Combining PK and PD data; a 20 year perspective

Twenty years ago there were already many publications about population PK/PD analyses, but no research had been done on the best way to combine PK and PD data. A fundamental assumption of PK/PD analyses was that the PK response drives the PD response. Analysing PK and PD simultaneously was considered to be the gold standard, avoiding that potential PK model miss-specification would inflate PD parameter standard error estimates (SEs), but could be computationally difficult. Sequential PK/PD analysis was computationally simpler but estimation error in the PK is ignored and so PD parameter SEs may be underestimated. A body of simulation work was presented looking at 4 different methods to combine PK and PD data.

  1. Simultaneous PK/PD analysis.
  2. Fit PK data. Fix individual PK parameters. Fit PD data.
  3. Fit PK data. Fix population PK parameters. Fit PD data.
  4. Fit PK data. Fix population PK parameters (but retain the PK data in the analysis data file). Fit PD data.

The simulation results of 20 years ago found that methods 1 and 4 performed equally well, and method 2 performed least well.

Moving forward 20 years the fundamental assumption that PK response drives the PD response hasn’t changed; or has it….? Two examples will be presented where there is a three-way interaction between the disease/host, PK and PD, and which renders the question of how to best combine PK and PD data still a relevant discussion even today.