Date post: | 14-Jul-2015 |
Category: |
Data & Analytics |
Upload: | unicef-office-of-research-innocenti |
View: | 215 times |
Download: | 0 times |
Not all individuals who are recruited into a study are followed for the duration of interest
Deaths from causes that are not of primary interest
Migration away from the study area
Loss of interest in study
Atole Fresco
Ingredients (g/180 ml)
Incaparina 13.5 -
Dry skim milk 21.6 -
Sugar 9.0 13.3
Flavoring agent - 2.1
Nutrients per 180 ml
Energy (kcal) 163 59
Protein (g) 11.5 -
2 smaller villages
Atole Fresco
19 21
110
751 785
153 25
0
200
400
600
800
0
5
10
15
20
25
30
Atole Fresco Atole Fresco
kca
l/d
ay
g /
da
y
PROTEIN ENERGY
INTAKE FROM DIET
INTAKE FROM
SUPPLEMENT
Intakes in children 15-36 mo
2 larger villages
Atole Fresco
Composition of supplement
Boys Girls
Mother
Height (cm) 149.5 149.6
Non-pregnant BMI (kg/m2) 21.7 21.5
Child
Birth weight (g) 3140 3050
0
10
20
30
40
50
1969 1970-71 1972-73 1974-75 1976-77
% s
evere
ly s
tunte
d
Survey year
Atole Fresco
Habicht et al., J Nutr 1995
19
21
23
25
27
29
31
10th 25th 75th 90th
Vo
cab
ula
ry
SES (Percentile ranking)
Atole Fresco
10th 25th 75th 90th
Grade (Percentile ranking)
Pollitt et al J Nutrition 1995;124:111S-1118S
Women living in villages who delivered 1996-1999 (n = 130)Achievement Numeracy Literacy SIA Comprehension SIA Vocabulary General knowledge
0
20
40
60
80
100
<6 >= 6
Su
mm
ary
ach
ievem
en
t te
st
sco
re
Highest grade attained
Atole Fresco
Li et al, Pediatrics 2003; 112: 1156-1162
1855 (78%)
Living in
Guatemala
2392
Individuals in
1969-77 study
272 (11%)
Died between
1969 and 2002
102 (4%)
Untraced
2018 (85%)
Known to be
alive in 2002
163 (7%)
Living
outside
Guatemala
1112
Original villages
154
Nearby villages
419
In or near
Guatemala City
170
Elsewhere in
Guatemala
Grajeda et al., FNB 2005
1196
267
108
1571 (65.7%)
Males Females
Age (y) 32.1 32.5
Schooling (grade) 5.1 4.1
Rural residence (%) 74.3 69.1
Height (m) 1.62 1.51
BMI (kg/m2) 24.7 26.8
Data are means or %Sample sizes vary slightly due to item-missing data
Maluccio et al, Econ J 2009; Stein et al., APAM 2008
Reading comprehension
scores improved by 0.28 SD
Ravens Progressive Matrices
scores improved by 0.24 SD
Hoddinott et al. Lancet 2008
Wages
34% to 46% higher
Annual hours worked
222 Lower
▪ (CI: -572 to 128)
Annual income
US$ 914 higher
▪ ( CI: -$190, $2018)
0.00
0.15
0.30
0.45
0.60
0.75
0-24 0-36 36-72
0.670.62
0.22
US
$ / h
r
Window of exposure (months)
P < 0.01 P < 0.01
P > 0.4
Participation rates are 95% for village residents, 63% for internal migrants, and 0% for external migrants and untracedMigrants, untraced individuals and non-respondents probably differ from respondents Can one assume MAR or MCAR? Does failure to account for attrition bias our
estimates?
Step 1: Is attrition associated with intervention group?Step 2: What else predicts attrition?
Born in Atolevillage
(n=1269)
Born in Fresco village
(n=1123)
Alive in Guatemala, %
77.2 77.9
Migrated outside Guatemala, %
7.0 6.5
Died, % 12.4 10.4
Not traced, % 3.4 5.2
Data represent distribution of cohort members (n= 2392), by intervention group and tracing status
Born in Atole village (n=980)
Born in Fresco village (n=875)
Location (%)
Participated(%)
Location (%)
Participated(%)
In study village, % 57.9 94.9 62.4 95.1
In nearby village, % 11.8 90.5 4.5 89.7
In or near Guatemala City, %
20.9 63.9 24.5 63.4
Elsewhere in Guatemala, %
9.4 65.2 8.7 61.8
Data represent distribution of those eligible for contact (n=1855) and the proportion of those who completed at least one instrument (total n=1571), by intervention group and region of Guatemala
The pattern of attrition, from tracing through location and consent, is not markedly differential by intervention group.
This is reassuring.
But is it convincing?
Interviewed Died Untraced External
migrant
Traced, not
interviewed
n 1571 272 102 163 284
Year of birth 1970.1 1971.5 1970.1 1969.5 1970.2
Male, % 48 59 43 55 65
Atole village, % 53 58 43 55 52
Age of mother at birth, y 27.6 28.3 25.8 26.4 27.2
Age of father at birth, y 32.9 35.0 31.9 31.6 32.5
Mother’s schooling, y 1.3 1.1 0.9 1.5 1.5
Father’s schooling, y 1.7 1.3 3.0 1.7 2.0
SES, SD -0.19 -0.27 -0.60 0.36 0.16
Grajeda et al., FNB 2005All variables p≤0.05 by ANOVA, except for ‘Atole village’ (p=0.15)
Died Untraced External
migrant
Traced, not
interviewed
Male (ref = female) 1.47* 0.65 1.25 1.85*
Year of birth (per y) 1.09* 1.02 0.98 1.02
Age of mother at birth (per y) 1.00 1.02 0.99 1.01
Age of father at birth (per y) 1.02* 0.99 0.99 1.00
Mother’s schooling (per y) 0.97 0.79 1.03 1.05
Father’s schooling (per y) 0.92* 1.28* 0.98 1.04
SES (per SD) 0.99 0.72 1.14* 1.05
Data are odds ratios from polytomous regression models. * P<0.05. Reference category is ‘completed at least one instrument’ (n=1571). Model includes terms for village of birth and dummy indicators for missing values, with missing values imputed to the sample mode or mean, as appropriate Grajeda et al., FNB 2005
Predictors of attrition differ across type of attrition
These variables are likely to feature in any modeling work
Control for these variables will resolve some MAR concerns
Substantial attrition is likely to plague all longitudinal studies Allow for attrition when planning and assembling
cohort Collect data on factors that might predict attritionEnsure multiple contact points (parents, neighbors, population registers) Plan for periodic updates of key details and contact
informationEngage children as they mature Acknowledge their increasing autonomy and need for
privacy
In Guatemala: Generations of prior and
current investigators in the INCAP Longitudinal Study
Three generations of residents of the study villages
The INCAP field and support staff
In USA: Colleagues and students at
Emory University, University of Pennsylvania, IFPRI, Middlebury…
Funding: National Institutes of Health,
USA
▪ PH-43-65-640
▪ TW-005598
▪ HD-046125
▪ HD-075784