Conditional versus unconditional covariate effects in pharmacometric models: implications for interpretation, communication, and reporting

As model-informed drug development (MIDD) becomes standard, it is central that complex statistical dependencies do not lead to clinical misinterpretation.

A new publication by E. Niclas Jonsson, Siv Jönsson, Emma Hansson, and Joakim Nyberg, and medical writing by Viviana Moroso, explore the distinction between conditional and unconditional covariate effects, an important nuance when models include correlated predictors like body weight and age. This work provides a framework to improve the clarity of dosing recommendations, ensuring they are robust and actionable for prescribers across diverse patient populations.

• Conditional covariate effects provide important mechanistic information but are sensitive to model context and correlations. 
• Interpreting conditional effects in isolation can lead to biased dose predictions and incorrect conclusions about dosing needs. 
• Unconditional covariate effects are derived from simpler models and do not depend on other covariates in the model. 
• Despite their simplicity, unconditional covariate effects can provide robust dosing information with less risk of misinterpretation. 

The authors recommend reporting unconditional effects for communication tools like forest plots and drug labels to maximize interpretability, while using the full conditional model for mechanistic understanding and extrapolations.