The Data to Model to Decision Pipeline

Understand the core PMx workflow that connects observations, models, and decisions.
Tip

What you’ll build today: a clear understanding of how data flows into models and ultimately drives decisions.

Learning Objectives

  • Describe the data → model → decision pipeline
  • Explain how errors propagate across steps
  • Identify where validation is critical

Key Ideas

Pharmacometrics is a sequential reasoning process:

  • Data → Model → Decision

Each step transforms information. That transformation introduces assumptions.

Insight: Because models compress data into parameters, they can hide upstream issues rather than reveal them.

Warning

Most modeling problems are actually data problems in disguise.

Worked Example: Unit Error Cascade

You receive a dataset:

  • Concentrations recorded in ng/mL, but assumed to be mg/L
  • You fit a model → clearance appears 1000× higher than expected
  • You interpret this as rapid elimination

Decision:

  • You lower the dose unnecessarily

What actually happened:

  • A unit error at the data stage propagated into a wrong model and then into a wrong decision

Insight

The earlier the mistake:

  • the harder it is to detect
  • the larger its downstream impact

Strategies

  • Validate data before modeling
  • Check assumptions explicitly
  • Ask what decision is being supported

Common Mistakes

  • Jumping straight to modeling without validating the data
  • Misinterpreting model output as ground truth
  • Ignoring upstream data issues when results look unexpected
  • Being overconfident in model outputs without checking assumptions

Practice Problems

  1. What are the 3 steps?
  2. Where do most errors originate?
  1. Data → Model → Decision
  2. Data stage

Summary

The PMx pipeline is only as strong as its weakest step.

  • Always identify your current stage.
  • Validate before modeling.
  • Ask what decision is being supported.
  • Errors propagate forward.