Development of a Mechanistic Distribution Model for Predicting Drug Partitioning into Human Milk

Limited clinical data on drug safety during lactation complicate benefit-risk assessment for breastfeeding women who require pharmacotherapy. Mechanistic prediction of the milk-to-plasma ratio within physiologically-based pharmacokinetic frameworks offers a complementary and efficient approach to traditional clinical lactation studies. We aimed to develop a mechanistic milk partitioning model for milk-to-plasma ratio prediction that explicitly accounts for protein binding, pH-dependent ionization, lipid partitioning, electrostatic interactions with acidic phospholipids present in milk fat globule membranes, and membrane potential effects on drug disposition.

Model performance was compared with both the phase distribution model and its log-transformed variant using a verification dataset of 107 observations from 57 compounds with diverse physicochemical properties.

For acids, moderate-to-strong bases and ampholytes, the milk partitioning model demonstrated the best performance (absolute average fold error (AAFE): 1.99, 1.70 and 3.34), outperforming both comparator models. For weak bases, the milk partitioning model achieved comparable accuracy to the phase distribution model and its log-transformed variant (AAFE: 1.63 vs. 1.62 and 1.64) while showing lower bias (average fold error (AFE): 0.92 vs. 0.72 and 0.73). Finally, for neutral compounds, the phase distribution model showed superior performance to the milk partitioning model (AAFE: 2.35 vs. 2.60).

When integrated within a physiologically-based pharmacokinetic framework, the milk partitioning model provides a mechanistically robust and practical tool to support quantitative assessment of drug transfer into human milk and inform safer dosing strategies for breastfeeding populations.