1.1 Reference Values

Much of our understanding of context is driven by reference values. A doctor knows whether your cholesterol is high or low based on large population studies of other people. Every health study is essentially just a way to calculate reference values: of the n people exposed to this treatment, some fraction will improve. If that fraction is large enough, we say the treatment works. If not, the treatment doesn’t work.

So the real question in any medical condition is: what is the reference value? What is the standard by which I am judging my current condition?

For many (most) situations, the reference values have been pre-computed based the medical community’s long experience treating patients like you. We know that X% of people with your type of cancer respond well to this drug. We know that Y% of people who smoke develop this disease. And on and on.

But for some situations – like data from microbiome tests – there is no reference value. Nobody knows what a “healthy” microbiome looks like. We need more data before we can say definitively that such-and-such abundance levels are “healthy” or “unhealthy”.

In other cases, there are reference values for the general population, but not necessarily for you. The average height of a 3-year-old girl, for example, is based on data from umpteen thousands of 3-year-olds, but what about among 3-year-olds of your ethnic group, or your family, or people of your socio-economic class, or those in your neighborhood? Whether to consider your 3-year-old for special treatment depends entirely on which reference group you are using.

How can we get those reference values?

In other cases, a treatment may be too new, or too crazy, for there to be reference values. A terminal cancer patient who tries an experimental treatment, for example, is living in a world of unknown reference values. Importantly, after they try the treatment, they become one of the reference values. And that’s great! we now have a reference value for that treatment — but only if somebody bothers to record it. Often that data simply falls on the floor with nobody to catch it.