Masters candidate in Biostatistics, Guanchao Wang, will present:
“Estimating Covariate-Specific Reference Ranges from Meta-Regression”
Plan B Adviser: Haitao Chu
Abstract: A reference range is an important tool in medical screenings and research. An appropriately defined reference range can help differentiate abnormal measurements from a healthy population. Many studies have established methods to estimate reference ranges from single studies. Other studies that reported reference ranges through meta-analysis usually used pooled mean and its confidence interval as the reference value, but these statistics do not capture the full variation across studies. Recently, methods of estimating normal reference ranges through meta-analysis for a new individual were proposed. However, these methods do not allow for estimating covariate-specific reference ranges without fitting separate models. We propose a method that estimates covariate-specific reference ranges through meta-regression for a categorical covariate using the Bayesian random-effects model. We performed a case study to demonstrate the proposed method. Age group-specific reference ranges were estimated for pediatric overnight sleep duration using models with different variance assumptions. The results suggest that different variance assumptions may be preferred based on how precise the within and between-study variances each are estimated. Variance assumptions can be made based on clinical evidence and DIC can be used as a criterion to choose among all candidate models.