merlin: An R package for Mixed Effects Regression for Linear, Nonlinear and User-defined models

TL;DR

merlin is a versatile R package that enables joint modeling of complex hierarchical multi-outcome data, including various types of outcomes and random effects, facilitating advanced biological and medical analyses.

Contribution

It introduces a flexible framework for joint modeling of unlimited outcome types and random effects, with user-defined families and comprehensive prediction capabilities.

Findings

Successfully modeled complex patient data post-heart valve replacement.

Demonstrated ability to handle multiple longitudinal and survival outcomes.

Provided tools for individual and population predictions in complex models.

Abstract

The R package merlin performs flexible joint modelling of hierarchical multi-outcome data. Increasingly, multiple longitudinal biomarker measurements, possibly censored time-to-event outcomes and baseline characteristics are available. However, there is limited software that allows all of this information to be incorporated into one model. In this paper, we present merlin which allows for the estimation of models with unlimited numbers of continuous, binary, count and time-to-event outcomes, with unlimited levels of nested random effects. A wide variety of link functions, including the expected value, the gradient and shared random effects, are available in order to link the different outcomes in a biologically plausible way. The accompanying predict.merlin function allows for individual and population level predictions to be made from even the most complex models. There is the option to specify user-defined families, making merlin ideal for methodological research. The flexibility of merlin is illustrated using an example in patients followed up after heart valve replacement, beginning with a linear model, and finishing with a joint multiple longitudinal and competing risks survival model.