This piece (on the contributions of medical expenses to bankruptcy) has so e important lessons for epidemiology as well. It's easy to forget that a "risk factor" may be neither a sufficient nor a necessary cause of a clinical endpoint.
For example, smoking is associated with increases in the rates of myocardial infarction. However, some people may smoke heavily for a lifetime and never experience the endpoint (despite living to an old age). Conversely, a smoker who suffers a myocardial infarction may have been fated to have the event anyway. It's even tricky to really estimate the attributable fraction of the risk as confidence intervals are often broad and we are never certain that the conditions for an unbiased estimate (namely, no unmeasured confounders and a clear counter-factual case) are present.
So saying that exposure X contributes to the outcome Y does not make the strong claims of causality that it might seem on first glance. In fact, exposure X can be an real predictor and be increasing but not explain much of the variability in Y. Even worse, there can be issues of meqasurement error. In the financial example, what if a medical expense were to be placed on a credit card? In the smoking example it is clear that estimating pack-years of exposure is likely to be very approximate with typical levels of record-keeping.
So it's not trivial to assess the absolute importance of associations -- even if they are shown to exist in data.
No comments:
Post a Comment