Masters candidate in Biostatistics, Monica Iram, will present:
“Comparing Statistical Models for Analysis of Early Fungal Activity Data”
Plan B Adviser: Dr. Biyue Dai
Abstract: Cryptococcal meningitis (CM) is an infection of the brain that causes more than 100,000 HIV-related deaths each year. In therapeutic clinical studies that aim to evaluate treatment efficacy for CM, early fungicidal activity (EFA) during the first 2 weeks of therapy has been used as a standard quantitative measure of the rate of Cryptococcus clearance from cerebrospinal fluid (CSF) collected through a lumbar puncture. Historically, EFA is estimated for a particular subject using a simple linear regression to calculate the slope of the decrease in the log transformed CSF fungal count, and EFA is estimated for a group by taking the average of the slopes from the individual regression lines. However, some recent CM studies have implemented linear mixed models (LMMs) in the estimation of EFA as opposed to the historic simple linear regression (SLR) method. In this work, we compare the use of the simple linear regression method to the use of a linear mixed model for EFA estimation using real data from CM Phase II Clinical Trial ENACT. We then use simulations to empirically assess the performance and operating characteristics of each model under various scenarios, including single-arm studies that compare EFA estimates to a historical threshold, and two-arm studies that seek to compare EFA estimates across concurrent treatment groups. Results from fitting the models to ENACT data found that SLR provides steeper estimates of EFA compared to LMMs. ENACT data analysis also found that the two methods produce very different EFA estimates for individuals whose EFA declines rapidly at the beginning. Simulation results show that the use of LMMs for EFA estimation within a specific arm produces a biased estimate towards the null. However, when comparing the treatment difference across two arms, the LMM has higher statistical power than the SLR approach in scenarios with presence of outliers.
