“…For this reason, the use of clustered data during prediction model development and its subsequent validation offers a critical opportunity to inspect whether this heterogeneity would actually be a concern when the model would be implemented in clinical practice. [226,116,221,15,26,23,20,21,227,25] However, actually resolving the presence of heterogeneity (and thus ensuring model predictions are accurate for all clusters) remains a difficult challenge for which limited guidance is available. [228] For this reason, we here explore an alternative approach that aims to reduce this heterogeneity and minimize the need for estimating setting-specific model parameters, to thereby improve its generalizability.…”