After a closer look at the research it doesn't look like obesity is much of a signal for predicting Alzheimer's disease after all

Photo by Henrikke Due

Sorry it looks like I’ve made a mistake in blindly trusting a research study. I have two biases: first I am very skeptical about scientific research. You can read about my takedown of my own field of marketing research. Most people fall into one of two camps: “science deniers” or “science believers.” I’m one of the rare people that is something in between. I verify. I have the training to do that and I have been very, very successful at finding the problems in my own field.

Unfortunately my skepticism was outweighed by my other bias that healthy living is the solution to most problems of health, so when I saw a claim that obesity causes Alzheimer’s I bought into it without fully checking it. It turns out they were probably wrong and so was I (for believing them).

In that post, I wrote that obesity is a massive predictor of Alzheimer’s disease. According to Whitmer et al. (2007) obesity at mid-life caused a “3.10 fold increase in risk” of Alzheimer’s disease.

There are two issues with this—one minor and one major:

Unclear language around the amount of change

This is a minor issue but it’s not 3.1 times higher; it’s 3.1 times as much. The “3.10 fold” language is misleading and unclear. Using the language in their way, a 1-fold increase would be no change. Perhaps that’s proper usage of the phrase but to me I read it and it says 3.1 times more (4.1 times as much). More clear language would be great.

Not much difference in the raw values

It is kind of a problem that they had to use a whole bunch of control variables in their model to find the obesity issue. If you look at the raw values there’s actually not much difference in the different weight groups:

• Normal weight at mid-life: 4.43% chance of Alzheimer’s disease

• Overweight at mid-life: 5.10% of Alzheimer’s disease

• Obese at mid-life: 4.71% chance of Alzheimer’s disease

If you look at it that way, all the numbers are too close to draw any real conclusions. The differences look like rounding errors. It’s not linear either. Why would it be worse to be overweight than obese? It’s not worse. The problem is a lack of precision due to the sample size of 477 total people with Alzheimer’s disease in the dataset. That number is not large enough to detect minute differences between these three groups.

So how did they find that massive difference then?

They used what’s called “control variables.” In a statistical model, you can include all the variables you are interested in into a model and basically let the model decide which factors are the most important. In some ways this is good but in some ways it’s bad. Here are the control variables used (footnotes of Table 3):

1. Fully adjusted for age (as the time scale),

2. age in midlife,

3. education,

4. race,

5. sex

6. marital status,

7. smoking,

8. hyperlipidemia,

9. hypertension,

10. diabetes,

11. ischemic heart disease and

12. stroke

But here’s the problem. Many of these issues are caused by obesity, or are caused by the same attitudes, lifestyle and behaviors that cause obesity. So if you control for all of these issues then your statistical model is nonsense. You should really only control for the variables *not* caused by your variable of interest (obesity).

Conclusions

While obesity could play a role in Alzheimer’s disease it’s sure not clear from the data used in this research. This indicates that there is no reason to panic about an up-surge in Alzheimer’s disease in the future—at least not based on this research.

Source

Whitmer, R., Gunderson, E., Quesenberry, C., Zhou, J., & Yaffe, K. (2007). Body Mass Index in Midlife and Risk of Alzheimer Disease and Vascular Dementia. Current Alzheimer Research, 4(2), 103–109. doi:10.2174/156720507780362047