MPH Master pieces

Beware observation!

There are a couple of mega-trials in this week's JAMA. Both looked at whether selenium or vitamin E could prevent cancer. It all started with a trial of whether selenium prevents skin cancer - the answer was no, but it was noticed that prostate cancer was reduced (by a massive 65%) in the intervention arm. Prostate cancer was also down 35% in another trial - of vitamin E supplements. Hence the current trials.

The actual trials reports are solid and worthy but the covering editorial is a good place to start. Various points of interest - these were very big trials so (a) statistically they put the issue beyond doubt and (b) no-one will ever want to fund a 'selenium and cancer' trial again. The editorial points out that the subjects were 'health well-motivated volunteers who had good access to care' - which may have reduced the scope for selenium to make any difference in this group with a startlingly low incidence, as it turned out, of prostate cancer in both the intervention and the control groups. 

But it all reminds us to be wary of observation only evidence and of secondary analyses in trials designed for other purposes.

Screening in Finland

A good piece on colorectal cancer screening in Finland. Unusually (uniquely?) for a national programme they decided to implement as a randomised trial. 

One of the interesting features of the write-up is the concept of 'test', 'episode' and 'programme' sensitivity. Sensitivity is the percentage of cases that you manage to detect. We normally think about test sensitivity - of the people with cancer, what percent screen positive?  - but test sensitivity only reckons people who were actually tested. Programme sensitivity takes account of people who don't turn up for screening - the non-responders - and so reports what percent of people in the target population whose cancers were picked up. I haven't managed to work out yet what 'episode' sensitivity is.

The other point to note is that we never really know how many people had cancer - the best we can do is to count how many people screened negative and were later diagnosed with cancer.

Finally three unfortunate people screened positive, were given the all clear at colonoscopy and then turned out to have cancer after all. Where would you put them in statistics of sensivity and specificity? (on reflection, perhaps that's what 'episode sensitivity' is about).

Worsen inequalities - make alcohol cheap

Here is an important report from Finland.

Alcohol became a lot cheaper in Finland in 2004 (a side effect of the European Union). The effect? Substantial increases in alcohol related mortality, especially among the unemployed and those with low education, social class or income.  It's a well written paper with careful attention to the details of data quality and easy-to-understand statistical analysis.

Governments usually make policy with plenty of debate and little evidence. So remember this paper next time there's a debate about the level of duty on beer, wine and spirits.