Tuesday, August 06, 2013

What could possilby go wrong with meta-analysis and guidelines (think pre-op beta blockers)

Epidemiology 101 describes reasons for a correlation, namely :" causation,bias,confounding and chance".But fraud should also be on the list. Reports indicate that fraud played a major role in a Dutch study which in turn determined the outcome of at least one meta-analysis and from that a major recommendation for pre-op beta blockers in non cardiac patients.And now it seems that rather than prevent peri-operative deaths the beta-blockers might have lead to more deaths.

 The European Society of Cardiology issued a strong recommendation for the use of peri-operative use of beta-blockers in 2009. Their analysis that lead to that recommendation was apparently heavily influenced by the DECREASE trial that showed a significant decrease in perioperative heart attacks in the treatment group. On the other hand ,the POISE trails showed that the control group had fewer deaths. When the two were combined and included in the meta-analysis and sprinkled with magic statistical fairy dust benefits were shown to be greater than the risks. When the fraud issue was raised and another Meta-analysis was done excluding the DECREASE data the risks were greater than benefits and more deaths occurred in the treatment arm.

See here for details and for more links.

As tragic as however- many- deaths occurred as a result of these guidelines there is reason for great optimism moving forward. Surely we will not need to worry about this sort of thing happening with the advent and proliferation of Accountable Care Organizations (ACOs) which will be catalyzed by the great crony capitalism victory social justice generating Accountable Care Act.We look forward to have wise "leaders with ideas " leading the way  in good Don Berwicken fashion who will be immune to the multitude of Kahneman-Tversky type cognitive biases to which the hapless individual medical practitioners and their selfish patients are so susceptible.

What is  the big deal here anyway-surely fraud is rare in medical studies.I hope so too but  there are other reasons to be wary of meta-analysis and should sweeping guidelines be based on meta-analysis that are driven by one study. Agent Mulner believed the "truth is out there somewhere"-maybe  but it is elusive and premature conclusions that are magnified by being enshrined in guidelines-particularly those that fortified in a P4P setting-can do more than a little harm.When the wisdom of the day is 180 degrees from the wisdom yesterday you can hear the plaintiff attorney asking :"Doctor, were you wrong then or are you wrong now ?"

1 comment:

Michel Accad said...

Hi James,

I wrote a little blurb on this a while back. The late Alvan Feinstein had aptly called the meta-analysis "statistical alchemy."

Michel

http://alertandoriented.com/the-statistical-alchemy-of-meta-analyses/