Abstract Background Cryptococcal meningitis is a life-threatening infection of the central nervous system. Early fungicidal activity (EFA), which measures how quickly Cryptococcus yeasts are cleared from cerebrospinal fluid (CSF) over time, is a key endpoint in phase II antifungal clinical trials. EFA is commonly estimated via a two-step approach, where simple linear regressions are first fit to each individual’s longitudinal measurements, and the slopes are then aggregated for statistical inference. Alternatively, a linear mixed model can estimate individual and group-level effects in one model. A systematic comparison of these methods is needed to understand and clarify their differences in analyzing EFA data in clinical trials. Methods We conducted a systematic literature review to summarize the usage and reporting of both methods in cryptococcal meningitis clinical trials. We then evaluated their performance and operating characteristics under various simulation scenarios and applied both approaches to data from a phase II trial. Results The two-step approach and mixed model approach yielded discrepant CSF EFA estimates across literature review, simulation experiments, and real data analysis. The two-step approach using linear regression produced steeper estimates and larger estimated treatment effects. Linear mixed model yielded smaller standard errors but estimates biased towards the null, especially when early culture sterility was common. Conclusions The two-step approach is preferred for estimating EFA in cryptococcal meningitis studies, especially when early culture sterility occurs. EFA estimates should not be interpreted or compared across studies without accounting for the underlying statistical models and data pre-processing.
Fuszard et al. (Wed,) studied this question.