Background: The development and (external) validation of diagnostic and prognostic prediction models is an important area in contemporary medical research. During the past few years, evidence synthesis and meta-analysis of individual participant data (IPD) has become increasingly popular, not only for intervention research but also for improving the development, validation and generalizability of diagnostic and prognostic prediction models. IPD meta- analyses (IPD-MA) provide unique opportunities to improve development and enhance the applicability of prediction models across (sub)populations and settings. There is, however, little guidance on how to conduct an IPD-MA to develop and validate diagnostic and prognostic prediction models, or how to interpret their findings.
Objectives: To describe how IPD-MA of diagnostic and prognostic modeling studies differ from IPD-MA for assessing treatment effects.
Methods: We identify key advantages and challenges in IPD-MA of prediction models. Subsequently, we provide recommendations for the design of such IPD-MA including the selection of relevant studies, Finally, we discuss statistical methods for handling between-study heterogeneity, missing data, and other issues regarding prediction model development and validation. We illustrate all these concepts using various empirical examples across medical disciplines.
Conclusions: The guidance provided in this work may help meta-analysts in prediction modeling research to decide upon appropriate strategies when conducting an IPD-MA, and assist readers, reviewers and practitioners when evaluating the quality of resulting evidence.