Thread by @PWGTennant

Excited to share exchange with Willett, Stampfer, & @deirdre_tobias published in @AJCNutrition. Hopefully interesting to all in nutrition Epi!

Paper: academic.oup.com/ajcn/article/1…

Willet et al: academic.oup.com/ajcn/advance-a…

@GeorgiaTomova et al: academic.oup.com/ajcn/advance-a…

1/12
#EpiTwitter

Our paper examined common approaches to adjusting for energy intake using a causal framework. Willet et al raised 4 points of disagreement with our conclusions.

I'll try to summarise with our responses. Beware, it jumps straight into technical details!

2/12 cc @GeorgiaTomova

Q1: Partition models

Willet et al: 'the energy partition model is not appropriate because it does not ultimately control for total energy intake" & this "is not consistent with the isocaloric diet/disease relation of greatest interest'

3/12 cc @GeorgiaTomova

A1: Partition models

@GeorgiaTomova et al: 'Nutritional data are compositional, with the total being mathematically determined by its components. A model containing all components of energy intake therefore... controls for the total, making it necessarily isocaloric'

4/12

Q2: Standard energy-adjusted models

Willet et al: '(The claimed issues with the 'standard model' are) not a limitation or bias of the standard energy-adjusted model as the authors conclude. Rather, this reflects a poorly specified exposure'

5/12 cc @GeorgiaTomova

A2: Standard energy-adjusted models

@GeorgiaTomova et al: this is a 'genuine statistical bias' because '1) fewer components... leads to greater residual confounding; & 2) more components summarized by the total energy variable (may yield) composite variable bias'

6/12

Q3 Nutrient densities.

Willet et al: 'energy density... represents an interaction, but such an interaction is often biologically likely... it is (therefore) fully appropriate to hypothesize and model diet composition expressed as a percentage of calories'

7/12 cc @GeorgiaTomova

A3 Nutrient densities

@GeorgiaTomova et al: 'We profoundly disagree' & quote Willet & Stamfer: 'Since a nutrient density variable contains the inverse of caloric intake... nutrient densities will tend to be associated with disease in the opposite direction to total'

8/12

Q4 Measurement error

Willet et al: 'correlated errors in the numerator & denominator tend to cancel out when calculating nutrient densities or energy-adjusted intakes... (in) partition model, measurement errors are accumulated...'

9/12 cc @GeorgiaTomova

A4 Measurement error

@GeorgiaTomova et al: 'adjustment for all components should... offer the maximum information about the... error structure and, therefore, return the most accurate coefficients... adjusting for total energy intake does not provide as much information'

10/12

We are thrilled at the interest that our paper has attracted & hope it will increase awareness of the issues with different approaches to modelling energy intake.

If this is your area, please consider a reading our paper & letter exchange!

11/12
#EpiTwitter cc @GeorgiaTomova

PS the notes to Willet et al include a curious/amusing claim that we 'chose' not to reply to Willet et al's letter! This is untrue, & hopefully is something @AJCNutrition can correct.

I'm glad both the letter & our response can now be read!

12/12
#EpiTwitter cc @GeorgiaTomova