Sunday, October 11, 2015

Some Further Thoughts on Observational Studies

Confounding Variables in Observational Studies

A while ago I had a look at the observational studies regarding saturated fat (SFA) and coronary heart disease (CHD).  I haven’t blogged about it because my experience was pretty disappointing as very few of the papers referenced by meta-analyses included much detail besides risk ratios (RR) for SFA and CHD, and some didn’t even do that.  The paper Jakobsen et al cited for the Adventist Health Study didn’t report the RRs for nutrient intake (SFA, etc), but reported the RR for nuts and other foods.  Not to mention that this is a pretty inappropriate cohort, as Seventh Day Adventists are supposed to vegetarian and so vegetarian food patterns would be associated with even greater adherence to lifestyle factors that promote health.  Notably this cohort was an outlier for SFA>PUFA and SFA>CHO RRs (see figure 1).  This is similar to Jim Mann’s health conscious cohort that included a large number of vegetarians and semi-vegetarians (see figure 2 from Siri-Tarino et al)

Anyway, only a few observational studies reported baseline characteristics which can be sources of confounding variables.  The following tables come from a new paper published on the Nurse’s Health Study (NHS) and the Health Professionals Follow-up Study (HPFS)*.  The first shows a consistent trend whereby people who eat more SFA tend to engage in less health conscious behaviours.  The trends for each factor alone aren’t very impressive, but the results suggest they add up quite a bit

The trend between age-adjusted SFA intake and CHD is highly significant in both studies and the RR in the highest quintile is 1.54 for NHS and 1.32 for HPFS.  However, in the multivariate analysis, which adjusts for many of the baseline characteristics (and therefore attempts to control for confounding variables), there is no significant association between SFA and CHD and the RR in the highest quintile is 1.01 for NHS and 0.87 for HPFS**

This adjustment doesn’t work all the time as evidenced by examples such as hormone replacement therapy, but it does bring us much closer to the real relationship between SFA and CHD (I don’t know what it is but it’s probably close to neutral)

* NHS and HPFS have typically being the studies driving the inverse association between PUFA and CHD in a couple of the meta-analyses of observational studies.  This paper doesn’t change the overall evidence base much as it’s hardly a new cohort of people

** The results of this paper suggest the RR for 5% SFA is 0.94 and RR for 5% MUFA is 1.00, so I don’t know how they came up with MUFA being better than SFA in a substitution analysis

What Do We Mean By Saturated Fat?

One of the problems with the focus on nutrients is that they don’t necessarily tell you much about food quality.  We can get some information on this from the 2010 USDA dietary guidelines, which included some data from NHANES on the sources of saturated fat in the average US diet (below)

When these sources of saturated fat are roughly grouped into ‘higher’ and ‘lower’ food quality (based roughly on added sugars, refined starches and added fats) it seems that probably about half the saturated fat in the US diet comes from lower quality foods (higher quality = 37.4%, lower quality = 37.9%, unknown – 24.5%) (see below).  This is really not that surprising since the average US diet contains a lot of calories from ‘lower quality foods’, but it brings up two points

1.      Any association between saturated fat and disease could easily be due lower food quality rather than saturated fat
2.      Since consumption of refined sugars, starches and fats, etc is high, dietary advice should focus on reducing those rather than the current focus on things with a really debatable effect on health (such as SFA and salt).  It’s insane that dietary guidelines allow up to 3 serves of refined grains, but you better not eat fatty meat and full fat dairy, because SFA

This applies to other nutrients as well.  This discussion has occurred with carbohydrates.  It’s time for it to happen more widely with protein, fats, etc as well.

Higher Quality Foods (%)
Lower Quality Foods (%)
Regular cheese


Grain-based desserts

Dairy desserts

Chicken and chicken mixed dishes

Sausage, franks, bacon, ribs


Tortillas, burritos, tacos

Beef and beef mixed dishes

Reduced-fat milk

Pasta and pasta dishes

Whole milk

Eggs and egg mixed dishes



Potato/corn/other chips

Nuts and seeds, and nut and seed mixed dishes

Fried white potatoes

All other food categories (24.5%)


  1. Excellent post, Steven. You made several very good points in here.

    Why are burgers low-quality foods?

    1. Thanks Zooko

      The bun and other typically nutrient poor things that might be added (tomato sauce, mayonnaise, etc)