Monday, June 27, 2011

The Future Burden Of Disease From Obesity May Be Underestimated (plus, how to sound like a very smart statistician)

No doc office is complete without this
In yet another instance of the medical literature being an endless font of obscure statistical jargon, check out this article in that wacky policy journal, Health Affairs, and say hello to the term "three dimensional forecasting."

Against all odds, the intrepid Disease Management Care Blog will attempt to wrestle this tricky theoretical tomfoolery to the ground, examine the implications for the obesity epidemic and, best of all, once again demonstrate how obscure epidemiologic phrasing can be used against foes and fools alike.

The Past As Prologue To The Future

According to authors Eric Reither, Jay Olshansky and Yang Yang, the accuracy of of "two dimensional" statistical trending is blunted by an underlying assumption that the trajectory of past data trends will continue into the future.  One example is the unrelenting upward march of national health care costs (which explains our national interest in "bending the trend").


In this article, the authors apply their more accurate "three dimensional" approach to assessing the future national risks that are associated with obesity.  By way of background, the DMCB found this classic two dimensional graph that projects a 70% rate of obesity in the U.S. by 2020.  When the association between obesity and premature heart disease is considered, it's easy to conclude that that there could be a significant shortening of life expectancy in the coming decades.

Using The 3rd Dimension To Check Out The Future

Reither et al says it's much worse than that.  Their "three dimensional" methodology mathematically includes the burden of additional risk factors (for example, the prevalence of childhood obesity) that are present today that, in turn, could act as yet-unseen or "latent" drivers of more risk in the future.  This is a step up from the basic "two dimensional" modeling that bakes in past assumptions about improving overall life expectancy and ignores today's simmering time-bombs built of pizza, hot dogs and pop.


Their conclusion?  They applied their modeling by testing it on past heart disease risk factor data and compared their predictions of the incidence of coronary artery disease with today's real incidence.  They found a surprisingly good fit.  Based on the fit of these data, they warn that the U.S. public health community's reliance on old statistical approaches is probably underestimating what we should be planning for in the future.  They recommend that "three dimensional modeling" be more broadly used and that we start by looking at obesity.  Based on some other research, they point out that we may be underestimating the projected decrease in life expectancy from obesity by as much as much as five years.

Enter The Institute Of Medicine

Speaking of which, the IOM has just released a report that lists interventions that have been shown to blunt the incidence of childhood obesity.  Children should be screened (two measures are weight for length or BMI at the 85th percentile), be encouraged to increase their physical activity (15 minutes per hour in day care for example; community outdoor recreation areas), engage in healthy eating habits (for example, attention to portion sizes) and be protected from predatory food industry marketing.  You can find lots of good stuff here; no obesity prevention program is compete without the IOM recommendatons.

3rd Dimension Verbal Swordsmanship

Armed with the concept that standard projections based on linear-statistical trending may be unequal to the task of assessing future risk, DMCB readers can now challenge know-it-all speakers, faux-expert consultants and tiresome academics by stopping them dead in their tracks - as early as PowerPoint slide 2.  Anytime you see a graph that has a dotted line extending up and into the future, you can raise your hand and ask if the plot is based on shabby two-dimensional modeling.  No boring meeting is complete without the threat of this kind of showstopper:

Er, excuse me Doctor Pintminded, but you seem to assume that the future prevalence of diabetes will be [insert number here] but does that account for present day risk factors that are typically included in a three dimensional analytic estimate of future burdens of disease?

Happy Hunting!

Image from Wikipedia

1 comment:

Anonymous said...

From the abstract, it looks like our esteemed colleagues in sociology and public health have re-''discovered'' the same autoregressive/autoregressive distributed lag models for time series forecasting that are taught in introductory econometrics courses to undergraduate economics majors across the country.

Of course, if this is novel for them, I have to wonder how well they did the analysis. As an aside, I'd also hazard that contrary to their assertions, the issues facing state pensions plans have their genesis in much more than merely bad forecasts of life expectancy.


Perhaps they can take solace in the fact that they weren't the ones who rediscovered calculus in their medical scholarship, like this poor fellow did in the 1990s: