Is Adjusting for Weight in TEE Calculation Reliable in Regression Analysis?

[wywong]

The article Energy expenditure in adults living in developing compared with industrialized countries: a meta-analysis of doubly labeled water studies has the following shocking conclusion:

Conclusion: TEE (total energy expenditure) adjusted for weight and age or PAL (physical activity level) did not differ significantly between developing and industrialized countries, which calls into question the role of energy expenditure in the cause of obesity at the population level.

The authors argued that the lack of physical activities in industrialized countries had little effect on the people's energy expenditure and thus current increases in energy intake were sufficient to explain the significant increase in obesity prevalence.

My question is: how can TEE be adjusted for weight, when the variate (weight) is probably a function of the output (TEE)?

Thanks in advance,

wywong

 

[FactChecker]

You are right that weight and TEE are correlated, but not completely. The research would need to carefully explain what effects they think are remaining after they have adjusted for weight. It's not clear to me if the phrase "or PAL" means that they also adjusted TEE for PAL or if PAL is just renaming TEE. The link to the article is broken.

 

[mfb]

I fixed the link.

It is a meta-study, throwing together various different groups selected for various different purposes. They find "a positive association of weight (P < 0.001) with TEE for both sexes". If people with a larger weight have a larger total energy expenditure as claimed, then you cannot get overweight simply by reducing your physical activity (at least not on average). You have to eat more.

 

[wywong]

I fixed the link.

It is a meta-study, throwing together various different groups selected for various different purposes. They find "a positive association of weight (P < 0.001) with TEE for both sexes". If people with a larger weight have a larger total energy expenditure as claimed, then you cannot get overweight simply by reducing your physical activity (at least not on average). You have to eat more.

Thanks for fixing the link.

It is true that a lot of overweight people overeat and overeating is a contributing factor, but I don't think it logical to exclude lack of exercise as a significant factor as claimed by the authors. Suppose 60% of overweight people overeat, but the other 40% don't and lack of exercise is solely responsible for the latter's problem. Further assume the latter's weight is steady, i.e., energy intake = energy expenditure (the energy saved in reduced physical activities is all spent on maintaining a heavier body). In this scenario, the energy expenditure of the 40% is the same as normal-weight people, but the energy expenditure of the 60% is higher due to an even heavier body. So the average TEE of the population is higher due to the former 60%, but the overweight/obese problem cannot be attributed solely to overeating.

Since both abundance of cheap fast food and lack of exercise correlates with human development index (HDI), I expect the meta-study would always get more or less the same TEE after adjustment for weight regardless of the percentage of the overeating population. In other words, even if everyone avoid excess energy intake by calorie counting, a lot of people still get overweight if they don't exercise.

 

[wywong]

You are right that weight and TEE are correlated, but not completely. The research would need to carefully explain what effects they think are remaining after they have adjusted for weight. It's not clear to me if the phrase "or PAL" means that they also adjusted TEE for PAL or if PAL is just renaming TEE. The link to the article is broken.

Sorry for the broken link.

PAL is defined as the ratio of TEE to the resting energy expenditure. Since resting energy expenditure is rather similar for different populations, it is natural that TEE and PAL follow the same pattern.

 

[mfb]

I guess we would need data about the distributions in the individual samples.

 

[wywong]

I guess we would need data about the distributions in the individual samples.

Due to the high cost of the doubly labelled water method, study sizes tend to be small. In the meta-study, there were 4972 individuals in 183 cohorts, an average of only about 27 per cohort, which was rather small. Another problem is the shape of the weight bell curves. In low HDI countries, the median and standard deviation of weight tended to be smaller. The overlap between the bell curves of low and high HDI countries represented mostly the normal weight population and it is not surprising that their TEEs were similar. In the overweight region where the bell curves differ so much, is there a reliable way to compare?

 

[Svein]

Drawing conclusions from regression analysis is a very dangerous activity, especially if you assume correlation⇒causality.

 

[FactChecker]

Drawing conclusions from regression analysis is a very dangerous activity, especially if you assume correlation⇒causality.

I agree. But ignoring the data is worse. Regression and other statistical techniques are tools that must be used wisely, but they must be used.