Friday, October 06, 2006

More viral load analysis

A recent discussion online made me perform a little back-of-the-envelope math on viral loads. As shown in the recent paper discussed below, on an individual basis the predictive value of viral load on any one individual's rate of CD4 T cell loss is poor. In fact the authors found it responsible for about 4-6% of the variation. But they also supported the original data showing a good correlation between subgroups of viral load and rate of CD4 T cell loss.

But how good?

Well, I performed the same kind of analysis on the subgroup data.

For those with viral loads under 500 I used a value of 250, for those between 500 and and 1000 I took 750, etc. I graphed the log10 value of these viral loads against the average CD4 T cell loss given in the paper for these subgroups.

The result?

A whopping 93.5% link between the subgroup viral load and rate of CD4 T cell loss. Just a tad better than the 4-6% for the individual data points.

The upshot of this? Well, clearly the paper is still saying that at the individual level the predictive value of viral load is very limited - but the SAME ANALYSIS performed on the bigger picture clearly shows why clinicians and scientists the world over concluded that viral load was a good predictor of T cell decline.

It should also but a sock in the mouth of any dissidents who say otherwise.

I'll freely admit that this analysis is far from the comprehensive dataset looked at in the paper by Rodriguez et al, but it IS the same data ;-) Regardless of its limitations, it's pretty striking.