MPH Master pieces

Air pollution

You don't often see a nice simple linear regression nowadays but here's one.

The study looks at life expectancy in counties of the USA to see the effect of air pollution. The authors assembled data on life expectancy and on air pollution in 211 counties of the USA at two time periods - around 1980 and around 2000. In general life expectancy improved over the 20 years, as did air pollution.

Of course other things might account for the change in life expectancy. People got wealthier. Smoking decreased. In the USA black people are known to have lower life expectancy than white people, so a large influx or outflow of black people from a particular county might have accounted for some of its change in life expectancy. Measures of these are included in the regression models.

The measure for smoking is interesting. The researchers couldn't get actual smoking data for the 211 counties for the two time periods, so they used a proxy - mortality from lung cancer and from COPD. That seems sensible but lung cancer and COPD arefairly major causes of death so using them as predictors of life expectancy (which is just a summary of mortality rates) seems rather incestuous to me. This is the NEJM so I guess the statistical referees would have complained if it was a serious problem.

Meta analysis and fish oils

Fish oil is good for you - right? Well, maybe.

An excellent commentary, prompted by a new meta-analysis, tries to make sense of the data. And as Brunner points out, this stuff matters - we may be driving tuna to extinction on the back of our public health messages.

Part of the problem is that 'fish oil' is not a very precise scientific category. This doesn't matter if you can eat fish twice a week, but most of us can't. And once you move to tablets, you need to know exactly which fish oil to put in them. So we need at least to distinguish eicosa from docosa - eicosapentanenoic from docosahexanoic acid.

The two biggest trials, which dominate the meta-analysis, were in Italy and Japan. It's probably not sensible to combine these two studies in a meta-analysis. Japanese and Italians have very different baseline diets, particularly in respect of fish. A good meta-analysis will show in the Forest plot the weighting of each study - in this case the Italian study contributes 60% and the Japanese one 26%, so the others hardly count for anything. 

Finally and biologically fish oils may produce their effect on mortality after myocardial infact by preventing arrythmia - or they may not. If fish oil reduces sudden death but not non-fatal cardiovascular events, that suggests an effect on arrythmia rather than thrombosis. But the evidence is mixed.

Screening for congenital heart disease

In theory a paediatrician examines all newborn babies, at least if born in hospital, to check that all is well. But of course nobody is perfect and things get missed - about 30% of the time in the case of congenital abnormalities of the heart. You can get this down to about 10% if you add in routine pulse oximetry according to a large  study in Sweden.

The careful discussion gives insight into the real world of screening, which is a lot more complex than a nice 2x2 table of screen positive and negative versus true positive and negative. Firstly of course there are plenty of decisions about what kit to use, what the cutoffs are for declaring a case to be screen positive, how often to repeat the screening test and so on. Then when you start to run the programme, some patients who have disease and screen positive get sent home by mistake (a 'protocol violation'). You detect things which were worth finding but which weren't the aim of the programme - so does that count as a false positive or a true positive? And you detect other things (e.g. some degree of pulmonary hypertension) that you really don't know what to do about.

The results section includes the highest relative risk I've ever seen in print (previous record holder - risk of hepatoma in people who are HBcAg positive) - screen positive on pulse oximetry gives a relative risk of 719.8 for duct-dependent heart disease.

Finally see what you think about the claim that early diagnosis can be cost saving, and overall the programme may be cost neutral. 

All in all, much food for thought.

Hand washing

This isn't exactly 'current reading' but I've only just come across it. An excellent account of a project to increase handwashing - and hence reduce infant mortality - in Karachi. Note that the intervention isn't just a bar of soap: it's the whole programme e.g.

'for the first two months of the intervention, neighbourhood meetings were held for the mothers in every neighbourhood two or three times a week.' 

This has relevance in the developed world if as we suspect (and to some extent supported by this paper) handwashing interrupts the transmission of pandemic flu and other respiratory viruses.

Suicide and the organisation of mental health services

This is an interesting one. The question is obviously important - does the way you organise services affect suicide rates? - and the method chosed to answer the question as good as one could ask for. And yet...

The basic method is simple, and ecological. Take a measure of how mental health services are organised - in this case a validated method of describing services. Then get the best possible estimate of suicide rates - in this case a Bayesian estimate, which is right for small area studies because it damps down extreme rates based on small numbers. Finally adjust for an obvious confounder: since suicide is closely associated with socio-economic conditions, an elegant factor analysis was used to identify two factors used in the control for confounding. Conclusion - yes, suicide rates are lower in places with community based services, and constant re-organisation is a bad thing. 

So what's the problem? Well, for one thing I like the conclusion - it's what I was hoping for. This is always an alert sign that critical faculties may have been dulled. I'm also suspicious that the study would not have made it into print if the results had been negative so perhaps we're looking at the tip of an unpublished iceberg of studies with less positive results.

The areas studied were tiny in population - average 5000, or about the size of a two-man UK general practice. How much sense does it make to try to describe the mental health services available for such a small population, particularly the inpatient component?

And finally, of course there is a lot more to mental health outcomes than suicide. so let's congratulate the authors on the clarity of their thinking and the elegance of the study, but as readers let's not be dazzled. An accompanying editorial sounds some sensible cautions.