Sunday, July 8, 2007

or maybe scary statisticians?!

So last time I wrote about three large scale lung cancer screening trials that concluded unanimously x-ray screening for lung cancer does not decrease disease specific mortality. Now, the question becomes what would be the right parameters in making conclusions from a diagnostic clinical trial and what factors shall be taken into account when evaluating the results of these studies?
An article published in 1999 by Dr. Henschke concluded that CT screening for lung cancer is more efficient than chest x-ray since in a small observational study she observed that 85% for cancers are missed using CXR. This motivated the NCI to put together what is considered to be the most expensive RCT, the National Lung Screening Trial (NLST) where they've screened nearly 50,000 people in two arms (CT vs. x-ray) for three years and then they follow up patients for 7-8 years. The final measure of screening successfulness is the mortality rate and the final results are supposed to come out in 2008-2009.
One of the issues associated with setting up an RCT is that it costs so much. You need to have a large population in order to avoid statistical biases and population heterogeneity. So far it is estimated that the NLST has cost $250,000,000, and it seems this figure could be doubled (or maybe it is already doubled). Because the study is really costly, screening takes place in 2-3 years because the lengthier the screening, the more costly the study will be. An important question here is that will a 2-3 year screening necessarily decrease the "mortality"? An answer to this involves complicated statistical modeling, but it has been estimated that if NLST or previous RCTs screened for 18 years, then the decrease in mortality would have been significant between the control and screened arm. Now, no one in the right mind will screen 50,000 people for 18 years because the country may very well go bankrupt. So instead these trials cut on the number of years of screening and no wonder they do not detect a significant decrease in mortality. After you initiate screening, you have to take into account how long it takes for effects to be observable, specially since in the first few years the mortality would not be different because when you start the screening there are already a lot of patients with late stage lung cancer that will die.
More on the other issues later....

No comments: