Immortal time bias: authors' reply.

In a recent issue of AIDS, Galli et al. [1] reported that, among HIV-1 individuals receiving combined antiretroviral therapy (cART), statin use is associated with a lower risk of cancer with an adjusted hazards ratio estimated as 0.45 [95% confidence interval (95% CI) 0.17–0.71]. However, their analysis illustrates the well known immortal time bias [2]. As a matter of fact, they defined baseline as time of cART initiation and compared the risk of cancer among those who will initiate statin at a later time versus those who will never initiate statin. Because there is, by construction, a delay between cART initiation and statin initiation, no cancer can be observed in the ‘statin’ group in the first few years after cART initiation as illustrated in the Figure 1 of their article [1]. This is the most likely explanation of the association they described. A correct analysis would have considered statin exposure as a time-dependent covariate. This type of analysis was precisely developed to overcome the issue of immortal time bias [2]. Moreover, adjusting for updated cholesterol values, which might be influenced by statin exposure, is also an error. It is a pity that the review process did not detect these obvious mistakes.