Misreporting: Case Study

Misreporting
Case study: 2007 Australian Children’s Survey
Dr Anna Rangan
School of Molecular Bioscience
Misreporting
›  Misreporting refers to under- and over-reporting of intake
›  Under-reporting includes under-eating and under-recording
›  Common issue in all nutritional surveys
-  OPEN study found 12-16% under-reporting of EI (24HR)
›  Why worry?
›  Introduces bias into studies
›  Usually reduces associations between diet (foods/nutrients) and disease
›  Provides inaccurate estimates of nutrient inadequacies
›  Therefore important to assess the validity of EI in dietary surveys to
interpret data/results correctly
Validation of energy intake
›  DLW method is gold standard (not commonly used in NNS)
-  based on the principal of energy balance EE = EI, assuming stable BW and
composition
›  More common approach is to use Goldberg cut-offs
-  less expensive and practical but only excludes the most biased records
›  Alternative (but crude) method include set cut-points
-  e.g. allowable EI for women 500-3500 kcal, for men 800-4000 kcal
-  or rejecting top and bottom 2% of the distribution
Goldberg cut-off method
›  Often used to identify misreporters (esp under-reporters)
›  Can be used at individual or group level
›  Based on the principle of minimum EI required to be compatible with life
›  Compares the reported EI with TEE (both expressed as a multiple of BMR)
›  BMR is estimated from equations (e.g. Schofield)
›  EI/BMR = TEE/BMR (assuming BW stable)
›  EI/BMR = PAL
›  Then CL (or cut-offs) are applied around PAL
›  Values that fall outside these cut-offs are considered ‘misreporters’ of EI
Goldberg cut-off method
›  95% CI = PAL x exp ( +2 x S/100 )
√n
where S (28.7) takes into account the variability in EI (including number of days),
BMR, PAL (Black et al, 2000)
where n=number of days (i.e. 1 day of 24HR)
›  For example: for a PAL of 1.55,
the 95% CI are 0.87 and 2.75
If EI/BMR < 0.87, considered an under-reporter
If EI/BMR > 2.75, considered an over-reporter
2007 Australian Children’s Survey
Aims
1.  Identify misreporters using Goldberg cut-offs
2.  Describe characteristics of misreporters
3.  Explore impact of misreporters on the relationship between EI and BMI
4.  Examine food types most likely misreported
›  Rangan AM, Flood VM, Gill TP. Misreporting of Energy Intake in the 2007 Australian
Children’s Survey: Identification, Characteristics and Impact of Misreporters. Nutrients 2011,
3, 186-199.
›  Rangan A, Allman-Farinelli M, Donohoe E, Gill TP. Misreporting of energy intake in the 2007
Australian Children’s Survey: differences in the reporting of food types between plausible,
under- and over-reporters of energy intake. J Hum Nutr Diet 2013, doi:10.1111/jhn.12182.
2007 Australian Children’s Survey
Methods:
›  4826 children aged 2-16 y
›  Dietary data: 24 hour recalls on all children (used 1st day)
›  Physical activity: validated use-of-time tool for children 9-16y
›  Used Goldberg cut-offs to identify misreporters
1.  single PAL of 1.55
2.  estimated PAL (based on questionnaire or median PAL 1.65 if n/a)
% Misreporters
Under-reporters (%)
PAL 1.55
estPAL
Over-reporters (%)
PAL 1.55
estPAL
Boys
2-3 y
1.2
1.7
3.1
1.3
4-8 y
1.9
2.6
2.9
1.8
9-13 y
5.2
8.2
2.9
1.5
14-16 y
8.2
10.7
3.0
2.1
2-3 y
1.1
1.3
5.5
3.6
4-8 y
1.3
2.4
2.7
2.3
9-13 y
6.1
9.5
3.3
2.6
14-16 y
15.3
15.3
2.0
1.7
TOTAL
5.0
6.7
3.0
2.1
Girls
Characteristic
Under-reporters
Plausible reporters
Over-reporters
p
Total, % (n)
Energy intake (MJ),
mean (SE)
Age group, %
2–3 years
4–8 years
9–13 years
14–16 years
Gender, %
Boys
Girls
Parental education, %
School/certificate
Diploma/degree
Area of residence, %
Urban
Rural
Day of the week, %
Weekday
Weekend day
BMI, mean (SE)
Boys
Girls
PAL ^, mean (SE)
Boys
Girls
Unusual intake on
survey day, %
Feeling unwell
Feeling well
6.7%
5.03 (0.08)
91.3%
8.50 (0.04)
2.1%
16.11 (0.50)
<0.001
1.5%
2.5%
8.8%
13.0%
96.0%
95.4%
89.2%
85.1%
2.5%
2.1%
2.0%
1.9%
<0.001
6.1%
7.3%
92.2%
90.3%
1.7%
2.4%
0.043
7.2%
6.0%
90.2%
92.5%
2.5%
1.5%
0.010
7.3%
5.3%
90.7%
92.5%
2.0%
2.2%
0.034
6.6%
6.7%
91.8%
90.7%
1.6%
2.6%
0.133
22.6 (0.46)
22.7 (0.41)
18.3 (0.07)
18.6 (0.08)
18.2 (0.37)
17.3 (0.42)
<0.001
<0.001
1.81 (0.03)
1.67 (0.02)
1.71 (0.01)
1.60 (0.01)
1.63 (0.06)
1.56 (0.03)
<0.001
0.001
29.5%
6.0%
69.2%
92.0%
1.4%
2.1%
<0.001
From 2007 Children’s Survey (Rangan et al, Nutrients 2011.)
Impact of misreporting on diet-disease relationship
MLR model : variables associated with EI
All children
Excluding misreporters
β
P
β
P
Age
0.332
<0.001
0.381
<0.001
Gender
-1.442
<0.001
-1.439
<0.001
BMI z-score
-0.034
0.41
0.157
<0.001
R2=0.25
R2=0.38
Shows that misreporters have a significant impact on the association
between EI and BMI z-score
Differential reporting of food types
›  Foods or beverages may be omitted, added or substituted
›  Portion sizes may be under/over-estimated
›  May be due to
-  inability to recall foods eaten (forgotten)
-  inability to estimate portion size
-  due to lack of food knowledge or food preparation methods
-  intentional (‘socially desirable’)
›  Literature suggests high levels of selective reporting among adults
›  Selective reporting affects the quality and interpretation of data, and
should be investigated
Differential reporting of food types (% consuming)
2007 Kids Survey
Differential reporting of food types (g/consumer)
2007 Kids Survey
Differential reporting of food types (% consuming)
2007 Kids Survey
Differential reporting of food types (g/consumer)
2007 Kids Survey
Food types reporting summary
›  Under-reporters reported less frequent consumption and smaller quantities
of consumption of both core and non-core foods (compared with plausible
reporters)
›  Over-reporters reported similar percentages of consumption of many core
and non-core foods with some exceptions but generally consumed much
larger quantities
›  Compared to plausible reporters, under-reporters had
-  higher intakes of protein (%E) and starch (%E)
-  but lower intakes of sugar (%E) and fat (%E)
›  And over-reporters had
-  higher fat (%E) and lower carbohydrate (%E) intakes
Limitations
›  Only analysed 1 day data, therefore not ‘usual’ intake
›  Missing PAL data for younger children
›  Limitations of Goldberg equations
-  assumes individuals are in energy balance, weight stable
-  children have additional requirements for growth (small after 2y)
-  BMR equations do not perform well for higher BW
-  only detect extreme misreporters (not minor deviations)
-  cannot distinguish between under-eating and under-recording
Dealing with misreporters
›  No easy answer but should not be ignored
-  Can’t use statistics to correct for biased reporting
-  Determine the extent, who, why and what foods
›  Include or exclude?
1.  Analyse data both with and without misreporters
-  and use the difference as part of uncertainty evaluation
2.  Energy adjustment
-  e.g. apply energy residual method (Willett et al 1997)
-  or NRC-B method including biomarker data
-  but may not be appropriate if biased food reporting
Thank you