Tuesday, October 8, 2019

Am I at 'risk'? What might that mean?

What does it mean to be 'at risk' of some disease--say, heart disease?  We tend to think it's something like a probability, something that might but won't necessarily happen.  In a sense, except accidents that may have to do with one's job, say, or sex-specific diseases, we are all at risk for every disease.  That is, we could each, in principle, get the disease.  But the average risk isn't usually what we care about: we want to know our specific risk.  What does that mean and how could it be known, or even estimated?

One sense of the idea is that if I am 'at risk', my probability of experiencing the trait is higher than others'.  Perhaps everyone is at some risk, in which case I am not just 'at risk' but want to know what my particular probability of getting the disease is.  And, of course, one can (should--must) ask what 'risk' means--an absolute probability?  a relative probability?  a probability dependent on some criteria that might be changeable, such as diet?  And extrapolation to the future from what past individuals with my measures-of-whatever experienced?

Let's say that someone with certain blood-profile measures (cholesterol, pressure, say) is at higher risk of heart disease, and presumably because of what that measure measured.  That could mean that, based on some past observation, a fraction f of people with the measured 'risk profile' experienced a heart attack in, say, some 5 year period. That does not mean, though it is often carelessly interpreted as meaning, either that all subjects were once at the same probability, p, of getting the disease, or it could mean that a fraction equal to f of people were doomed, and everyone else immune to the disease.  Or it could be some sort of average among people, the net result observed later after all the events have happened.  How can we tell?

There are lots of ways such kinds of risk might be framed, but they will I think necessarily be specific to various measures, that is, apply to persons characterized by the measures on them as individuals:  male, age 70+, cigar smoker, charcoal steak eater, .....

Seems rather straightforward even if, for someone in the 'risk' category, it may be terrifying.  That's a bit odd, since a risk means only some such people will be victims of the possible event.  Which ones? Why not all?  Is everyone in a given category at equal risk?  How can we know--if we even think about these issues?

Time traveling: How such measures, such p's, are determined and how they're used
Risks are something like the probability or chance that an event will occur in the future, given some conditions in the present.  With disease, it may be the probability of experiencing the disease with, or without, employing some preventive measure--medicine, diet, exercise, and so on.  It all sounds straightforward and, one might say 'scientific' since these p's are presumably estimated from data that includes measures of exposure or genotypes, and of subsequent outcomes.

But probability to whom?  Me?  You? Everyone?  Typically, as just noted, there will be various adjustments depending on age, sex, and perhaps other variables.  These are, of course, approximates based on statistical regression studies.  But were all people with the same such profile at equal risk of getting the disorder?  How could we tell?  And a critically important fact is that these are risks that have already been experienced: they are from past data, often reflecting who did in fact get the disease among those having had similar profiles; yet what we most want to know is our risk that is, for our personal, individual future. And to apply some fraction from the past to the future we have to make some, I think fundamental and often unprovable or indeed often very implausible, assumptions, such as that everyone is at the same risk, given their measured risk factors profile.

The only way such time traveling to the future based on the data we have from the past is useful, is if we assume that what is past is prologue, that is, that causal conditions won't change in any relevant way. But even just based on your own experience in our society, isn't that a rather patently naive, or at best wishful-thinking assumption?

Have lifestyle factors changed perceptibly in your lifetime?  They surely have in mine!  And do you seriously think you can predict lifestyle factors that might be relevant 10, 20....50 years from now?  If not, how reasonable can we assume it is to use past exposures to compute future risks? And, this is not to mention the obvious fact that disease rates or risks have clearly changed, usually substantially, since adequate epidemiological and health records have been kept (and the risks vary among populations, too).  This means that extrapolation to the future is, at the very least, resting on strong but untested (and untestable) assumptions.

Naturally, geneticists, obsessed with DNA sequences that are safe to measure without knowing anything except how to read a sequencer, want to ignore these things, and treat genetic variants as absolute causal entities; but history very clearly shows that the lifestyle factors are by far the more important.  After all, in recent history, even during the lifetimes of many or most readers of this post, risks, as well as exposure to risk factors, for late-onset disease have changed a lot.  We have more dietary advice, more health-care interventions, and so on, than we did decades ago, and there's no sign this won't continue.  We also indulge in plenty of, if not more junk-food binging than the past.

So, when 'risk' is made by regression, on group data, measuring only some of the potential variables (and even that inevitably with some measurement error), it is hubris at best to simply extrapolate that past experience to the future.

We could feel much more positive if these issues were addressed directly, rather than generally being ignored.  This would mean that those routinely computing future risks (based on genotype or whatever) by simply extending past risks to the future....would be confronted with the problem and forced to think more carefully and more circumspectly about biomedical risks and what they mean--and don't mean.  That would be both a scientific and a public service, and far more honest than what we have today (though 'honest' implies the probable counterfactual that the scientists making the predictions understand what they are doing). This applies, of course, to the very notion of 'precision genomic' medicine.

But if you think about it even just a little, it is easy to see that risk is a risky concept, an assumption about the nature of causes and about the degree that we have measured or understand them, and about their degree of context-specificity and about the past's translatability to the future.

Can the prospect of getting some disease be truly probabilistic, in which, say, everyone in a given risk class has the same probability of getting the disease--or is that just a gross oversimplification, an average of past cases, due to lack of knowledge about causation on our part?  This is a key, if perhaps almost unanswerable--perhaps inherently philosophical--question, one that has no 'answer' in the usual sense.  Unfortunately, experts are not paid to express the realities of such doubts, even if they think about them.  We want answers!

But perhaps since the proper answer involves unknowable facts about the future, there are none, and the past is all that we have.  At least we should acknowledge the uncertainties that plague predictions.