Fixing Medicare

Last week I was chatting with a friend and colleague, and over wine we were talking of matters deep and complex. In our weighty discussion was something that has been playing on my mind recently, and it provides the basis for this week's entry.

When you're looking at a complex biological system (such as human beings) trying to discern the cause or correct treatment of a disease, there are many, many factors to consider such as is the disease more prevalent in Caucasian women aged 35-45 who are post menopausal, have not had children, who smoke and drink occasionally, and that consume a low-fat diet and are not overweight? In that constraint alone we've identified 9 variables, and we haven't yet delved into an accurate medical history or lifestyle questionnaire, after which, you could easily be looking at 200-300 variables.

In early education, we're taught to think in 2 dimensions, and then 3, but very rarely do we spend much time thinking in more than 3 dimensions. We're taught to interpret and create 2d graphs that have a correlation between the x axis and the y axis, and from this we generally infer outcomes. In the above case of trying to determine disease characteristics, we're looking for a correlation that spans 300 dimensions, and even then, the 'signal' we're looking for is very small compared to the 'noise'. Whilst mathematicians are capable of multi-variant analysis, I think they'd laugh if you asked them to solve this problem, and yet, this is what the pharmaceutical and healthcare industries are trying to do on a daily basis. Since the problem is essentially unsolvable using scientific / mathematical methods what's left? Well, non scientific methods such as 'trial and error'(*), intuition and just plain serendipity(#).

Rather than becoming overwhelmed with the futily of trying to solve this problem, I've come up with a solution: prison inmates. You see, 200-300 variables is simply too many. We need to study a population that lives a similar and well known lifestyle, and thus dramatically reduce the number of variables in the model. Why prison inmates? Their lifestyle is well known, tightly controlled, as is their diet, and there's lots of them, giving a large sample size of people that have a great deal in common. For example I believe($) that prisoners have a well defined exercise period, and since you generally know where they are, it's easy to monitor their vital signs and characteristics, and they (presumably) always turn up to clinic appointments. Want to see if the absence of broccoli from the diet makes a difference? Easy, exclude broccoli from half of your population and monitor the effect. It's still a difficult problem to solve, but at least I've simplified it greatly.

(*) In reality, mostly 'error' which is why so much money is spent on research and development, yielding very little in terms of outcome.
(#) If you don't believe me, look up the history of the development of Penicillin, it was good luck combined with awful science that led to this breakthrough.
($) Don't quote me on this, I've thankfully never spent time in the clink, although I did get detention a few times at school.