[importantbrian]

This study has lots of problems. I'm not totally convinced the results are statistically significant. The problem with using non-linear models and not showing P values is it is open to interpretation what a statistically significant result is. To me, these results don't seem that compelling. For the cholesterol question, the Adjusted Risk for CVD at the 95% confidence level is between 1.39% and 5.09%, and for All-Cause Mortality, it is 2.51%-6.36%. That means according to this study cholesterol consumption is more highly correlated with all-cause mortality than it is with CVD. That seems hard to believe, but might be true.

The biggest issue I have is with using non-linear models. I don't have access to the full article so I don't know why they chose the statistical methods they did, but I'm suspicious that they don't report P values for what it seems should be linear relationships. I don't buy the non-linearity argument in the summary. Once you get into non-linear regression models the risk of overfitting gets worse. There is a lot of work that has to be done to show you've accounted for this, and it's hard to tell from the summary if that was done.

They also didn't control for some things I would have liked to see. Namely, calories and weight gain. So, it could be that all this study is really showing is a relationship between overeating, weight gain, and CVD and all-cause mortality. It may simply be that people who eat more calories and end up being overweight are also more likely to eat a lot of eggs and cholesterol. In this case, it's gaining weight that causes health issues, not the eggs. It's just the case that people who eat too much and are overweight also happen to eat a lot of eggs. What I'd like to see is a study where they control for calories consumed, weight gain, lean mass etc. and then see how a diet with more cholesterol performed against a diet with less cholesterol.