Estimating Covariate-Adjusted Incidence Density Ratios for Multiple Time Intervals in Clinical Trials Using Nonparametric Randomization-Based ANCOVA

Abstract

Current methods for estimating incidence density ratios across multiple intervals with covariate/stratification adjustment generally rely on strong parametric assumptions regarding underlying distributions. Some nonparametric methods exist for such settings, but these methods focus on incidence densities for each patient separately and have the limitation of not taking into account the extent to which some patients provide more information than others. Alternatively, the proposed nonparametric methodology captures the mean number of events and the mean person-time for each time interval for each treatment group within each stratum, together with the mean for the vector of covariables and the corresponding covariance matrix. After stratum adjustment, nonparametric randomization-based ANCOVA is applied to produce covariate-adjusted estimates for the log incidence density ratios through forcing the difference in means for covariables to zero. Such covariance adjustment can be invoked either directly for the log incidence density ratios, or for the stratification-adjusted mean number of events and mean person-time, from which the log incidence density ratios are determined. The method is illustrated on exacerbation data from a clinical trial of chronic lung disease.

Keywords: :

• Analysis of covariance
• Linear models equating covariate means
• Lung disease
• Weighted least squares