Monday 30 August 2010

On the level about epidemiology

[This is a guest post by Rob Lyons, deputy editor of Spiked

Richard Wilkinson and Kate Pickett have made great play of the fact that The Spirit Level is based on what they see as a science: ‘social epidemiology’. Leaving aside the dubious nature of their conclusions, it is worth noting that since the field of epidemiology has expanded, its explanatory powers have shrunk. While the ability to examine the spread of infectious disease has been invaluable, employing epidemiology beyond this has been - for the most part - bad news for our understanding of health. It's even worse as a method of understanding society.

The most famous early example of the application of epidemiological ideas is the ingenious manner in which John Snow, a London physician, managed to identify the source of a cholera outbreak in Broad Street, Soho in 1854. Simply by mapping the cases of cholera, he found they were clustered around one particular well. Whether Snow’s intervention to disable the well by removing the handle of its pump really saved the day is a matter of some dispute, but his method of plotting cases and looking for concrete causes was inspirational.

Fast forward to the 1950s, where a classic study by Austin Bradford Hill and Richard Doll identified for the first time (at least, outside Germany) the link between smoking and lung cancer. Bradford Hill and Doll sent a questionnaire to doctors to measure their smoking habits, then followed up the questionnaire some years later. Those doctors who smoked were much more likely to have developed the disease and no one would seriously disagree now that active smoking is a very significant risk factor for lung cancer.

There is a proviso, however: even with smoking, the most famous of these epidemiological connections, most of those engaged in the risky behaviour - smoking - do not die from lung cancer while there are a small number of non-smokers who die from lung cancer, too. Other explanatory factors, like a genetic predisposition to developing certain cancers, a less-than-ideal diet and sheer luck seem to come together, too.

Despite the search for other similar environmental health risk factors, from passive smoking to bacon butties, epidemiology has proven to be a fairly crude tool. Austin Bradford Hill, despite the success of his groundbreaking study, advised that epidemiological methods should be applied cautiously. He listed a variety of tests that should be applied to any putative connection between a risk factor and a disease, summarised in John Brignell’s book The Epidemiologists (see my review here: http://www.spiked-online.com/index.php/site/article/2446/).

1. Strength: Is the association strong enough that we can rule out other factors?
2. Consistency: Have the results been replicated by different researchers, and under different conditions?
3. Specificity: Is the exposure associated with a very specific disease as opposed to a wide range of diseases?
4. Temporality: Did the exposure precede the disease?
5. Biological gradient: Are increasing exposures associated with increasing risk of disease?
6. Plausibility: Is there a credible scientific mechanism that can explain the association?
7. Coherence: Is the association consistent with the natural history of the disease?
8. Experimental evidence: Does a physical intervention show results consistent with the association?
9. Analogy: Is there a similar result to which we can draw a relationship?

When these tests are applied rigorously, nearly all the postulated links between risk factors and harmful effects fall apart. For example, in the case of passive smoking and lung cancer, the association is weak, it is not consistent, exposure is usually assumed rather than measured, and so on. In other words, an association between passive smoking and lung cancer is either non-existent or too weak to worth worrying about.

Let’s now turn to this new field of ‘social epidemiology’. We can see in The Spirit Level that Wilkinson and Pickett’s claims fall flat on their faces when Bradford Hill’s tests are applied. The associations are not strong, rarely rising much above ‘apparently random’; other explanations are usually both available and more plausible than the idea of inequality leading to status anxiety; other researchers, simply by minor tweaks in the data, have seen the associations disappear; and the wide selection of social problems deemed to arise from inequality is a long way from ‘specific’.

As someone who would broadly see themselves as left-wing, I find there is a more pressing problem with The Spirit Level: that it dumbs down the fight between different groups in society over the carving up of the wealth we produce from a genuine struggle for material betterment to a psychological disorder demanding intervention from on high. Seeing inequality in terms of status envy is thus actually disempowering rather than liberating. But regardless of our attitude to the political question of wealth distribution, The Spirit Level is well worth criticising as bad epidemiology, too.

15 comments:

Fredrik Eich said...

"Other [than smoking] explanatory factors, like a genetic predisposition to developing certain cancers, a less-than-ideal diet and sheer luck seem to come together, too."
And let's not forget that the volitilty of lung cancer rates has a shocking disregard for smoker prevelance over time and space and is far more strongly attracted to areas of
high precipitation. here and here. And rain does not cause lung cancer either!

Rollo Tommasi said...

Interesting article. But I don’t recognise Rob Lyons’ assessment of how Bradford Hill’s tests apply to evidence about the association between passive smoking and lung cancer.

The strength of association is not strong but, in Bradford Hill’s own words, it does not necessarily need to be (“We must not be too ready to dismiss a cause and effect hypothesis merely on the grounds that the observed association appears to be slight.” - http://www.edwardtufte.com/tufte/hill ). The association IS consistent – being found in both case control and cohort studies; both home and workplace studies. And the association satisfies most of the other criteria, other than specificity and possibly experiment.

Here are a couple of assessments which determine that the Bradford Hill criteria are satisfied. Can Mr Lyons provide professionally reputable sources which back up his claims?

http://ije.oxfordjournals.org/cgi/reprint/36/5/1048 - page 1055
http://www.advisorybodies.doh.gov.uk/coc/ETS.PDF

Anonymous said...

Rollo Tommasi.
Bradford Hill gives Weil's disease and Meningitis as his examples of cause but with a low association. However these differ from passive smoking and lung cancer in the following respect. Exposure to rat's urine and the harbouring of the meningococcus are, respectively, NECESSARY conditions. You have to imagine a world in which there exists a special kind of cancer, whose only cause is passive (not active) smoking, and where there is some critical necessary level of passive smoking exposure, which can be measured for any individual.

I don't have much confidence in the journal article you cite. In the late 90s the whole issue of passive smoking became almost a religion. I cannot take seriously a scientific paper which ends with the paragraph,
"The tobacco industry tends to understate the effect of passive
smoking on occurrence of lung cancer, and has tried to
introduce ventilation as a solution for passive smoking in
enclosed public areas.143 But this does not prevent exposure
of workers in these areas to passive smoking. The preferable
public health policy is still a total ban on smoking in
public areas."

To my mind, Doll, on Desert Island Discs and Peto before the Parliamentary Committee, sum up best the beliefs of rational people who aren't on a mission. That is: frequenting smoke-filled bars is like smoking 10 cigarettes a year - probably not as good for you as trekking through fresh Alpine air, but no more serious than the health scares we read about every week: about which people are allowed to make up their own minds. I notice today's is that walking would save 10000 cancer deaths a years. If that's the case, desk jobs should be immediately outlawed.

Anonymous said...

Ah! ROLLO,where ya been dudette..I see your still trying to sell shs/ets as deadly.But I see youve fallen back a tad.....ever heard of an insignificant health risk.

Here I will provide some incite!

There have been 34 studies into lung cancer and exposure to cigarette smoke as a child. 3 suggest a raised risk, nearly four times as many 11 suggest PROTECTION with 20 suggesting no raised or reduced risk. The most famous is the World Health Organization 1998 study which concluded:

"Results: ETS exposure during childhood was not associated with an increased risk of lung cancer (odds ratio [OR] for ever exposure = 0.78; 95% confidence interval [CI] = 0.64–0.96)."

"Conclusions: Our results indicate no association between childhood exposure to ETS and lung cancer risk."

Rollo Tommasi said...

Anon (16:10): We are all entitled to our own views about interpreting the Bradford Hill criteria. However, the difference to date is that neither you nor Rob Lyons (assuming you are not he) have provided professionally reputable sources to back up my claims.

What exactly is your concern about the Taylor et al quote?

"The tobacco industry tends to understate the effect of passive smoking on occurrence of lung cancer” – True, at least it was when the report was produced.
“The tobacco industry ….. has tried to introduce ventilation as a solution for passive smoking in enclosed public areas.” - True
“But this does not prevent exposure of workers in these areas to passive smoking.” – Also true.
“The preferable public health policy is still a total ban on smoking in public areas." – Reasonable conclusion

All in all, that quote seems pretty reasonable to me.

It’s interesting you are happy to place such weight on the views of Doll and Peto. So may I assume that you also agree with these statements of theirs:

“As a responsible citizen, I believe that nobody should have to work in an atmosphere polluted by other people’s smoke” – Doll (http://www.bma.org.uk/images/Smokefree_tcm41-20691.pdf)
“Environmental tobacco smoke that people experience at work or at home is definitely a cause of lung cancer.” – Doll (http://news.bbc.co.uk/1/hi/health/2053840.stm)

“The definite statement is that some people are killed by breathing other people's smoke, and then there is reasonable uncertainty about the number killed.” – Peto to House of Lords Economic Affairs Committee
“It is when you look at the totality of the evidence and try to get all of the evidence together, and try to assess the extent to which selective publication of positive results could cause problems, then I think you do get a consistent picture that there is some excess risk of lung cancer among those who are exposed to passive smoking, from other people's smoke, on a regular long term basis.” - Ditto

Rollo Tommasi said...

Anon (19:21): Delighted you’re so keen to find me here. I have to ask, what exactly is the relevance of studies into risks of lung cancer from only childhood exposure to this debate?

For what it’s worth, I don’t accept your conclusion about what these studies tell us. Beside the fact that you have cherry-picked one study for no better reason than that it is “famous”, looking at the evidence as a whole tells a somewhat different story to what you state. The US Surgeon General’s 2006 report found that:

“When a pooled RR estimate in association with maternal and paternal smoking was calculated, in addition to a calculated combined index that represented childhood exposure from either parent, there was some increase in risk in association with secondhand smoke exposure to maternal smoking (OR = 1.15 [95 percent CI, 0.86–1.52]), paternal smoking (OR = 1.10 [95 percent CI, 0.89–1.36]), or smoking by either parent (OR = 1.11 [95 percent CI, 0.94–1.31]).”

Anonymous said...

Rollo,what part of insignificant do your numbers show! Fancy dancing and cherry picking of studies as the SG 2006 report used leaving out certain other studies Brown comes to mind and enstrom and kabats for another....Besides knowing that many others in tobacco control actually wrote the
Sg report to include stanton glantz and the ACS for another it would leave no doubt that the report itself is more propaganda and intellectual prostitution to come up with the few things they did. Yet there is no proof of any claim except for odor anoyance.

Had the sg included other studies and not done fancy math the actual RR would have been closer to a .70 maybe a .80.But,hey tobacco control has millions or had to waste on psuedo-science and propagandace.But when it came down to actual science OSHA had to stand firm. No rules here,but of course the clinton administration made sure the OSHA head trumped out some propaganda to cover their butts in tobacco control and keep the SHS/ETS junk alive. Yet the OSHA page directs people to the debunked EPA study.It just doesnt get more desperate than that.

Rollo Tommasi said...

I’ve seen claims that the Surgeon General’s report “cherry-picked” results, but I’ve never seen any proper substantiation for this. At least you refer to a couple of examples which you claim were cherry-picked out (although you don’t offer any evidence that they were deliberately omitted).

Of these, sorry but I don’t recognise a study by Brown. As for Enstrom & Kabat, it was not cherry-picked out. It is referred to in the appendix as one of many studies excluded from the report because they were published too recently. Several of these other reports were published before Enstrom & Kabat, so it is not as if it was deliberately ignored.

Your claim that “had the sg included other studies and not done fancy math the actual RR would have been closer to a .70 maybe a .80” is both unsubstantiated speculation and totally wrong. Including E&K makes no difference to the overall conclusion that passive smoking increases the risk of lung cancer. Want proof?

Here’s some. A member of the IARC working group has shown that adding E&K to the IARC analysis of the increased risk of lung cancer would have changed the IARC estimate from 1.24 to 1.23. - http://www.bmj.com/content/327/7413/501.3.full

And here’s more. The 2007 Taylor et al meta-analysis included E&K. And what were its results? A pooled relative risk of 1.27 (95% CI 1.17–1.37). - http://ije.oxfordjournals.org/cgi/content/full/dym158v1#B51

If you want your criticism of the Surgeon General’s report to be taken seriously, you need to offer more than unsubstantiated smears.

Carl V Phillips said...

I definitely agree with Mr. Lyons's analysis and conclusion, but I have to take issue with the reference to Hill's criteria. Epidemiology, as widely practiced, is junk science in a lot of ways. The notion that there can be a set of criteria for assessing whether a correlation represents causation is one bit of junk thinking that is popular among many people doing epidemiology (though not the leading thinkers). I and others have written about this at length. A couple of my articles are among the leading comments:

http://www.epi-perspectives.com/content/1/1/3

http://www.ete-online.com/content/3/1/5

(though mine are not the most read -- that would be the criticism in the main advanced textbook in the field).

One of the clearest criticisms of the use of "criteria" (credit to Weed for this one, not me) is that they are used idiosyncratically. That is, when someone invokes the concept, they invariably pick and choose from the list to find ways to support their claim (that there is or that there is not whatever causal relationship). This post, and especially the comments, illustrate this point nicely. The equally fundamental problem is that there is no rule for determining whether a "criterion" has been "met" (which is to say that they are not actually criteria).

...

Carl V Phillips said...

...

Another relevant point (credit to Poole primarily for this one) is that most people who talk about Hill's considerations miss "number 0" on the list, the fact that there needs to be a correlation in the first place. In the case of Wilkinson, numerous critics have pointed out that the main problem is that there is not really such (unless you cook the data in just the right way, and then only barely). The same is the main criticism against ETS exposure causing geriatric diseases. It is extremely plausible that it would, but the hitch is that if you look at the data, it does not (on average, when considered honestly) show much of a correlation.

In any case, none of this takes away from this and other analyses about what is flawed about The Spirit Level. The arguments here are good -- its just that they are not made any better by trying to match them to items on Hill's list. As I try to make clear in one of the above papers, someone who understands scientific reasoning will find that the Hill considerations cover many of the points one might want to reason about. But invoking his list as other than just a reminder list for things to think about, as if it is "criteria" that can be shown to be true or not, is not a good strategy unless your goal is to try to show much of what passes for epidemiology to be junk science because a large portion of people doing it believe this is possible.

Anonymous said...

The Science of ETS


All of studies on ETS are either Epidemiological Studies

or Meta Analysis Studies.


Conditions required for epidemiological validity


A study must warrant that its numerical representations of individual lifetime PS exposure recalls are true measures of actual exposures.
A study must warrant that an exposure recall bias affects cases and control groups, and exposed and non-exposed groups at the same rate.
A study must warrant that subject selection and misclassification biases affect cases and control groups, and exposed and non-exposed groups at the same rate.
A study must warrant that known causal confounders affect cases and control groups, and exposed and non-exposed groups at the same rate.
A study must warrant the accuracy of pathological and diagnostic records.
The results from different studies addressing the same subject must be consistently reproducible.
In any study, the statistical margin of error of reported risks should reach no less than the 95% level of significance.
If the above criteria are met, the results of a study should also be consistent with Hill’s criteria of causality.
Meta-analysis summations shall not be credible unless performed on the basis of all available studies, which studies also must be of homogeneous design and conduct, and must have met the above criteria of validity.


Note all major studies on ETS have been conducted in smoking households.

Anonymous said...

Hill’s criteria for Epidemiology Studies


Sir Richard Doll (1912-2005), the father of epidemiological studies on tobacco and disease and his mentor Sir Austin Bradford Hill (1987-1991) established the criteria for case controlled studies in epidemiology.


Hill's explicit statement that cause-effect decisions cannot be based on a set of rules. Overlooked are Hill's important lessons about how to make decisions based on epidemiologic evidence. He advised epidemiologists to avoid over-emphasizing statistical significance testing, given the observation that systematic error is often greater than random error. His compelling and intuitive examples point out the need to consider costs and benefits when making decisions about health-promoting interventions


• Statistical significance should not be mistaken for evidence of a substantial association.

• Association does not prove causation (other evidence must be considered).

• Precision should not be mistaken for validity (non-random errors exist).

• Evidence (or belief) that there is a causal relationship is not sufficient to suggest action should be taken.

• Uncertainty about whether there is a causal relationship (or even an association) is not sufficient to suggest action should not be taken.


In 1953: Sir Richard Doll establishes a correlation between actively smoking cigarettes and lung cancer risk. However, over subsequent decades, in 2005, Doll drastically changed his views and gradually emerged as a major defender of corporate industry interests. As a member of a recent IARC scientific working group, convened to review evidence relating tobacco smoking and cancer, Doll finally admitted: "It does look as if it's the cancers that are principally caused by hormones that are not affected by smoking. Most of the other cancers throughout the body are induced by exposure to chemicals, often environmental ones“.

Anonymous said...

Validity of Epidemiological Studies


Last July, introducing his office's latest report on secondhand smoke, then-U.S. Surgeon General Richard Carmona asserted that "there is no risk-free level of secondhand smoke exposure."
In an effort to circumvent this capital obstacle, all secondhand smoke studies have estimated risk using a misleading marker of "lifetime exposure." Yet, instant exposures also vary uncontrollably over time, so lifetime summations of exposure could not be, and were not, measured.
Typically, the studies asked 60--70 year-old self-declared nonsmokers to recall how many cigarettes, cigars or pipes might have been smoked in their presence during their lifetimes, how thick the smoke might have been in the rooms, whether the windows were open, and similar vagaries. Obtained mostly during brief phone interviews, answers were then recorded as precise measures of lifetime individual exposures.
In reality, it is impossible to summarize accurately from momentary and vague recalls, and with an absurd expectation of precision, the total exposure to secondhand smoke over more than a half-century of a person's lifetime. No measure of cumulative lifetime secondhand smoke exposure was ever possible, so the epidemiologic studies estimated risk based not only on an improper marker of exposure, but also on exposure data that are illusory.
In addition, results are not consistently reproducible. The majority of studies do not report a statistically significant change in risk from secondhand smoke exposure, few studies show an statistically insignificant increase in risk, and astoundingly some show a reduction of risk.

Rollo Tommasi said...

It seems rather ironic that, after Dr Phillips’ carefully considered views, we are subjected to a series of cut-and-pastes by Anonymous, all taken from a cranky and jaundiced presentation by a US pro-smoking group: www.virginiasmokersalliance.com/VASArev2.ppt

Anonymous said...

Rollo, very much unlike your uncranky, unjaundiced views.

Puhhhh-leeeez!