A0760
Title: Comparison of dose-response meta-analytic models using empirical and simulation studies
Authors: Joseph Beyene - McMaster University (Canada) [presenting]
Abstract: Dose-response relationship studies are common in clinical as well as epidemiological studies and are critical for understanding how varying levels of exposure impact the risk of an outcome. Using meta-analytic approaches to synthesize dose-response data from multiple studies presents several challenges. One-stage and two-stage dose-response meta-analysis (DRMA) methods are commonly employed. The one-stage method employs a linear mixed model, while the two-stage method involves estimating key model parameters within each study and synthesizing them across studies. The underlying dose-response relationship can be linear or nonlinear, depending on the specific exposure-outcome pairing. Various exposure-outcome relationships are investigated using the one-stage DRMA method, employing linear, quadratic polynomial, and restricted cubic spline (RCS) models. Knot selection in RCS models, model selection using different functional forms, and the influence of outlying studies are investigated in detail. Various models are assessed using an empirical study based on a large collection of published DRMA datasets. With model parameter selection informed by the empirical data, extensive simulations are designed and implemented to evaluate comparative performance across several realistic scenarios.
