Objectives : This workshop will introduce participants to the statistical methods than can be applied for meta-analysis of previously published prediction models. We will discuss opportunities and challenges of combining existing literature models into a new, single, model by using a minimal set of patient-level data. Subsequently, we will introduce the required statistical methodology and describe common software packages, using real example data.
Description: Many risk prediction models are commonly developed and validated for predicting the future occurrence of a particular outcome (prognostic prediction models) but also to predict the presence of a certain disease (diagnostic prediction models). Prediction models aim to provide absolute probabilities of a certain outcome or disease in an individual given his/her observed medical history, physical examination, and additional results from, e.g., imaging test or biomarker assays. Prediction models have become abundant in the literature. Consequently, for the same outcome or target population, commonly various prediction models have been developed, even over 100 models for predicting outcome after traumatic brain injury, over 60 for breast cancer and over 40 for diabetes type 2, to mention just a view. Rather than developing the next prediction model for a particular outcome or target population, systematic reviews of risk prediction models have become timely. The question arises whether and how previously published prediction models should and can be combined in a meta-analytical manner. Recently, innovative methods have been developed to meta-analyse (combine) previously published prediction models, given that particular information from these studies is available.
In this workshop we will discuss why and how meta-analysis of risk prediction research is opportune. We describe how previously published prediction models can appropriately be combined for this aim. We will also demonstrate how researchers can accommodate for heterogeneity across study populations. Finally, we will illustrate how researchers should interpret the achieved meta-analytical model performance and provide strategies for improving upon its generalizability and applicability.