Meta-learning studies how learning systems can increase in efficiency through experience. Its purpose is to understand how learning itself can become flexible according to the tasks or contexts under study. These properties appear particularly interesting in safety field and in Dynamic Risk Management (DRM) process. Indeed, in such a process, the meta-learning strategies can support safety managers in the recommendation of models estimating a particular risk in the presence of various types of uncertainty due to lack of knowledge, omissions, incomplete analysis, and/or simplifying assumptions. The aim of this paper is to provide a narrative review about available meta-learning approaches in DRM process and, particularly, applied for assessing uncertainty in models able to estimate safety risks. No documents are available in the literature that deal with meta-learning approaches for safety topics, for DRM process, or for uncertainty assessment in models for estimating risks. However, the variety of existing studies related to meta-learning principles allows to address the development of a framework applicable in a general DRM process to reduce uncertainty. We discuss some main elements of a novel and preliminary metalearning framework that should help safety managers to select a subset of models assessing a risk that assures desired uncertainty conditions. Particular emphasis should be given to the identification and definition of relevant metaattributes, such as data-based, domain-based, uncertainty-based, and sensitivity-based meta-features.