Thursday, January 29, 2015

Observational Studies Regarding Saturated Fat and Coronary Heart Disease

There are a few main meta-analyses that pooled the results of observational studies.  I won’t go into much detail or list all the observational studies, though I may look into them later


The Jakobsen, et al meta-analysis included 11 cohort studies (4-10 years of follow up) with a total of 344,696 people, 5249 CHD events and 2155 CHD deaths.  The hazard ratios (HR) come from substitution models to see the whether replacing 5% of calories from SFA with MUFA, PUFA or carbohydrate (CHO) is associated with CHD (taken from data in table 2).  See figure 1 for the plots on the individual studies.

CHD Events
All
Women
Men
SFA>MUFA
1.19 (1.00-1.42)
1.15 (0.84-1.58)
1.23 (0.98-1.55)
SFA>PUFA
0.87 (0.77-0.97)
0.85 (0.68-1.06)
0.87 (0.76-1.01)
SFA>CHO
1.07 (1.01-1.14)
1.00 (0.89-1.12)
1.11 (1.02-1.20)

CHD Mortality
All
Women
Men
SFA>MUFA
1.01 (0.73-1.41)
0.88 (0.51-1.54)
1.10 (0.71-1.69)
SFA>PUFA
0.74 (0.61-0.89)
0.61 (0.37-1.01)
0.80 (0.64-0.99)
SFA>CHO
0.96 (0.82-1.13)
0.86 (0.65-1.13)
1.03 (0.86-1.24)

Their conclusion was: “The associations suggest that replacing SFAs with PUFAs rather than MUFAs or carbohydrates prevents CHD over a wide range of intakes.”  However, their results suggest that replacing MUFA with PUFA may be more beneficial*

* An odd explanation they provide for SFA>MUFA being generally unfavourable is that: “the main source of MUFAs was animal fat, whereby confounding from other dietary components in meat and dairy products cannot be excluded.”  Because (1) animal fat is apparently low in SFA; (2) we just know meat and dairy is bad; and (3) there’s definitely no confounding going on in the SFA>PUFA results


The Mente, et al meta-analysis used the used the Bradford Hill criteria to “derive a causation score based on 4 criteria (strength, consistency, temporality, and coherence)”, see table 1A score of 4 was considered strong evidence of a cause-and-effect relationship between the dietary exposure and disease. A score of 3 was deemed to indicate moderate evidence of causation. A score of 2 or less was considered a reflection of weak evidence of causation”.  They pooled the results of many observational studies looking at a large variety of dietary factors and these results can be seen in table 2.  The RRs came from comparing the highest quintile/quartile with the lowest.


Total N
No. of Studies
CHD
Events
(1)
(2)
No. of Criteria Met (of 4)
TFA
145132
4
1.32
(1.16-1.48)
3/4
(75%)
3/6
(50%)
4
MUFA
101521
4
0.80
(0.67-0.93)
2/4
(50%)
3/5
(60%)
3
LCO3
301780
13
0.86
(0.75-0.97)
4/14
(29%)
7/22
(32%)
2*
Total Fat
126439
7
0.99
(0.88-1.09)
1/9
(11%)
1/9
(11%)
2
SFA
160673
11
1.06
(0.96-1.15)
4/12
(33%)
1/6
(17%)
2
PUFA
102937
6
1.02
(0.81-1.23)
1/6
(17%)
1/8
(12%)
2
(1) Consistency in Coronary Risk of Mortality
(2) Consistency in Coronary Risk, Mortality or MI
* “Bradford Hill score is 3 when restricting analyses to cohort studies of high methodologic quality (low risk of bias)”


The Skeaff & Miller meta-analysis pooled studies that compared the highest vs. lowest quintile/quartile and those with 1% and 5% increments in calorie intake from SFA.  The summary for all fats is presented below in table 3.  The forest plots for the SFA analysis are presented in the paper table 6, 7 and 8 (paper is open access).  They found SFA was not significantly associated with CHD events or CHD mortality.



The Siri-Tarino meta-analysis included 11 cohort studies (5-23 years of follow up) with a total of 347,747 people and 11,006 CVD events.  The RRs came from comparing the highest quintile/quartile with the lowest.  The characteristics of the included studies are presented in table 2 and table 3.  They found SFA wasn’t significantly associated with CHD or stroke



The recent Chowdhury meta-analysis included 20 cohort studies with a total of 283,963 people and 10,518 CHD events.  The RRs for the cohort studies (figure 1) and circulating fatty acids (figure 2) came from comparing the top third with the bottom third.  They found SFA wasn’t significantly associated with CHD in their analysis of cohort studies and those measuring circulating fatty acids.  It’s interesting that: (1) stearic acid had the highest RR among circulating SFA even though unlike other SFAs, stearic acid doesn’t increase LDL-C (it actually slightly lowers it) [1]; and (2) margaric acid and pentadecanoic acid (from dairy fat) had some of the strongest inverse associations with CHD in the circulating fatty acid analysis 



The most recent meta-analysis by Farvid, et al included 13 cohort studies (5.3-30 years of follow up), with a total of 310,602 people, 12,479 CHD events and 5882 CHD deaths.  RRs for linoleic acid (LA) came from: (1) comparing the highest vs. lowest tertile/quartile/quintile; (2) 5% energy increments from LA; (3) substituting 5% energy from CHO with LA; and (4) substituting 5% energy from SFA with LA


CHD events
CHD mortality
LA (high vs. low)
0.85 (0.78-0.92)
0.79 (0.71-0.89)
LA (+5%)
0.90 (0.85-0.94)
0.87 (0.81-0.93)
CHO>LA (5%)
0.89 (0.80-0.98)
0.87 (0.81-0.94)
SFA>LA (5%)
0.90 (0.80-1.01)
0.87 (0.82-0.94)

Similar to Jakobsen, the Farvid meta-analysis isn’t really anti-SFA, but rather is pro-LA.  They found LA is inversely associated with CHD and association was the same whether CHO or SFA replaced LA

 

In summary, all the meta-analyses agree that SFA isn’t significantly associated with CHD.  The Jakobsen and Farvid meta-analyses found that replacing 5% of total calories from SFA with PUFA (which means almost halving current SFA intake and almost doubling current PUFA intake) would reduce both CHD events and CHD mortality, but the replacement doesn't seem to be specific for SFA and appears to be an independent effect of PUFA.  These meta-analyses, even Jakobsen and Farvid, should not be used to argue that SFA are toxic (etc) as you must then come to the conclusion that MUFA (particularly in Jakobsen) and carbohydrate are equally as toxic (which is pretty rediculous).

* When looking at the Forest plots you can see how it’s easy it is to cherry pick studies to support any preconceived notions, which is a major reason why meta-analyses are important

7 comments:

  1. Great summary of the important meta studies.
    Last year's paper on linoleic acid from Harvard spells it out - PUFA is equally protective, association wise, whether it replaces SFA or CHO. MUFA too when they calculate that. So this is a non specific effect of PUFA, which is both scarcer and essential.
    It doesn't affect total mortality, and the effect of increases can't really be predicted because the average intake (in NZ) s 5% and few people would eat 10%E.
    Keys noted that 3-7% was the normal human range of PUFA and this was too low to influence cholesterol levels, which were mainly reflecting SFA.
    Thus the lipid hypothesis doesn't explain anything.

    ReplyDelete
    Replies
    1. Thanks. I just added that paper.

      I agree, despite repeated claims that SFA is toxic (which to Harvard’s credit they don’t seem to do as much of), the only evidence against SFA is really just pro-PUFA.

      Total mortality is definitely an issue for advice to replace SFA with PUFA as none of the meta-analyses of clinical trials found any benefit*. It’s not even a dilution effect as the RR in Hooper and Mozaffarian is ~1.0. But I didn’t see total mortality reported in Farvid, or any of the other meta-analyses of observational studies

      * Skeaff & Miller almost did, mainly because their misinterpretation of the Finnish Mental Hospital Study (more on that later)

      Delete
  2. Steven, do you have the appendices for that Mente review?

    ReplyDelete
    Replies
    1. No, but they say the appendices are available on request from the authors

      Delete
    2. Yea, I requested them from the authors sometime ago but they never replied.

      Anyway, nice post as usual.

      Delete
    3. Thanks.

      Have you tried emailing them again? I've been trying to track down some old papers and I've had to email people a few times to get a response

      Delete
    4. You're right, I should try again. Thanks.

      Delete