Good afternoon,
Sorry if this question seems to newbie for you.
We have a nutritional dataset with energy (kcal), carbohydrates (g),
proteins (g) and tota fats (g).
We know that
Energy = carbohydrates*4 + total fats*9 + proteins *4 + alcohol*7 (but our
dataset includes only children - so alcohol value are zero).
We want to know the effect of each macro-nutrient on obesity (defined as BMI
being >=95e percentile on USDA CDC Growth curves) when adjusting for
socio-economic factors.
I used macro-nutrients in quintiles.
I made a logistic regression with all these variables (socioeconomic factors
+ carbohydrates (g) (in quintiles) + total fats(g) (in quintiles) +
proteins(g) (in quintiles) ) and I got adjusted odds ratios.
Before adjusting , I got OR for each variables separately and everything
(socioeconomic factors + carbohydrates (g) (in quintiles) + total fats(g)
(in quintiles) + proteins(g) (in quintiles) ) was significantly associated
with obesity.
After adjusting, only carbohydrates was still significant. It is because the
3 macro-nutrients were all related to each others ? In fact, a children eats
X amount of carbohydrates, X amount of total fats and X amount of proteins.
All the amounts are >0 and never 0. and no children does not eat at all, or
in other words, everyone must have eaten something (life and death rule).
Should I use some corrections here ?
Just to let you know that there is no significant interractions between the
3 macro-nutrients in the multivariate analysis and their correlations (carbo
vs fats, carbo vs proteins, proteins vs fats) are less than 25%.
Thanks
Manon Girard
|