Research Commentary
Simon Thornley, Grant Schofield, Caryn Zinn, George Henderson
https://doi.org/10.1111/imj.14325
Abstract
Introduction
Evidence from meta‐analyses has been influential in deciding whether or not limiting saturated fat intake reduces the incidence of cardiovascular disease. Recently, random effects analyses have been criticised for exaggerating the influence of publication bias, and an alternative proposed which obviates this issue: “inverse‐variance heterogeneity”. We re‐analysed the influential Hooper meta‐analysis which supports limiting saturated fat intake to decide whether or not the results of the study were sensitive to the method used.
Methods
Inverse‐variance heterogeneity analysis of this summary study was carried out and the results contrasted with standard methods. Publication bias was also considered.
Results
Inverse variance heterogeneity analysis of the Hooper combined‐CVD endpoint results returned a pooled relative risk of 0.93 (95% confidence interval: 0.74 to 1.16). This finding contrasts with the traditional random-effects analysis with the corresponding statistic of 0.93 (95% confidence interval: 0.88 to 0.98). Egger tests, funnel and Doi plots along with recently published suppressed trial results suggest that publication bias is present.
Conclusions
This study questions the use of the Hooper study as evidence to support limiting saturated fat intake. Our re‐analysis, together with concordant results from other meta‐analyses of trials indicate that routine advice to reduce saturated fat intake in people with (or at risk for) cardiovascular disease be reconsidered.1
Comment
There are a number of meta-analyses that do not support the idea that saturated fat is a cause of heart disease. Whereas the 2015 meta-analysis by Hooper and colleagues suggested that there was a positive effect on cardiovascular mortality from reducing saturated fat intake.2
Did the Hooper study actually show benefit from reducing saturated fat?
While there was a small, yet potentially important effect on cardiovascular events associated with reduced saturated fat diets (by approximately 17%; risk ratio (RR) 0.83; 95% confidence interval (CI) 0.72 to 0.96), there was no effect on cardiovascular mortality or on stroke incidence. The only effect seen was when saturated fat was replaced with polyunsaturated fats, and not when saturated fat was replaced with carbohydrate or monounsaturated fats.
Was the effect observed in the replacement analysis larger because of the statistical methods they used?
In the paper by Thornley et al., the authors explain that the ‘random effects models’ used by Hooper and colleagues to pool their data, has the effect of biasing the results in favour of smaller trials where those disagree with larger. This seems counter-intuitive to the idea that larger numbers are more likely to have greater relevance to populations (‘the law of large numbers’).
In their paper, Thornley and colleagues suggest that “inverse-variance heterogeneity’ analysis is more suitable because it widens confidence intervals, “yet retain[s] the favourable weights of the fixed effect method”. Using this method of statistical analysis, a pooled relative risk of 0.93, with a 95% confidence interval of 0.74 to 1.16 is produced. This means (because the 95% CI overlaps ‘1’) that there is no definitive effect of saturated fat on cardiovascular disease using this analysis.
What does this all mean?
The one analysis from the Hooper study which showed benefit from reduced saturated fat was for cardiovascular events, however, there are several considerations as to why this may not be a meaningful finding:
Greater weighting given to small studies
When smaller studies with a higher risk of publication bias are given a smaller weighting, the effects are no longer significant.
No effect when carbs substituted for saturated fat
When carbohydrate is consumed in greater amounts, with a lower intake of saturated fat, there is no reduction in CVD incidence or mortality. This suggests that it’s not the saturated fat that is the problem!
Positive benefits are only seen when polyunsaturated fats substituted for saturated fat
There was an effect shown when more polyunsaturated fats were consumed and less saturated fats. But because the same effect wasn’t seen with carbs, or with monounsaturated fats, it doesn’t make sense to label saturated fats as ‘bad’ but instead to look deeper into why polyunsaturated fats might be ‘good’ (i.e. greater intake of omega-3 fats, and/or greater intake of vegetables.)
No effect on cardiovascular or all-cause mortality
The most important outcome is death. And the most important ‘death’ outcome is overall rate, not necessarily grouped into different diseases. Put it this way, if we want to see what type of diet is ‘best’ overall, we need to see how many people get diseases, or die, from all causes, not from a particular illness. We don’t see a risk for all-cause or even CVD mortality from saturated fat in the diet.
No substantive analysis of food ‘quality’
While ‘quality’ is vague, it is becoming clearer by the day, that more important than macro split in the diet, is how much of the diet is made up by refined and ultra-processed foods. In most observational data we can see a clear trend towards there being greater impact of processed food and our modern food environment on mortality when compared to the amounts of macros, or sub-groups of macros that are eaten.
References
1. Thornley S, Schofield G, Zinn C, Henderson G. How reliable is the statistical evidence for limiting saturated fat intake? A fresh look at the influential Hooper meta-analysis. Internal Medicine Journal. 2019;0(ja).
2. Hooper L, Martin N, Abdelhamid A, Davey Smith G. Reduction in saturated fat intake for cardiovascular disease. The Cochrane database of systematic reviews. 2015(6):Cd011737.
