Sunday, June 28, 2015

The National Diet Heart Study

The National Diet Heart Study (NDHS) was mentioned occasionally by some of the old trials and was included in the meta-analysis by Hooperet et al (2012).  Tracking down information on the NDHS was not easy.  There are a few easily accessible papers that mention it, but the actual study was published as a monograph in a supplement of Circulation in 1968, which was not online or for sale.  Fortunately I had access to a university library that had it.  The monograph is 428 pages long (which includes a 154 page appendix), so as you can imagine there’s a lot of detail, most of which I won’t cover (but is probably quite relevant for someone who is conducting diet trials)

Studies Associated with the Trial

The national diet-heart study final report (1968) (the monograph, which is also mentioned here)

Purpose

The NDHS was a feasibility study in preparation for the possibility of a huge trial involving 100,000 middle-aged men followed for 4-5 years, which they calculated “would be required to detect a 20 per cent reduction in incidence with statistical reliability.”  Among their objectives, looking at CHD events and mortality wasn’t really one of them and as you’ll see the study was very underpowered to do so

Methods

The participants for this study were apparently healthy middle-aged (45-55) men without CHD, abnormal ECGs or very high cholesterol levels.  Participants in each study were stratified based on cholesterol, BP, BMI and whether they smoked, then randomised to one of the groups.  The groups were similar at baseline and the participants appeared to be more health conscious than the general population based on lower rates of smoking

The NDHS includes three studies (first, extended and second) and many different experimental diets to examine the effect on acceptability, cholesterol levels, using a double blind design, and structured vs. unstructured dietary advice.  The NDHS took place in several US cities with an open centre each city, and in some ‘cottages’ in Faribault that housed mental patients.  Information about the different groups is presented in the table below:

Diet
Study
Exp/Con
Blind
SFA
PUFA
Chol
Fat
P:S
B
1st
Exp
Double
<9%
≥15
350-450
30%
1.5
C
1st
Exp
Double
<9%
≥18-20
350-450
40%
2.0
D
1st,Ext,2nd
Con
Both
16-18
7
650-750
40%
0.4
E
1st
Exp
Double
“low”
“very high”
“low”
40%
4.4
X
1st, Ext
Exp
Single


350
25-30%
1.0
BC
Ext, 2nd
Exp
Both
Reduced
Increased
350-450
30-40%
1.5-2.0
F
Ext, 2nd
Exp
Double
Low
High
350-450
40%
3.0
G
Ext, 2nd
Exp
Double
Mod Red
Mod Inc
350-450
40%
1.0
Y
2nd
Exp
Single
Like X
Like X
350
25-30%
1.0
Z
2nd
Con
Single
No advice
No advice
No advice
No advice
No advice

Some further points:

·         The diet for group D was designed to be similar to the usual American diet, which was described as having an average fat intake of 43%, cholesterol intake of 750 mg and a P:S ratio of 0.3
·         The double blind design was accomplished by most of the fat coming from food products (filled meats, dairy, etc) sold at diet heart open centres (DH-foods).  The added fat in the filled foods came from vegetable oils (experimental diets) and “either animal fat or hydrogenated shortening” (diet D)
·         Cholesterol intake was lower than planned in diet D because they couldn’t include more eggs without compromising the double blind design and extra cholesterol from other sources “was explored and abandoned as impractical”.  In addition diet D also didn’t meet its target for SFA, and as a result cholesterol levels in this group decreased, but not to the same degree as the other groups
·         Group Z was described as an ‘open control group’, the participants knew they were in a control group

The first study was relatively simple, just diets B, C, D, E and X.  Diets B, C and D were double blind and used in the open centres.  Diet X was used in the twin cities (Minneapolis–Saint Paul) open centres and “resembled diet B in composition, but utilized diet instruction alone, and purchase of foods exclusively in the open market, without special D-H foods.”  Diet E was also double blind and tested alongside diets B, C and D in Faribault

The extended and second studies were designed to compare the effects (feasibility, compliance, cholesterol response, etc) of the double blind design with the DH-foods against diet plans without the DH-foods.  The structure of the extended and second studies is more complex and is probably best explained by the diagrams below.  The second study in Faribault examined whether differences in fatty acid composition affected the response to increases in dietary cholesterol and found that higher SFA diets increased the response.


Results

The changes in cholesterol were covered in a 43 page long chapter, but basically what happened is what you would expect (as replacing SFA with PUFA reduces cholesterol), although the degree was less than expected from the equations used at the time*, which is probably due to a combination of issues of compliance and inaccurate reporting.  Some things that may be of interest include:

·         The calculated average intra-individual standard deviation in cholesterol levels was approximately 12-13 mg/dl (0.33 mmol/l) (which is a bit lower than the 17 mg/dl Chris Masterjohn mentions here)
·         In the open centres (not Faribault) cholesterol levels were lowest 2 weeks into the study, which tells us that changes in cholesterol from dietary interventions occur very quickly (much like other metabolic parameters) and that adherence declined over time
·         The researchers didn’t find a seasonal effect on cholesterol levels (the idea being summer >> increased extracellular volume >> decreased concentration of cholesterol (not absolute amount).  This mechanism may be related to how low salt diets slightly increase cholesterol)
·         The participants who either weighed more or had higher cholesterol levels tended to have larger decreases in cholesterol on all diets
·         Changes in weight, triglycerides, blood pressure and smoking were pretty similar between the relevant diet groups

More importantly, the results for CHD and non-CHD events are presented below.  We aren’t told what the CHD events were (was it only heart attacks or was it angina as well?).  The ‘other events’ were usually ulcers, gout, and cancers

First Study
Group B
Group C
Group D
Group X
Participants
385
390
382
54
CHD Events

4
1

Other Events
9
8
5
3*
Total Events
9
12
6
3
* 1 of these events was the development of peripheral vascular disease, which would be classified as a CVD event but not a CHD event

Extended Study
Group BC
Group F
Group G
Group X
Participants
357
128
114
28
CHD Events


1

Other Events
1
1

1
Total Events
1
1
1
1

Second Study
Group BC
Group D
Group F
Group G
Group Y
Group Z
Participants
194
304
127
120
38
38
CHD Events

5
1



Other Events

4

2


Total Events

9
1
2



To make things simple, let’s categorise the groups as experimental or control, then combine them, where all the groups except D and Z are experimental.  In summary the combined experimental group had lower incidence of CHD events which would not reach significance.  Putting the values into RevMan gives me a p value of 0.18 for CHD events and 0.56 for total events (these P values may not be accurate)


Experimental Groups (N = 1935)
Control Groups
(N = 724)
Risk Ratio
CHD Events
6
5
0.449
Non-CHD Events
25
9
1.039
Total Events
31
14
0.828

The combination of the small number of events and the fact that the first and second studies only lasted 1 year (the extended was a continuation of the first study, but only 4 events occurred in this study) means that any difference between the events in both groups could simply reflect differences in health status at randomisation, rather than any effect of the diet.  Something similar occurred in STARS where 3 of the participants in the control group had severe CHD and had events in the first several months (can’t remember where I read it though).  Also, as mentioned above, the filled foods for group D contained animal fats and hydrogenated shortening so it’s really a trial of SFA and TFA vs. PUFA (like so many of the others)

There were a number of things discussed in the paper that didn’t make the researchers enthusiastic about the large field trial.  These included the difficulties in recruiting participants, the double blind design, expense of the diet heart filled foods, drop outs, issues of adherence, a lower than expected incidence of CHD events, that cholesterol quickly returned to baseline levels in the post-intervention period (people weren’t interested), etc.  From here Ivan Frantz began the Minnesota Coronary Survey in 1968, which was the last diet heart trial conducted in the US**

Hooper et al included the NDHS as a modified fat trial and modified and reduced fat trial, which is completely appropriate.  However, I can’t figure out where they got most of their results from (to be honest I haven’t spent too long trying, but this stuff should be clear).  Fortunately the NDHS scarcely affected their overall result

CVD Events
Exp (n/N)
Con (n/N)
Weight
RR (CI)
Modified Fat
NDHS Open 1st Mod
4/348
1/170
0.3%
1.95 (0.22-17.35)
NDHS Open 2nd Mod
1/112
2/140
0.2%
0.63 (0.06-6.80)
Modified and Reduced Fat
NDHS Open 1st L&M
1/378
1/171
0.2%
0.45 (0.03-7.19)
NDHS Open 2nd L&M
0/179
2/140
0.1%
0.16 (0.01-3.24)

Total Mortality
Exp (n/N)
Con (n/N)
Weight
RR (CI)
Modified Fat
NDHS Faribault
4/143
0/52
0.0%
3.31 (0.18-60.49)
* No deaths due to CHD

* Equation 1 (S = SFA (%TE), P = PUFA (%TE), C = dietary cholesterol (mg)):
ΔChol = 2.16 ΔS – 1.65 ΔP + 0.0677 ΔC – 0.5
Equation 2 (Z = (dietary cholesterol (mg) per 1000 kcal)^0.5):
ΔChol = 1.35 (2ΔS – ΔP) + 1.5ΔZ 

** A lot of the problem of the 100,000 number is that it was apparently healthy middle-aged men without CHD, rather than older people in secondary prevention.  But in any event, trials of that kind of size aren’t impossible as evidenced by the Women’s Health Initiative, which also used its participants efficiently by testing many interventions for many diseases