Journal clubpopulation PK, methods
Learning one drug from a thousand scattered blood samples
Sheiner LB, Rosenberg B, Marathe VV. Estimation of population characteristics of pharmacokinetic parameters from routine clinical data. J Pharmacokinet Biopharm. 1977;5(5):445–479.
The problem
Classic pharmacokinetic studies take ten or more blood samples from each of a dozen healthy volunteers, then fit a curve to each person. That works beautifully for the people who least need the drug.
The patients who do need it are older, sicker, on other drugs, with failing kidneys. From them, the clinic collects one or two levels at whatever time the ward round allowed. You cannot fit a curve to two points per person, so for decades these data were thrown away.
The core idea
Stop treating each patient as a separate puzzle. Assume every patient’s clearance is drawn from one population: a typical value plus a personal deviation. Then estimate the population directly, pooling every scattered sample at once.
My take
This is the paper that made routine clinical data scientifically useful: pool the patients, model the variability, stop discarding the sick.