On 2017 Jan 19, Jameson Voss commented:

This paper identifies an interesting association between calories in the national food supply per capita and average population body weight. It was recently recognized by an independent review as an important work in the field. The recognition is appropriate as the analysis was based on worldwide comprehensive data, stratification of countries by income, assessment of longitudinal changes, incorporation of a mathematical model of weight gain and other strengths. The association is plausible and builds on other work connecting the food supply and body weight.

Despite these strengths, readers should be cautious about how they interpret the title and discussion. First, the title uses a causal phrase “major driver” which was also echoed in the recent review (linked above). Readers should be aware this title refers to an ecologic association and avoid making causal inferences about outcomes at a population or individual level based on this type of observational study design. Secondly, readers might misinterpret the term “major” as though the authors found the association to be stronger than other obesity correlates, but the paper did not provide an empiric comparison with any other obesity correlate. Finally, the discussion about opportunities for future research could also be misunderstood. The authors argue against testing the association with a study that moves people to different food systems by claiming it would be impractical. That might discourage readers from pursuing probative designs like “packet randomized experiments” which can utilize quasi-random assignment of migration (e.g., international adoptions, foreign exchange students, military household moves, etc.). If done properly, these studies can eliminate confounding among individuals, but confounding at the location persists. When participants are assigned to locations with higher energy in the food supply, they would simultaneously face other local, regional, or national exposures. That is, all exposures found at a location are assigned together as a single “packet,” so national food supply could be confounded by other differences between places (e.g., infrastructure, hygiene, ambient light, pollution, temperature, etc.).

There are empiric methods of handling packet level confounding, but it can be eliminated with a design called cluster randomization. Within the United States, an industry affiliated organization has worked to lower national per capita food energy supply and this has coincided with worsening average waist circumference, but it cannot be known if these trends are caused by the lower energy available in the food supply (as discussed here). Instead, system interventions could be randomized to different systems or subsystems or implemented at randomized start times or randomized locations.

While ecologic studies have confounding at both the location and subject level, there is still possible utility for prediction if the confounding remains stable across periods. Interventions, on the other hand, work based on causation and can have unintended consequences. Thus, readers should continue to feel encouraged to investigate net harms and benefits of altering the food supply.

The views expressed in this comment are those of the author and do not necessarily reflect the official policy or position of the Air Force, the DoD, or the U.S. Government.

On 2017 Jan 19, Jameson Voss commented:

This paper identifies an interesting association between calories in the national food supply per capita and average population body weight. It was recently recognized by an independent review as an important work in the field. The recognition is appropriate as the analysis was based on worldwide comprehensive data, stratification of countries by income, assessment of longitudinal changes, incorporation of a mathematical model of weight gain and other strengths. The association is plausible and builds on other work connecting the food supply and body weight.

Despite these strengths, readers should be cautious about how they interpret the title and discussion. First, the title uses a causal phrase “major driver” which was also echoed in the recent review (linked above). Readers should be aware this title refers to an ecologic association and avoid making causal inferences about outcomes at a population or individual level based on this type of observational study design. Secondly, readers might misinterpret the term “major” as though the authors found the association to be stronger than other obesity correlates, but the paper did not provide an empiric comparison with any other obesity correlate. Finally, the discussion about opportunities for future research could also be misunderstood. The authors argue against testing the association with a study that moves people to different food systems by claiming it would be impractical. That might discourage readers from pursuing probative designs like “packet randomized experiments” which can utilize quasi-random assignment of migration (e.g., international adoptions, foreign exchange students, military household moves, etc.). If done properly, these studies can eliminate confounding among individuals, but confounding at the location persists. When participants are assigned to locations with higher energy in the food supply, they would simultaneously face other local, regional, or national exposures. That is, all exposures found at a location are assigned together as a single “packet,” so national food supply could be confounded by other differences between places (e.g., infrastructure, hygiene, ambient light, pollution, temperature, etc.).

There are empiric methods of handling packet level confounding, but it can be eliminated with a design called cluster randomization. Within the United States, an industry affiliated organization has worked to lower national per capita food energy supply and this has coincided with worsening average waist circumference, but it cannot be known if these trends are caused by the lower energy available in the food supply (as discussed here). Instead, system interventions could be randomized to different systems or subsystems or implemented at randomized start times or randomized locations.

While ecologic studies have confounding at both the location and subject level, there is still possible utility for prediction if the confounding remains stable across periods. Interventions, on the other hand, work based on causation and can have unintended consequences. Thus, readers should continue to feel encouraged to investigate net harms and benefits of altering the food supply.

The views expressed in this comment are those of the author and do not necessarily reflect the official policy or position of the Air Force, the DoD, or the U.S. Government.