RChung

They wrote in their paper that they tried BMI in place of waist size, and it had "negligible" change in the fit of the equation. This is pretty common for these types of "predictive" studies: they're just looking for the best prediction rather than to understand the "causal path." If you're familiar with "technical time series analysis" for, say, stock prices, this is very similar to that approach: you don't really care about fundamentals or other "structural, causal" relationships or why a stock price goes up or down, you only care about whether you can predict changes in it. There are other times when you're interested in understanding why something changes, especially if you want to know how to modify behavior, but evidently prediction isn't really one of those situations.