ALSPAC Strong socioeconomic
patterning in breastfeeding
PelotasWeak socioeconomic patterning
in breastfeeding
BelarusRandomised
Trial
Association with any breastfeeding (per category)* Effect of breastfeeding intervention
Outcome β 95% CI P β 95% CI P Difference in outcome
95% CI
SBP, mmHg -0.35 -0.55, -0.14 0.001 -0.13 -0.83, 0.57 0.7 0.2 -2.9, 3.3
DBP, mmHg -0.16 -0.31, -0.01 0.04 0.05 -0.50, 0.60 0.9 0.2 -1.8, 2.2
BMI, kg/m2 -0.16 -0.22, -0.09 <0.001 0.14 -0.07, 0.36 0.2 0.1 -0.2, 0.3
IQ 0.97 0.62, 1.32 <0.001 1.97 0.88, 3.05 <0.001 5.9 -1.0, 12.8
* Adjusted for all indicators of maternal education, paternal education, family income, occupational social class
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Using the totality of the evidence
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Integrating cross-cohorts and additional
causal approachesMaternal prenatal smoking and child aggression:
Exploring intrauterine effects in UK, Australian and
Brazilian cohorts
Marie-Jo Brion, Monique Robinson, Alicia Matijasevich, Colin Steer, Luciana Anselmi, Ana Menezes, Craig Pennell, Lyle Palmer, Cesar Victora, George Davey Smith, Debbie Lawlor
Brion et al. (under review)
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Multi-method approach for assessing intrauterine
mechanisms
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Multivariable Regression
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Cross-cohort confounders
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Maternal-Paternal Comparisons
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Prenatal-Postnatal Comparisons
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Single Method Approach
Conventional multivariable regression independent associations of maternal smoking
and child aggression intrauterine effects as a possible explanation
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Totality of Evidence From Multiple Methods
Total evidence for intrauterine effects is weak Not consistent with SEP being primary / sole determinant Confounding by other familial factors
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
ALSPAC and Cross-Cohort Studies
ALSPAC integrated with additional cohorts to explore cross-cohort approach for improving causal inference based on conventional method alone
Value of integrating multiple methods for assessing causal mechanisms to arrive at conclusions based on totality of evidence
Breastfeeding and child BMI, BP and IQ Evidence supporting causal effects breastfeeding duration on greater IQ
Maternal prenatal smoking and child aggression Weak evidence for intrauterine mechanisms Not consistent with SEP being the main explanation for the association Other family-level confounders are likely
School ofSOCIAL AND COMMUNITY
MEDICINE
University ofBRISTOL
Acknowledgements CAiTE, University of Bristol
Debbie Lawlor George Davey Smith Colin Steer
Federal University of Pelotas Cesar Victora Alicia Matijasevich Bernardo Horta
Other Co-authors: Monique Robinson, Craig Pennell, Lyle Palmer Luciana Anselmi, Ana Menezes
Wellcome Trust UK Henry Wellcome Postdoctoral Fellowship