Tip:
Highlight text to annotate it
X
So, with the approach to analysis,
first of all, we worked variable by variable
and looked to see whether they were associated with WEMWBS.
So we used linear regression models for continuous,
when we'd got the GHQ or WEMWBS, the continuous GHQ,
and a regression analysis, logistic regression,
when we were looking for dichotomised outcomes with the GHQ.
And in every model we were adjusting for socioeconomic confounders.
I'll show you those in a sec.
And then the variables, the lifestyle variables,
that were associated at the 10% level,
which is a pretty lenient level,
we then put in a model together to say,
OK, of those that were associated, which were the most important?
And again we were adjusting for socioeconomic factors.
And then in the final denouement with that big multivariate model,
we bunged in the GHQ as well and said,
OK, so how much of this association
between this lifestyle factor and WEMWBS
is actually due to mental health problems as manifested by the GHQ?
So those were the socioeconomic factors we took into account
and those were the response categories we used.
When we did the univariable analyses with WEMWBS as the outcome,
we got four factors.
We got smoking, exercise,
oily fish consumption and fruit and vegetable consumption
coming out as associated with WEMWBS.
We didn't get alcohol consumption or the frequency of alcohol consumption
and we didn't get sugar.
Now, I am absolutely confident the reason we didn't get alcohol
is because of the way the variable was administered.
We know extremely well that high levels of alcohol consumption
are associated with mental health problems,
and I wouldn't for one minute want you to run away
saying I'd proved anything different.
But the sugar one is perhaps more interesting.
There is... I have done some work on obesity and well-being,
and, curiously and contrary to what one might expect,
people who are overweight and obese
tend to have slightly higher levels of wellbeing
than people who are normal weight.
And so there's something odd going on there
that doesn't really accord with the idea
that people eat 'cause they're stressed
or because of their emotional problems of some sort or another.
And that may be why we're getting
that difference with the sugary foods.
So when we put all these,
the factors that were associated at the 10% level,
the lifestyle factors, into a model with the socioeconomic factors,
long-standing illness and age,
if your age distribution is U-shaped, like ours,
you have to have a quadratic factor,
and people who are very statistically minded will know all about that,
but we don't really need to bother about it.
What I've put here is the F values,
which is the kind of multivariate correlation,
so the stronger correlation, the stronger the association.
And what I want you to see from this
is it was the fruit and veg that were coming over most powerfully
as predictive... associated with WEMWBS.
It's important that this is cross-sectional data
and people say, use the word "predict" in logistic regression
and ordinary regression models,
but actually in epidemiological terms,
it can mean something slightly different,
so I will try and avoid it.
Oily fish came out with significant at the 0.05 level,
which is the lowest level.
But exercise and smoking dropped out of the model.
Now, the exercise, I think there are measurement issues,
but it is interesting that smoking wasn't correlated.
If everybody who didn't smoke ate healthily all the time,
you would expect only one or the other to end up in the model.
But what's happening here is that this is the stronger predictor
and it's taking a lot of the effect away from these others.
So, once again, I don't think I've proved
and wouldn't want you to run away with the idea
that smoking and mental health are unrelated.
They are, you know, and people with mental illness
have much higher smoking rates them people without.
But in this model, that's what we got.
Long-standing illness was very highly correlated.
But what I wanted to show you was that economic activity,
which is a very powerful inequalities marker,
and one that we know is associated with well-being,
but if we look at the F values, actually fruit and veg
is more strongly associated with mental well-being
than economic activity.
Economic activity is negatively... economic inactivity,
so these are people who are not even seeking work.
They're retired
or permanently incapacitated, and that's negatively correlated,
but I haven't done those subtleties on that.
So that was that model.
If we look at what predicted... what was associated with the GHQ,
we get the same four variables
coming out in the univariable analyses and not these ones.
But you'll notice the P values are slightly different
and that when we go into the multivariable model with the GHQ,
you will start seeing those differences coming through.
Because it was smoking that was much the most predictive
of the lifestyle variables for mental health problems.
That's very nice, really, you know. It's the sort of...
You undertake these analyses
and you've no idea what you're going to find
and then you find this pattern that really makes some sort of sense.
Fruit and veg are coming through in this model a bit,
but they're not anything like as strongly,
and these two have dropped out.
But, once again, you can see, you know,
if you look at people who are economically inactive,
it's the same sort of order of magnitude as the lifestyle variables,
in terms of predicting...
having an effect on mental health.
So then, when we put into our WEMWBS model,
we put the GHQ, and we say,
OK, so how do things change now?
Well, the GHQ F is somewhere over here.
So I just put lots of little pluses here.
I mean, they're bound to be, 'cause we know they're correlated at 0.7,
so they're going to be in this model.
But what is fascinating is that fruit and veg and oily fish
were both predicting and associated with WEMWBS
at the same level they were with in the models without GHQ.
So whatever is going on between
fruit and vegetable consumption and wellbeing
is not accounted for by any relationship with mental illness.
So what does all that tell us?
Well, this is a cross-sectional survey
and you can't really start imputing causation
from cross-sectional surveys
'cause you don't know whether people who are mentally well
eat more fruit and veg because they're mentally well
or they're mentally well because they eat more fruit and veg.
You've got that chicken and egg situation
and the only way to unravel that
is to look in cross... longitudinal data sets.
And, indeed, as longitudinal data starts coming on board,
that's what we will start doing.
However, it really, it was interesting
how strongly these variables came out.
And it's very interesting that they are entirely independent
of either long-standing illness or the other lifestyle factors
or the various socioeconomic factors.
So at every level of socioeconomic...
every socioeconomic level,
those who ate more fruit and veg had more mental wellbeing.
And we get a very different picture with the GHQ.
Oh, I'm sorry, I didn't say on the way through,
we did all those analyses with the GHQ dichotomised as well,
and smoking was the only variable that came out as predictive,
and it's much more difficult to see any pattern there
because you lose so much power.
But the GHQ was the one...
Smoking was the one that was most closely associated with the GHQ
with mental... poor mental health,
and we did get a little bit of fruit and veg coming through.
Now, I think those findings are...
Well, they're a kind of starting point.
I mean, they don't... Sorry.
On the way through, yeah.
I think the fact we didn't find an association
with alcohol and physical activity, and indeed particularly these two,
is likely to be something to do with the way the variables
were measured in the survey.
And we also have this issue of covariance
between the different lifestyle factors.
If you put them all into a model,
the ones that are less strong are going to drop out.
That doesn't mean to say they're unimportant.
So just a little reflecting on those findings.
It would be very nice to come away
and conclude that we've got marvellous evidence
that all you need to do to improve mental wellbeing
is eat more fruit and veg,
and that would be very good, wouldn't it?
It may well be that these findings are borne out
in longitudinal surveys.
And, you know, what we've said is it's very important
to start looking and to see whether that is the case.
At the moment, all you can say
is that these two things are very closely associated
and we don't as yet know in which direction.
Thank you very much.
(applause)