Damn Statistics!

Damn StatisticsIf you’ve been following this blog for a while, then you will have spotted that I’ve written a lot about risk factors.  This is due to the glut of stories in the Media that have come out recently.  There was the “tall people are more likely to get cancer” one, then the “meat consumption causes cancer” one, then the “alcohol causes cancer” one and then the “Obesity causes cancer” one.

And if you’ve read through those posts, you will notice that I don’t seem all that convinced by the claims.  In each post, I seem to be downplaying the risks….saying that things are not as bad as the reports suggest.  Well, there is a reason for this.  Basically, I think that, in each case, the reports are overly-simplistic and that makes me take them with a pinch of salt.

So, I thought I’d use this post to try and explain why.  What do I think is wrong with these correlations?  Well… nothing, actually.  The correlations themselves are fine.  But when it comes to how the results are interpreted…. that’s a different story.

First off, how are these correlations identified?  Scientists & medics will survey cancer patients, to try and identify common factors in their environment, lifestyle, medical history or genetic makeup, which could be contributing to the onset of disease.  This is known as “Epidemiology“, and it was through the use of this type of scientific study that, eg. smoking and asbestos exposure were identified as risk factors for lung cancer.  It is also how meat consumption, obesity, height and drinking alcohol were studied to see if they correlated with increased risks of cancer.

So, to investigate a potential risk factor, researchers will look at all of the patients with a certain type of cancer.  Then they will pick one factor – let’s say obesity, but it could be anything – and split the patients up into those who are obese and those that aren’t.  Then, they will look to see if more of the patients are obese, or if more of them are of normal weight.  If more colon cancer patients are obese than not, this suggests there is a link between obesity and disease onset.  Or, alternatively, the numbers may be about the same, suggesting no link to disease onset, but they might discover that obese patients are more likely to have a more advanced form of the disease, suggesting a link between obesity and disease progression.

But there area couple of problems with these correlations, and therefore with the subsequent news stories.  The first one, I’ve covered before back in Just One Cornetto: Correlation is not Causation.  Just because obesity might correlate with increased cancer risk, it doesn’t mean that it is a cause.

But the second problem with these correlations is even trickier: they are too simplistic.  In order to make a correlation, you have to pick ONE thing to assess.  It might be obesity, it might be alcohol consumption, it might be meat consumption… Doesn’t matter.  The point is that you isolate ONE of them, and then check the records to see if high or low levels correlate with increased cancer risk, or poorer survival.

But the problem is, by isolating out individual characteristics like this, you run the risk of focusing on the wrong issue and missing what is REALLY important.  This is because individual traits don’t exist in isolation, but are often linked to other characteristics.  For example, in the list above, people who are obese often consume more red meat.  And more alcohol.  So, if you find a correlation between obesity and cancer, it could be that what the REAL risk factor behind this is that overweight people are also more likely to be heavy drinkers, and it is the underlying alcohol consumption which is the real issue.  Or, it may be that, being more likely to be overweight, heavy drinkers are ALSO more likely to eat more red meat and THIS is what you should focus on.

Now, you could try and get to the bottom of this by subdividing the patients further, eg. separating them into “obese + high alcohol”, “obese + low alcohol”, “not obese + high alcohol”, “not obese + low alcohol”.  So, you look at four groups rather than two.  You could go even further, by adding meat consumption into the mix.  But that would double the amount of groups again, so you’d have eight groups to analyse.  Ooh!  But then overweight people also eat more sugar!  So maybe sugar consumption could be looked at too!  So, now you have SIXTEEN groups to analyse….

It’s like cake.  Yes! I said it!  It’s like CAKE!  With a cake, the fewer times you slice it the bigger the portions are, but fewer people get a piece.  But if you slice it more times, more people can get a portion, but the less satisfied everyone will be, because the pieces are so much smaller.

And that is the issue with the type of statistics that are carried out to look for cancer risk factors.  There is only a finite number of patients to start with.  So, the more you split them into groups for analysis, the fewer patients will be in each one.  And you need a LOT of patients if you want to do proper statistics.  So, the more groups you have, the fewer patients are in each one, and the fewer patients per group, the less reliable the statistics will be.  It is a Catch 22.

And THAT is the issue I have with the stories I’ve spoken about before.  The studies themselves are absolutely fine, but in order to do proper statistics, the studies have to concentrate on one factor.  But the fact that only single things can be looked at means that you have to be VERY careful when you are drawing conclusions about what the data is actually telling you.

But I can’t stop thinking about cake now……..(goes off to look for cake).

————————-

Pérez-Hernández,A.I., Catalán, V., Gómez-Ambrosi, J., AmaiaRodríguez, A., & Frühbeck, G. (2014). Mechanisms Linking Excess Adiposity and Carcinogenesis Promotion Frontiers in Endocrinology, 5 DOI: 10.3389/fendo.2014.00065

Bagnardi, V., Rota, M., Botteri, E., Tramacere, I., Islami, F., Fedirko, V., Scotti, L., Jenab, M., Turati, F., Pasquali, E., Pelucchi, C., Bellocco, R., Negri, E., Corrao, G., Rehm, J., Boffetta, P., & La Vecchia, C. (2012). Light alcohol drinking and cancer: a meta-analysis Annals of Oncology, 24 (2), 301-308 DOI: 10.1093/annonc/mds337

Bouvard, V., Loomis, D., Guyton, K., Grosse, Y., Ghissassi, F., Benbrahim-Tallaa, L., Guha, N., Mattock, H., & Straif, K. (2015). Carcinogenicity of consumption of red and processed meat The Lancet Oncology DOI: 10.1016/S1470-2045(15)00444-1

ResearchBlogging.org

AG McCluskey (2016). Damn Statistics Zongo’s Cancer Diaries

Advertisements

One thought on “Damn Statistics!

  1. Pingback: Kill “Bill” !!! | zongo's cancer diaries

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s