Sunday, 22 June 2014
Epidemiology can be Interesting
Hat tip to Nigel Kinburn for pulling up two studies from Siri-Tarino et al.’s 2010 saturated fat meta-analysis that did show correlations with heart disease. These were also the studies with the widest range of intakes. So what can they tell us?
The first is by Jim Mann and colleagues from 1993, and straight up I am surprised that this has been included in any meta-analysis, because it uses a self-selected vegetarian cohort, with friends and family standing in for the rest of the population.
“The study differs from previous prospective studies of diet and IHD in that the volunteers were individuals whose self-selected diet resembled, in nutrient content, current dietary recommendations rather than the relatively high saturated fat diet typical of most affluent societies. The ﬁndings may not only help to explain which attributes of a vegetarian diet protect against IHD but also which foods and nutrients are important in the aetiology of IHD in populations who modify their diets along the lines of present guidelines.”
It’s odd that such high-bias methodology isn’t excluded from meta-analysis, and makes me wonder what else is included.
What does Mann et al. tell us?
“Results—IHD mortality was less than half that expected from the experience reported for all of England and Wales. An increase in mortality for IHD was observed with increasing intakes of total and saturated animal fat and dietary cholesterol—death rate ratios in the third tertile compared with the ﬁrst tertile: 3.29, 95% conﬁdence interval (CI) 1.50 to 7.21; 2.77, 95% CI 1.25 to 6.13; 3.53, 95% CI 1.57 to 7.96, respectively. No protective effects were observed for dietary ﬁbre, ﬁsh or alcohol. Within the study, death rate ratios were increased among those in the upper half of the normal BMI range (22.5 to under 25) and those who were overweight (BMI over 25) compared with those with BMI 20 to under 22.5.
(Relative risk figures have been converted from 100 to 1.0)
IHD was significantly associated with higher consumption of eggs, cholesterol, animal fat, and saturated fat.
But, here’s the surprising finding; none of those dietary factors was associated with any increase in total mortality, significant or non-significant. People who avoided dying of IHD by following the healthy eating guidelines were dying at the same rate – the same ages - as their less health-conscious friends and family. This wasn’t pinned down to any particular cause of death.
The fact that BMI under 20 was associated with as much increased risk of overall mortality as BMI over 25 (“total mortality was signiﬁcantly higher in those with an initial BMI under 20, and a similar though not statistically signiﬁcant trend is apparent for IHD mortality.”) wasn’t mentioned in the abstract, and is underplayed in the text (if it can be explained by undiagnosed pre-existing disease, so can the correlation with higher BMI). A bit like this dodgy AHF BMI calculator. Set this to BMI 7 (maximum height, minimum weight) and you still look healthy; muscular or curvy depending on gender. Results in real life may vary.
The main dietary finding pertaining to all-cause mortality in Mann et al.;
“all cause mortality for all subjects was signiﬁcantly lower in the middle and highest intakes of green vegetables (0.62, 95% CI 0.46–0.83; and 0.74, 95% CI 0.56–0.99) and among those consuming the highest intake of nuts (0.76, 95% CI 0.60–0.97) compared with the lowest intakes of these foods.”
The second paper is by Bonniface et al., and unfortunately doesn’t supply all-cause mortality data.
“Not consuming alcohol, smoking, not exercising and being socially disadvantaged were related to high saturated fat intake and CHD death. Cox survival analyses adjusting for these factors found that a level of saturated fat 100 g per week higher corresponded to a relative risk for CHD death for men of 1.00 (0.86-1.18) and 1.40 (1.09-1.79) for women. This difference between the effects of saturated fat in men and women was statistically significant (P=0.019).”
Mean intakes of SFA in Bonniface et al. - Men: 47.0 g/d Women: 34.4 g/d. A respectable ~20%E (similar to the consumption by Indians eating food prepared with ghee in Raheja et al. 1995).
“The cut-off points for the quintiles of saturated fat in grams per week were 220, 276, 337 and 427 for men and 159, 202, 252 and 319 for women. There was a clear trend to higher CHD death rates associated with higher total and saturated fats and Keys' fat difference in women.”
Keys' fat difference? This is the ratio between SFA and PUFA.
“The result for the Keys statistic indicates that a higher level of saturated fat can be compensated by a lower level of polyunsaturated fat, in the ratio 2:1.”
PUFA by itself showed no correlations, but the Keys difference did. In fact, the correlation between Keys' difference and CHD in table 3 is pretty striking.
Both populations were in Britain. Perhaps the take-home is, that in Britain, at least at a certain point in time, you could choose how you wanted to die to some extent by choosing your diet around heart guidelines. Or by watching your Keys' ratio if you were female (women today, with vegetable oil diets, would not have ideal Keys' ratios by these tables). But living longer than those around you by restricting saturated fat is not a prediction supported by this epidemiology, or by any meta-analysis, as was discussed by Simon Thornley, Grant Schofield and I in our letter to the NZMJ.
Diet epidemiology is interesting stuff. It’s incredibly hard to do well, and the things we can take away from it are sometimes unexpected. The papers that go into meta-analyses, even for something like SFA, are wildly heterogeneous in design and in quality. Jim Mann et al.’s 1993 paper told me just about everything I wanted to know that it was possible to tell from the data collected. Bonniface et al. were more obscure; critical data points for the Keys' difference were not included. What use are quintiles without means or cut-offs?
I was surprised, as I said, that the Mann et al. paper, good though it is, is being included in meta-analysis, because of the self-selection bias (so, self-selection in Paleo or LCHF diet studies shouldn't be a barrier to being taken seriously either). That it was included speaks to the impartiality of meta-studies like Siri-Tarino et al. 2010 that exculpate saturated fat. Meta-studies give the overall truth that is relevant for public health planning, but miss the finer details of what is happening in specific communities at specific times. For example:
In Mann et al., nuts are protective. This is a common finding, e.g. in Hu et al. 1998. In Bonniface et al. PUFA is not associated with harm, but the Keys' difference is. In Britain at the time of this study, among the mainly middle and upper classes, perhaps vegetable oils were not in common use. Perhaps nuts were a major source of linoleic acid, enough to attenuate its relationship with disease. And in Bonniface et al., with its more working class catchment (and this being Britain, class distinctions do matter), perhaps the ideal Keys' difference of 2:1 is what you get closer to eating nuts and fish with meat and dairy fat, and the adverse lower and higher ratios are what you get either not eating nuts and fish, or using vegetable oils and spreads instead of animal fats.
(the mean PUFA intake of 63.1g/week for women is ~4%E).
Because it may turn out that when diet-heart epidemiologists one day separate PUFA in nuts and fish from PUFA in oils they will get very different values, as these AMD researchers did.
Take home: For someone who has the disease of CHD, especially someone following a moderate fat, higher carbohydrate diet like most of the population (the dietary pattern at the heart of all epidemiology) it makes sense to follow these clues, as well as recognising the modern risk factors of sugar and refined flour; eat some nuts, fish, don’t eat red meat every day, eat only a few eggs per week, eat some full-fat dairy, and so on.
On the other hand, for the average person to eat a pleasurable diet that has been designed around avoiding CHD risk factors from animal foods risks inviting in a host of other diseases that they may be susceptible to in ways they were never susceptible to CHD. Advice to the general population should be limited to recommending those protective factors for CHD that a) supply micronutrients, and b) are also protective factors in a wider sense (nuts, fish, fruit and vegetables, and full fat dairy), instead of messing with Keys' difference based on theories about blood lipids, as opposed to consistent findings based on real food inputs and hard endpoints.