Owen Yang
Jargons and indicators are quite dangerous things. I have had complains about jargons, but I do not remember I commented on indicators before.
Think: What is BMI and What is obesity?
Obesity is quite a broad and multi-dimensional concept. We will never be able to define it perfectly. This is because we use it for different purposes.
If we only speak about one look obese, the definition is likely to have emphasis on the visual property.
BMI does not always indicate obesity
If we speak about health, there is a predominate focus on cardiovascular health and use of BMI as an indicator for obesity or body fat. BMI works well when body fat is the main reason of weight variation in our target population, such as an average middle age men in your neighbourhood. BMI does not work that well among athletes because fat may not be the main reason of weight variation.
Some would say additional indicators are needed to compare people with different level of ‘central obesity.’ Central obesity itself is another concept created to describe why BMI may not work that well in predicting cardiovascular risk if the risk is caused by body fat around abdomen, and if some people only have fat around abdomen but not elsewhere, so that BMI misrepresent this risk.
From this concept of central obesity we then create another indicator (waist circumference for example) and then we are so proud and say waist circumference should be added on top of BMI in order to predict the risk better.
Here we have an example of using BMI to define obesity to begin with, but we should have known that BMI is just a number from height and e weight. Then we create additional indicator to measure obesity and give it another name (central obesity).
If we look at an athlete, say a basketball player, who is 180cm tall and 98 kg (BMI 30) with a lean-looking body and waist circumference, the person should not be considered obese at all because the weight is contributed mainly by lean mass (eg bone, water, muscle). There is no justification to mention this person is ‘obese’ but not ‘centrally obese.’
Fat is not the only issue in BMI
It is not to say that all health problems related to BMI are due to central obesity. Some may be related to body fat in general. For example, postmenopausal endometrial cancer is related to obesity because adipose tissue produce excess oestrogen that may inappropriately contribute to the development of cancer. One can reasonably assume that al adipose tissue could produce oestrogen.
Some risk related to obesity may not be related to fat directly. Knee arthritis, for example, is likely to be largely due to weight itself. Therefore, if you are an athlete with a very lean body, your knee is perhaps protected by your muscles that support your knee, but the weight itself can contribute to wear and tear of your knees. If you over-use your knee, this can contribute to arthritis, but this is probably not related to BMI directly.
Therefore, some health risk associated with BMI may not be due to fat, but due to weight itself even if the person is lean.
Indicators should not be confused with concepts: other examples
So each indicator has its limitation. So what?
Sometimes indicators can be so powerful that it overtakes the original concept it tries to describe, and this is not great.
This applies to p value and false discovery rate. The fact that you call them false discovery rate does not mean it is the rate of false discovery.
In a prediction model, it is common to use C-statistic as an indicator of model performance. The fact that it is an indicator of model performance does not mean it is the model performance.
