Transcript Fellows Presentation FFQs and Dietary Pattern Analysis

FFQs and Dietary
Pattern Analysis
The road to better understanding the contribution of diet
towards maternal and offspring health
S
Diet and Health
Incident of Diabetes, IDF 2013
Diet and Health
Incident of Diabetes, IDF 2013
Diet and Health
kCal per day, 2014
Diet and Health
Diet and Health
S Uncover food patterns associated with increased and reduced
incidence of disease, their biomarkers (e.g., body weight), and/or
their internal regulators (e.g., gene expression).
S Using:
1.
2.
Food Frequency Questionnaires (FFQs); and
Diet pattern analysis using Principal Component Analysis (PCA).
Dietary Analysis
S FFQs are questionnaires used to determine the food and
beverages, and their quantities, consumed by an individual;
S For the NutriGen study, FFQs from each of the four cohorts
(ABC, CHILD, FAMILY, and START) have been processed.
Dietary Analysis
S FFQs are questionnaires used to determine the food and
beverages, and their quantities, consumed by an individual;
S For the NutriGen study, FFQs from each of the four cohorts
(ABC, CHILD, FAMILY, and START) have been processed.
Dietary Analysis
SHARE (ABC, FAMILY, and START)
CHILD
Origin
McMaster (Kelemen LE, et al., 2003) and the
Food Processor nutrient analysis software
Fred Hutchinson Cancer Research Center
and Nutrition Data Systems for Research
Items
~160 (variation between ethnicities)
~150
Food
Grouping
NO
YES (e.g., doughnuts, pies, pastries)
EthnicSpecific
YES (White European, South Asian, Chinese,
and Aboriginal/First Nation)
NO
Consumption
Frequency
Self-defined
Ranged (e.g., 1-2x/week)
Serving Size
Equal between ‘SHARE’ studies
Some differences with ‘SHARE’
Dietary Analysis
SHARE (ABC, FAMILY, and START)
CHILD
Origin
McMaster (Kelemen LE, et al., 2003) and the
Food Processor nutrient analysis software
Fred Hutchinson Cancer Research Center
and Nutrition Data Systems for Research
Items
~160 (variation between ethnicities)
~150
Food
Grouping
NO
YES (e.g., doughnuts, pies, pastries)
EthnicSpecific
YES (White European, South Asian, Chinese,
and Aboriginal/First Nation)
NO
Consumption
Frequency
Self-defined
Ranged (e.g., 1-2x/week)
Serving Size
Equal between ‘SHARE’ studies
Some differences with ‘SHARE’
Requires standardization
Dietary Pattern Analysis
1.
Standardize CHILD food portions to that of the SHARE FFQ.
•
e.g., ½ cup versus 1 cup servings, change from 2/week to 1/week
Dietary Pattern Analysis
1.
Standardize CHILD food portions to that of the SHARE FFQ.
•
2.
e.g., ½ cup versus 1 cup servings, change from 2/week to 1/week
Create standard food groups to reduce number of variables and ease
interpretation of dietary patterns
•
e.g., canned meat lunch meat, breakfast sausages => processed meat
Dietary Pattern Analysis
1.
Standardize CHILD food portions to that of the SHARE FFQ.
•
2.
Create standard food groups to reduce number of variables and ease
interpretation of dietary patterns
•
3.
e.g., ½ cup versus 1 cup servings, change from 2/week to 1/week
e.g., canned meat lunch meat, breakfast sausages => processed meat
Built upon food groupings from previous studies* analyzing dietary
pattern analysis and cardiometabolic conditions, allergies, and
indicators (e.g., FPG, HOMA-IR, CRP, cholesterol and TG).
*Hu et al AJCN 1998, Fung et al AJCN 2001, Nettleton et al AJCN 2009, Gadgil et al JAND 2013.
Dietary Pattern Analysis
•
•
•
•
•
•
•
Meats
Meat Dishes
Organ Meats
Processed Meats
Poultry & Waterfowl
Eggs
Fish & Seafood
•
•
•
•
•
•
•
•
•
•
•
Leafy Greens
Cruciferous Vegetables
Starchy Vegetables
Vegetable Medley
Other Vegetables
Fresh Seasonings
Legumes
Tofu
Fruits
Non-Meat Dishes
Stir-Fried Noodles and
Rice
•
•
•
•
•
•
•
•
•
Full-Fat Dairy
Low-Fat Dairy
Fermented Dairy
Fats
Fried Foods
Refined Grains
Pasta
Pizza
French Fries
• Whole Grains
• Nuts and Seeds
•
•
•
•
•
Snacks
Sweets
Condiments
Sweet Drinks
Artificial Sweet
• Tea
• Coffee
• Coolers, Spirits,
and Mixed Drinks
Dietary Pattern Analysis
S Principal Component Analysis (PCA)
S Reduces complex data into fewer dimensions
S Are there underlying patterns that distinguish groups of individuals?
S
e.g., dietary pattern
S Performed in R, using ‘psych’ package
S To uncover that we need to consider three PCA parameters:
1.
Number of dimensions/factors (i.e., number of diet patterns)
2.
Rotation method (i.e., diet patterns)
3.
Loading scores (i.e., foods within each diet)
Dietary Analysis
1. Number of Dimensions
S Scree plot (“breakpoint” or
“breakpoint” -1)
S Arbitrary cutoff (e.g.,
eigenvalue of 1.0)
Dietary Analysis
2. Rotation Method
S Groups the data in a specified manner, that best tells the story
S Oblique - assume that the variables are correlated
S Orthogonal - assume that the variables in the analysis are
uncorrelated
S Multiple choices but ‘varimax’ is most common dietary analysis
S
Aims to load food strongly in one dimension only.
Dietary Analysis
3. Loading Scores
S How strongly a specific food item/group contributes to a
dimension/dietary pattern
S Typical cutoff range from 0.20-0.30.
S In this case, 0.30 was used as the cutoff as it provided a clear
contrast between dietary patterns (e.g., prudent and Western)
ABC
Western: Red meats, processed meats,
fried foods, refined grains, snacks,
pasta, pizza, french fries, sweets and
condiments.
Prudent: Red meats, seafood, non-red
meats, legumes, leafy greens, fruit and
vegetables.
CHILD
Prudent: Non-red meats, legumes,
leafy greens, fruit, vegetables, nonmeat dishes.
Western: Fats, processed meats, fried
foods, refined grains, , pasta, pizza,
french fries, snacks, sweets and
condiments.
FAMILY
Prudent: Fermented dairy, non-red
meats, legumes, leafy greens, fruit,
vegetables, whole grains, non-meat
dishes.
Western: Fats, red-meat, processed
meats, fried foods, refined grains,
pasta, pizza, french fries, snacks,
sweets and condiments.
START
Prudent: Low-fat dairy, fermented
dairy, legumes, fruit, vegetables, nonmeat dishes.
Western: Full fat dairy, red-meat,
processed meats, fried foods, refined
grains, snacks, sweets and
condiments.
NutriGen
Prudent: Fermented dairy, legumes,
fruit, vegetables, non-meat dishes.
Western: Full-fat dairy, red-meat,
processed meats, starchy vegetables,
refined grains, pasta, pizza, french
fries, snacks, sweets and condiments.
Pollo-pescetarian: Eggs, fish, poultry,
leafy greens, fruit, vegetables, stir-fried
dishes, nuts and seeds.
Next Steps
S Compare loading scores to maternal outcomes such as GWG,
GDM status, FPG, and AUC glucose.
S If associations uncovered, does the diet also contribute to the
health of the offspring.
K-means (2 clusters)
K-means (2 clusters)
PCA Scores vs K-means Classification
AUC = 0.988