PREDIMED – what really is a risk factor?

We did such a bumper review of PREDIMED last week that this week will be a post script on some very interesting findings that I happened to observe going through the main PREDIMED paper (Ref 1).

PREDIMED claimed that “a Mediterranean diet supplemented with extra-virgin olive oil or nuts reduced the incidence of major cardiovascular event.” The study only included people without cardiovascular disease (CVD), but those considered to be at high risk for CVD.

To recruit ‘high-risk’ people, the inclusion criteria were that the participant should either have type 2 diabetes or three, or more, CVD risk factors. The CVD risk factors (remember that I said that I didn’t agree that all of these are risk factors) were considered to be: current smoking (>1 cig/day during the last month); hypertension (systolic BP ≥140 mmHg or diastolic BP ≥90 mmHg or being on blood pressure medication); LDL cholesterol ≥160 mg/dl or being on lipid-lowering medication; HDL cholesterol ≤40 mg/dl in men or ≤50 mg/dl in women; body mass index ≥25 kg/m2; and family history of premature coronary heart disease (CHD). (The LDL and HDL assumptions collectively were called “dyslipidemia”).

Baseline characteristics

Table 2 in the main paper reported many details about the participants at the start of the study: age; gender; BMI; smoking history; and so on. It reported the so-called risk factors too. We are thus able to see the number of people in each group with Type 2 diabetes, hypertension (high blood pressure by their definition), family history of premature coronary heart disease (CHD) and so on. I’ve summarised the key ones below:

Characteristics (Table 2)

EVOO

Nuts

Ctrl

Total

As a % *

Participants

2,543

2,454

2,450

7,447

Type 2 Diabetes

1,282

1,143

1,189

3,614

49%

Hypertension

2,088

2,024

2,050

6,162

83%

Dyslipidemia

1,821

1,799

1,763

5,383

72%

Family history premature CHD

576

532

560

1,668

22%

Smoking ever

971

989

923

2,883

39%

Male

1,050

1,128

987

3,165

43%

BMI >30

1,195

1,087

1,201

3,483

47%

* Rounded to the nearest whole number

The characteristics table is usually used to confirm that there are not significant differences between groups and there weren’t with the three PREDIMED dietary interventions. However, we can use these data to see how many of all the 7,447 participants had Type 2 diabetes for example, or were male. This number is shown in the total column. The final column shows the proportion of people who had each characteristic e.g. 43% of all participants were male.


Characteristics of people who had an event

Figure 2, later on in the main PREDIMED paper, contained the number of events for each characteristic in Table 2. This gives us the opportunity to calculate genuine risk factors. If, in a study of a number of people, 1% have diabetes and yet 99% of people who have an event have diabetes, we could say that diabetes is a huge risk factor. The numbers would virtually guarantee that someone with diabetes would have an event. I like to call this ‘punching above one’s weight.’ Conversely, if 99% of people are overweight and only 1% of events are experienced by overweight people, we could confidently say that being overweight is not a risk factor.

In the table below, I’ve taken the 288 events recorded in the PREDIMED study and shown the number that occurred in each risk factor group. e.g. 190 of the 288 events occurred in people with Type 2 diabetes. The final column is the interesting one. This is the ‘punching above one’s weight’ column. This is the proportion of people with this characteristic who had an event divided by the proportion of people with this characteristic in the study.

This is summarised below:

Participants

As a % *

Events

As a % *

Punching above?

Numbers

7,447

288

Type 2 Diabetes

3,614

49%

190

66%

1.36

Hypertension

6,162

83%

237

82%

0.99

Dyslipidemia

5,383

72%

175

61%

0.84

Family history premature CHD

1,668

22%

57

20%

0.88

Smoking ever

2,883

39%

154

53%

1.38

Male

3,165

43%

171

59%

1.40

BMI >30

3,483

47%

138

48%

1.02

* Rounded to the nearest whole number

If the ‘punching above’ number is 1.0, or very close to 1.0, the same proportion of people who had this characteristic had an event e.g. 47% of participants had a BMI over 30 and 48% of events were recorded among those with a BMI over 30, so this is not a particular risk factor. The events were as would be expected if all events were distributed evenly across participants. This means that hypertension was also not a particular risk factor, with a score of 0.99 – so close to 1.0.

If the final column is less than 1, fewer people who had this characteristic had an event than might be expected from even distribution of events across all types of people. e.g. dyslipidemia: 72% of participants were described as having dyslipidemia, yet only 61% of events occurred in people with dyslipidemia. So dyslipidemia was not a risk factor – arguably it was protective – ha ha. Interestingly, family history of premature coronary heart disease (CHD) was also not a particular risk factor in this study. Fewer people with this characteristic had an event than would have been expected with an even distribution.

If the final column is greater than 1, more people who had this characteristic had an event than might be expected. Here we can see genuine risk factors. These turn out to be two that absolutely won’t surprise us – the classic “being male” and smoking. The smoking options reported were “never smoked”, “former smoker” and “current smoker.” Adding the former and current smokers together gives the “ever smoked”, as opposed to “never smoked.” 39% of participants were “ever smokers” and 53% of events were recorded among those who had ever smoked. We can give up smoking. We can’t change the gender we were born with (we can change gender, but not what we were born with) and so the classic “being male” risk factor is a significant issue about which we can do nothing.

The final stand-out risk factor in PREDIMED is Type 2 diabetes: 49% of participants had Type 2 diabetes at the start of the study (this was one of the recruitment options, hence the high proportion) and 66% of the events occurred in the group with Type 2 diabetes. Type 2 diabetes thus punched significantly above its weight in PREDIMED.

Conclusion

In last week’s review, I reported that the authors’ conclusion was “Among persons at high cardiovascular risk, a Mediterranean diet supplemented with extra-virgin olive oil or nuts reduced the incidence of major cardiovascular events.”

I explained why my conclusion would have been: “Among Spanish people, aged 55-80, mostly women, at high cardiovascular risk, a low-fat diet increased the incidence of major cardiovascular events (only strokes significantly) by 3 events per 1,000 person years.”

Another conclusion could have been drawn from the seminal PREDIMED paper (Ref 1): If you wish to avoid a cardiovascular event: don’t be born male; don’t smoke – ever; and don’t get Type 2 diabetes. Don’t worry about having a BMI over 30. Don’t worry about so-called hypertension. Don’t worry about any nonsense you are told about your cholesterol, HDL, LDL, whatever… Don’t even worry that much about family history of CHD. Put all your effort into never smoking and not getting Type 2 diabetes. That’s what PREDIMED should have announced.


References
Ref 1 Estruch R et al. “Primary Prevention of Cardiovascular Disease with a Mediterranean Diet.” New England Journal of Medicine. (2013).

6 thoughts on “PREDIMED – what really is a risk factor?

  • July 15, 2020 at 11:51 am
    Permalink

    Risk factor , correlation, association is not causation. When will the medical world wake up to this?
    Nutritionists, dietitians (?) do not get that.

    “and the pig got up and slowly walked away”

  • October 23, 2019 at 8:14 am
    Permalink

    Aw, c’mon Zoe, you know that if there is sufficient funds, statistics can be found to prove anything. Harvard appear to find this useful.
    Lies, damn lies, and statistics. It is called advertising to attract more money.

  • September 19, 2017 at 2:51 am
    Permalink

    Very very interesting.
    The combining of low HDL and high LDL-or-statin confuses things. Those with diabetes will tend to have low HDL, but they could also have low LDL. Or, those with high LDL could be getting some benefit associated with statin use (possible with diabetes, at least in a non-randomised population). Presenting the component parts of “dyslipidemia” separately would have cleared this up.

    • September 22, 2017 at 3:35 pm
      Permalink

      Agreed! The only part of my lipid panels I take notice of is the trigs/HDL ratio which I consider to be an indicator of insulin resistance. Mine went from nearly 7 to 1 or generally less, and has stayed there consistently since low carbing so I no longer bother to get the lipids checked.

      Better of course would be an actual test of insulin, and HOMA calculation which is impossible to get on our NHS (in fact a lot of patients no longer even have their trigs reported). Would have been good to see how that changed in this trial too. IMO much more relevant.

  • September 18, 2017 at 5:58 pm
    Permalink

    Zoe, another great analysis! Your conclusions are spot on and pretty funny!

    On a side note, I was grabbed by your title and the use of the phrase “risk factor”. I’ve long proposed that the phrase should be eliminated because it is so misused and generally misinterpreted. Even the Wikipedia definition says both “…a variable associated with an increased risk of disease or infection”, and later says “…Some prefer the term risk factor to mean causal determinants…”. No wonder so many people automatically assume that an “association” is causal. If words such as “only associates with” were used instead, and “causally linked” reserved for real cases there would be a whole lot of scholarly articles that would be far less exciting.

    • September 18, 2017 at 8:55 pm
      Permalink

      Hi Robert
      Good point – let’s ban it!
      Best wishes – Zoe

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.