Prospective evaluation of a model for the prediction of milk:plasma drug concentrations from physicochemical characteristics.

1. Milk:plasma (M/P) drug concentration ratios predicted by a model utilizing pKa, plasma protein binding and octanol:water partition coefficients have been compared with actual M/P values for 10 basic drugs. 2. There was a close relationship between predicted and observed M/P ratios with a coefficient of determination r2 of 0.97. However, there was a proportional error. 3. The data were transformed by taking logs of predicted and observed (M/P + 1) values. Regression analysis resulted in an r2 of 0.95, an intercept on the Y-axis not significantly different from zero and a slope not significantly different from one. 4. The 95% confidence interval around a single prediction revealed an error between 150% for the lowest and 23% for the highest M/P ratios. The error is therefore lowest for the drugs likely to have the greatest transfer into milk. 5. There was no significant bias in the predictions. 6. The model was refined by multiple linear regression analysis utilising the observed M/P ratios for the 10 basic drugs in addition to those of the original drugs. The revised equation resulted in an improvement in the explained variance. 7. Protein binding was the most important single predictor. 8. The results confirm that M/P ratios for basic drugs can be predicted accurately from their physicochemical characteristics.