New Review of Controlled Trials Replacing Saturated fat with Industrial Seed Oils

Readers Stanley and JBG just informed me of a new review paper by Dr. Dariush Mozaffarian and colleagues. Dr. Mozaffarian is one of the Harvard epidemiologists responsible for the Nurse's Health study. The authors claim that overall, the controlled trials show that replacing saturated fat with polyunsaturated fat from industrial seed oils, but not carbohydrate or monounsaturated fat (as in olive oil), slightly reduces the risk of having a heart attack:

These findings provide evidence that consuming PUFA in place of SFA reduces CHD events in RCTs. This suggests that rather than trying to lower PUFA consumption, a shift toward greater population PUFA consumption in place of SFA would significantly reduce rates of CHD.

Looking at the studies they included in their analysis (and at those they excluded), it looks like they did a nice job cherry picking. For example:

• They included the Finnish Mental Hospital trial, which is a terrible trial for a number of reasons. It wasn't randomized, properly controlled, or blinded*. Thus, it doesn't fit the authors' stated inclusion criteria, but they included it in their analysis anyway**. Besides, the magnitude of the result has never been replicated by better trials-- not even close.
• They included two trials that changed more than just the proportion of SFA to PUFA. For example, the Oslo Diet-heart trial replaced animal fat with seed oils, but also increased fruit, nut, vegetable and fish intake, while reducing trans fat margarine intake. The STARS trial increased both omega-6 and omega-3, reduced processed food intake, and increased fruit and vegetable intake. These obviously aren't controlled trials isolating the issue of dietary fat substitution. If you subtract the four inappropriate trials from their analysis, which is half the studies they analyzed, the significant result disappears. Those four just happened to show the largest reduction in heart attack mortality...
• They excluded the Rose et al. corn oil trial and the Sydney Diet-heart trial. Both found a large increase in total mortality from replacing animal fat with seed oils, and the Rose trial found a large increase in heart attack deaths (the Sydney trial reported total mortality but not CHD deaths).

The authors claim, based on their analysis, that replacing 5% of calories as saturated fat with polyunsaturated fat would reduce the risk of having a heart attack by 10%. Take a minute to think about the implications of that statement. For the average American, that means cutting saturated fat nearly in half to 6% of energy, which is a challenge if you want to eat a normal diet. It also means nearly doubling PUFA intake, which will come mostly from seed oils if you follow the authors' advice.

So basically, even if the authors' conclusion were correct, you overhaul your whole diet and replace natural foods with industrial foods, and...? You reduce your 10-year risk of having a heart attack from 10 percent to 9 percent. Without affecting your overall risk of dying. The paper states that the interventions didn't affect overall mortality.


* Not even single blinded.  Autopsies were not conducted in a blinded manner. Physicians knew which hospital the cadavers came from, because autopsies were done on-site. There is some confusion about this point because the second paper states that physicians interpreted the autopsy reports in a blinded manner. But that doesn't make it blinded, since the autopsies weren't blinded. The patients were also not blinded, though this is hard to accomplish with a study like this.

** They refer to it as "cluster randomized", which I feel is a misuse of that term.  The investigators definitely didn't randomize the individual patients: whichever hospital a person was being treated in, that's the food he/she ate. There were only two hospitals, so "cluster randomization" in this case would just refer to deciding which hospital got the intervention first. I don't think this counts as cluster randomization.  An example of cluster randomization would be if you had 10 hospitals, and you randomized which hospital received which treatment first.  It's analogous to individual randomization but on a group scale.