MPH Master pieces

Economics of TB control

Here is an interesting evaluation of a tuberculosis control programme in London. The title says it all - it's the Find and Treat service. Table 1 in the paper shows that very few patients are lost to follow up if referred to the service early (though confusingly only 60 - 70% seem to complete treatment: or maybe I've misunderstood that row in the Table.)

Tuberculosis is particularly common among groups particularly likely to default from treatment: the paper cites prevalences of 

"788 per 100 000 in homeless people, 354 per 100 000 in people with problematic drug use, and 208 per 100 000 in prisoners. By comparison, the overall prevalence of tuberculosis in London was 27 per 100 000 people"

And outside London prevalence is likely to be even lower - for example the incidence cited in a recent report for Surrey was 7 per 100 000 (albeit with some worries about underreporting).  (Note that some of these figures are for incidence and some for prevalence. Incidence you get from reporting or notification of new cases, which should be available everywhere; prevalence estimates require a special survey, or some assumptions about the relationship between incidence and prevalence.)

But back to the economics. The paper uses a model, which hinges on transition probabilities between different compartments, for example between being treated and being lost to follow up , or between being treated and being cured. This concept of transition probabilities between different states is very common in economic modelling. Sometimes costings in economic papers are bold guesses but here we seem to have accurate budget figures. The effect of the programme is estimated in years of life and gain in quality, using the standard metric of an EQ-5D score converted into a utility.

And there is a happy outcome after all this modelling - the service comes in under NICE's cutoff of £30k per QALY.   

Alcohol in Cali

It seems that there is no problem so bad that alcohol cannot make it worse.

Cali, Colombia has one of the highest murder rates in the world: 68 per 100 000 in 2008. Just pause for a moment to think what that would mean in your own home town.

At various times during the period 2004 to 2008, depending on political will, alcohol sale and consumption in Cali was prohibited from either 2am, 3am or 4am till 10am. A simple analysis (see fig 2) displays homicide rates according to whether the policy on the day of the homicide was lax (2am), moderate (3am) or strict (4am). The results seem pretty clear but there is also an appropriately complicated time series analyses to take account of various other possible explanations (confounders).

The ecological fallacy

The ecological fallacy is a problem of cross level inference - wrongly transferring conclusions from one level of aggregation to another. Specifically, applying group data to individuals. Here is a contemporary example:

Dropout from substance abuse programmes is associated with worse outcomes, so programmes with high follow-up rates should be better for patients. Right? No, because that's the ecological fallacy at work. This paper shows you the maths of the fallacy. I must admit that even when shown the maths I still can't quite see how it works. I've got a niggling feeling that it may be because facilities with very high rates of follow up are putting all their efforts into that - hitting the target - and not into providing good treatment. 

Makes you think about all those other hospital-level indicators, doesn't it?

Meanwhile over at the IJE, some diligent Dutchmen have shown that Robinson's original work on the fallacy contained an error which lay undetected for 50 years.

 

Love canal

Analysing clusters is difficult, partly because of the  Texas sharpshooter problem. If you start with the cluster, statistical tests ( 'is this just chance?') become almost impossible. But it's important to remember that clusters are not just clusters - they are human lives blighted. And if you are already struggling to come to terms with that, scientists going on about Texas sharpshooters won't help. 

Here is an excellent account of being on the receiving end of scientific rigour / scepticism.  Love Canal incident was, 30 years ago, notorious as an example of toxic waste contamination.

I don't know where truth lies on this occasion but for me two facts stand out: (1) it takes a fairly special kind of idiocy to build an elementary school on land that someone is prepared to sell you for $1 and (2) the $1 land turned out to need a $145m solution.