Zombie ideas

zombie-hands.jpgPaul Krugman, the Nobel Prize-winning economist, recently wrote a column about a zombie idea, which he defines as “an idea that should have been killed by evidence, but refuses to die.” I would argue that we have zombie ideas in the omega-3 space as well. The most prevalent is the recent deluge of studies from cardiologists and doctors that purport to show omega-3s are not beneficial for cardiovascular disease. Perhaps EPA and DHA are maligned because of their association with the broader dietary supplement industry, which is not trusted by doctors or consumers, but that should not blind researchers from examining the evidence objectively.

The best way to test whether an idea is a zombie is to look at the totality of the evidence. In biological sciences, this usually means looking at meta-analyses and systematic reviews of randomized, controlled trials. This is not a perfect method for nutrition studies because study designs vary dramatically, and these methods basically roll all studies up into one. Regardless, if your evidence base is large enough, you should still see something in theory.

The media would have you believe that these meta-analyses are finding no benefits at all. However, if you actually read the papers in-depth, they have all reported some very interesting (and statistically significant) beneficial effects. Meta-analyses are statistical studies, so they can easily try to analyze 20+ endpoints and subgroups in each study. There is literally no additional cost, unlike adding an additional arm to a human clinical study. The problem is that if you are cutting data 20+ ways, you are not likely to find 20 statistically significant effects, let alone 15 or even 10. So you have to ask yourself, what is the cutoff for concluding there are no beneficial effects? The media, researchers behind these studies, and the journals in which they are published seem to think that you need to find dozens of beneficial effects before concluding there is a benefit, but even if you find only one statistically significant beneficial effect, didn’t you find a benefit? Isn’t there a portion of the patient population that would find some benefit? Should we forsake these patients because the study found no effect in a different portion of the population?

The reason I bring this up is because if you look at some of the big meta-analyses that have garnered negative media attention recently, they have all actually found some statistically significant benefits worthy of talking about. To give you one example, Kotwal et al found that EPA and DHA consumption reduced vascular death in patients with pre-existing CVD by 14% (p=0.03). They also found meaningful reductions in CVD events in studies with patients under the age of 61 (p=0.0015), in studies with a majority of normotensive patients (p=0.0008), in studies with less than 20% diabetics in the study population (p=0.0014), and in studies with hypertriglyceridemic (p=0.0002) and hypercholesterolemic patients (p=0.01). 

The study by Kotwal et al is not alone. Meta-analyses by Chowdhury et al, Rizos et al, Larsson et al, and many others have concluded there are no beneficial links between EPA and DHA consumption and CVD outcomes, yet have all discovered statistically significant benefits in their pre-designed analyses. Surely these are interesting enough to dispel the zombie idea that EPA and DHA have no role in CVD health.  If they had no role whatsoever, would you find this many statistically significant effects in studies designed to look at the totality of the evidence?