Status on statistical methods in dietary assessment and Multiple Source Method

1

• Status on statistical methods in dietary
  assessment and Multiple Source Method
• Heiner Boeing
• German Institute of Human Nutrition
  Potsdam-Rehbrücke
• Department of Epidemiology

2
Dietary assessment

• Proper assessment of dietary intakes is
  critical in epidemiologic research that examines
  the relationships between diet and disease risk

NCI Web site
3
Long term average dietary assessment

• The concept of long-term average daily intake,
  or "usual intake," is important because
  diet-health hypotheses are based on dietary
  intakes over the long term. However, until
  recently, sophisticated efforts to capture this
  concept have been limited at best.

NCI Web site
4
Dietary assessment of usual intake for
large-scale prospective studies
5
Concepts of proper dietary assessment in cohort
studies

• Correction of the effect measures (Validation
  studies)
• Use of a reference instrument in a subgroup to
  apply information to the full cohort
  (calibration/standardisation)
• Best estimate of individual intake in the full
  cohort (reduction of assessment bias)

6
Calibration
Referenz
Instrument
Calibrated Instrument
7
Consequences of calibration
8
How to calibrate

• Use of a reference instrument (R)
• Selection of a subgroup with simultaneous use of
  Q and R
• Determination of a mathematical function for the
  Q value that fit the R value (Calibration
  function)

9
Calibration functions
Calibration function 1) Linear regression
calibration (1 day) 2) Linear regression
calibration (2 days) 3) Use of standardisation
functions (mean, variance, skewness, curtosis)
R ? Q ?
Qcal sqrt(qVAR(R)/VAR(Q))(Q-AM(Q))AM(R)
q 1/sqrt(1VARINTRA/VARINTER)
f(Q)
Calibration does not change ranking by Q
10
Multiple Source Method
11
Structure of the algorithm
Observed positive daily intake data
Observed consumption days
Parallel steps
First step
Second step
Probability of consumption
Usual intake on consumption days
Third step
Usual intake probability of a consumption day
usual intake on consumption day
12
First step
Observed consumption days

• Partition of observations
• Transformation
• Shrinkage of transformed data
• Back transformation
• Composition of components

Probability of consumption
13
Second step
Observed positive daily intake data

• Partition of observations
• Transformation
• Shrinkage of transformed data
• Back transformation
• Composition of components

Usual intake on consumption days
14
Third step
Usual intake on consumption days
Probability of consumption
Usual intake usual intake on consumption
day probability of a consumption day
15
Example fresh fruits
16
Example fish
17
Example breakfast cereals
18
Summary

• There are several strategies available to improve
  dietary assessment in cohort studies
• There is currently a lack of empirical data that
  have compared the strategies
• The Multiple Source Method is a new promising
  tool for dietary assessment in epidemiological
  studies