Intention to treat – what was the question again?

If you haven’t already heard there is a new low carb study out. It shows similar effects on weight compared to a low fat diet over 2 years, but better HDL results in the low carb arm. It’s a really interesting study, but I’ll leave it to others, like Jimmy to elaborate.

Once again this low carb study is from Gary Foster and colleagues. Foster has previously given us this one. Although containing a lot of interesting data the researchers once again insist on only presenting the data as intention to treat with missing data carried forward.

Statistics is definitely not my strong side, and if I’m way off on this one please show me why. And if there is a stat wiz out there who can tell me why this is the right way to present results, please do.

The researchers did do some sort of sensitivity analysis and tried to justify using the entire sample of 307 participants in the final analysis. But it still does not compare to actually giving us the data on the compliers vs. non compliers. A lot of people dropped out of the study. Of the 154 randomized to the low fat group, 49 was not assessed at 24 months. Of the 153 in the low carb group, 64 were not assessed at 24 months.

Now, intent to treat analysis is a perfectly fair method to use. But it means that the results cannot tell us which dietary approach is the more effective. If we want to know which diet causes the greatest weight loss we must look at the data from the participants that actually followed the diet, and only those. What the results of this recent study tells us, is the effect of being put on a diet as opposed to the effect of following one. Is it really so bloody impossible to include data on compliers vs non compliers? As interesting as it is to know the effect of being put on a diet I for one would also like to know the actual effect of following the diets.

Like many other studies using similar analyses, this one also show a regression to the mean, that is the results differ the most at 6 months and slowly merge into similarity as the low carbers eat more carbs, and the low fat group eat more total energy. Whether this is because of the type of analysis and the increasing number of participants discontinuing the treatment is beyond me.

The title of table 2 of the study says is all really, “Predicted Mean Changes…”. But what about the actual change and the actual effect of following the prescribed diet?

Richard Feinman has written this article, illustrating the trouble with using an intent to treat analysis.
Still, Foster has done a good job and provided interesting data. Looking forward to seeing how this study is going to be received.