Doug Lorenz, PhD, Department of Bioinformatics and Biostatistics, University of Louisville

"Development and Internal Validation of a Prediction Rule for Radiographic Pneumonia in the Pediatric ED"

We detail the steps in the process of developing a prediction rule for identifying radiographic pneumonia in children presenting to the emergency department with respiratory symptoms. We provide brief clinical background for the pr0blem and the context for the prospectively-collected data set. The main portion of the presentation will detail the process of developing the statistical model: variable screening by empirical missingness and incidence rules, multiple imputation of missing data, selection of appropriate statistical methodology, structuring of the statistical model including definition of covariates, penalization, and other considerations, model fitting, post-fitting internal validation and calibration, and examination of the diagnostic accuracy of model predictions. The focus will be statistical rather than clinical, in that we will detail the processes associated with each step in the development of the model, rather than the clinical implications of the model and the predictions it provides.

Stay connected TwitterFacebook LinkedIn YouTubeInstagram