In Reply We appreciate the letter by Figueiredo et al discussing our recent article.1 The letter's authors raised 3 important points.
First, they asked questions about our decision to analyze the 118 patients who received intracranial pressure (ICP) monitors at least 24 hours after admission with the no-monitor group. Observational studies attempting to make causal inferences when the treatment is administered at different times must consider the potential for immortal time bias.2,3 In our study, an ICP monitor could not be received at least 24 hours after admission unless the patient had survived until the monitor was placed. If we had counted those receiving the ICP monitor at least 24 hours after admission among the treated patients, we would have created an immortal time bias in favor of the treatment because some treated patients would have had guaranteed survival to the time of placement. To address this bias, we used a recommended analytic approach2 that mimics a randomized trial. We defined a treatment period, during which patients were placed into treatment groups according to the treatments they received during that period, and a subsequent observation period during which patients were observed for the occurrence of outcomes. Analogous to intent-to-treat analyses in randomized trials, our treatment effect estimates may have been attenuated slightly by the 118 crossovers of initially untreated patients who later received ICP monitors after 24 hours. However, because these 118 patients represent just 5.7% of our 2085 untreated patients, the crossover rate was too small to have appreciably affected our results. The external validity of the article’s main results was strengthened, not compromised as the authors suggested, by this analytic approach.