MPH Master pieces

Smokefree Scotland

 Here is an evaluation of the effect of the smoke-free laws in Scotland. This is one of those things we absolutely need to know about if we are in public health but the paper, particularly the methods section, is not an easy read. The basic design is a questionnaire survey.

There are two difficult bits in the methods section. Firstly the survey respondents were recruited in several different survey waves with top-up samples for drop out. All this will give plenty of material for the tobacco companies to rubbish the research. Our task is to form a balanced judgement about whether the resulting sample is likely to give us truth or not - and to accept that we may have to conclude with regret that it does not. Important to be alert to flaws, particularly when research appears to say what we want it to say. 

The other difficult bit in the methods section is the statistical methodology. There is a lot of weighting to make the samples representative of the base populations. But more tricky for the non-statistician is the use of generalised estimating equations (GEE). The main benefit of using GEE is to get right (or as right as possible) the confidence intervals on all the results. But it also permits analysis of interactions between variables. As the authors point out, the key question is whether the various measures changed more in Scotland than in the rest of the UK. They have before-and-after measures (i.e two waves of results) in two areas, so technically speaking they are looking for a wave-by-area interaction.

Aaargh! - a new definition

I'll just have a little rant.

There's a new definition of  acute myocardial infarct. All very necessary no doubt but it means everything you read from now on about myocardial must now be put in the 'perhaps' box. All those risk scores, for example, were constructed using the old definition. Trends must be viewed with suspicion - is it real or just an effect of the new definition? Papers on the effectiveness of treatments (and hence cost-utility analyses) likewise were based on the disease as defined the old way.

When the ICD changes, for example from ICD9 to ICD10, an exercise known as 'bridge coding' is undertaken by the Office for National Statistics. Every death certificate is coded twice: once using ICD9 and again using ICD10 - more about this here . That way we know exactly the effect of the coding change.  But no-one will do that for the change in definition of AMI.

'Must read' papers on prostate screening

...so if you are in public health you have to read these two papers: one from Europe and one from the USA on prostate cancer screening. There is also a good commentary.

Our task here is to have thought about it more carefully than the tabloid journalists! 

Fire death

This is an interesting paper on trends in fire injury and death. When you read a paper like this the key question is the quality of the data - particularly how complete and accurate they are. This can be difficult to judge if you' re not familiar with the primary source of data. For a good example of the problem see this paper by the two Mikes - Gill and Goldacre. Police statistics suggested a decline in traffic injuries which was contradicted by NHS data.

In this case the source data is captured by the fire service and collated for publication by the relevant government department. The authors discuss data quality at the end of their paper, and this section is worth reading carefully. Note for example that they had to exclude 'precautionary check' from the data on injury and that in about a quarter of the fatal fires the cause was not recorded.

Overall the mortality trends seem more certain than the analyses of injuries and cause of the fire. 

More on diets

Here's a solid study on weight loss diets: randomised, blinded and follow up for two years. The scientific question was whether the overall %age of fat or carbohydrate genuinely affects weight loss, given that overall you ensure a lowered calorie intake. I won't spoil the excitement by giving you the answer to that question, but there is also a lot of interesting info on the time course of weight loss and diet adherence over two years.

The last head-to-head trial of four diets was organised funded by that well known medical research institution the British Broadcasting Corporation. Not so sound scientifically - no blinding and only 12 month follow up - but similar take-home message.