From Data to Modeling Dataset

Consolidate exploratory analysis and prepare a reviewed dataset for structural PK modeling.
Tip

Big picture: Before building equations, we should pause and confirm that the dataset has been reviewed and that the next step is truly modeling—not more cleaning.

Learning Objectives

By the end of this lesson, you will be able to:

  • Consolidate dataset understanding.
  • Review event records and observations.
  • Summarize exploratory findings.
  • Create a reusable modeling object.
  • Identify readiness for structural modeling.

Key Ideas

  • EDA should lead to decisions.
  • Modeling datasets should be reviewed before estimation.
  • Visualization generates hypotheses.
  • Modeling starts after understanding the data.

Setup

library(tidyverse)
library(nlmixr2data)

data("theo_sd", package = "nlmixr2data")

theo <- theo_sd

obs <- theo %>%
  filter(EVID == 0)

dose <- theo %>%
  filter(EVID != 0)

What We Learned from the Dataset

At this point, we already know:

  • the dataset uses event records
  • observations and doses coexist
  • concentrations follow expected PK behavior
  • subjects show variability
  • weight varies across subjects
  • event ordering appears reasonable

The goal now is to consolidate—not rediscover.

Worked Example 1: Build a Modeling Summary

Create a compact dataset summary.

dataset_summary <-
tibble(
  n_subjects =
    n_distinct(theo$ID),

  n_rows =
    nrow(theo),

  n_observations =
    nrow(obs),

  n_dose_rows =
    nrow(dose),

  min_time =
    min(obs$TIME),

  max_time =
    max(obs$TIME)
)

dataset_summary
# A tibble: 1 × 6
  n_subjects n_rows n_observations n_dose_rows min_time max_time
       <int>  <int>          <int>       <int>    <dbl>    <dbl>
1         12    144            132          12        0     24.6

Worked Example 2: Build a Modeling Object

Create a reviewed modeling object.

theo_model <-
theo %>%
  arrange(ID, TIME)

head(theo_model)
  ID TIME    DV     AMT EVID CMT   WT
1  1 0.00  0.00 319.992  101   1 79.6
2  1 0.00  0.74   0.000    0   2 79.6
3  1 0.25  2.84   0.000    0   2 79.6
4  1 0.57  6.57   0.000    0   2 79.6
5  1 1.12 10.50   0.000    0   2 79.6
6  1 2.02  9.66   0.000    0   2 79.6

Confirm dimensions.

dim(theo_model)
[1] 144   7

This object will serve as the working dataset in later modules.

No rows were removed.

No transformations were applied.

Worked Example 3: Final Concentration Profile

Generate one final subject-level plot.

ggplot(
  obs, aes(TIME, DV, group = ID)) +
    geom_line(alpha = 0.35) +
    labs(
        title = "Reviewed Concentration–Time Profiles",
        x = "Time",
        y = "Concentration"
    )

Ask:

  • Does concentration behave reasonably?
  • Is variability visible?
  • Are observations plausible?

If yes, proceed.

Worked Example 4: Readiness Checklist

Create a simple readiness table.

readiness <-
tibble(
  check = c(
    "Events reviewed",
    "Observations inspected",
    "Profiles visualized",
    "Dose records reviewed",
    "Working dataset created"
  ),
  status = "Complete"
)

readiness
# A tibble: 5 × 2
  check                   status  
  <chr>                   <chr>   
1 Events reviewed         Complete
2 Observations inspected  Complete
3 Profiles visualized     Complete
4 Dose records reviewed   Complete
5 Working dataset created Complete

Transition to Structural Modeling

Until now we have focused on:

Data
↓
QC
↓
Visualization

The next module introduces:

Data
↓
Structural PK Model
↓
Parameters
↓
Predictions

This is where equations begin.

Strategies

  • Review findings.
  • Freeze assumptions.
  • Move forward intentionally.

Common Mistakes

  • Repeating QC endlessly.
  • Changing datasets during estimation.
  • Losing reproducibility.

Practice Problems

  1. Recreate the summary table.
  2. Create a reviewed modeling object.
  3. Write three observations about the concentration profiles.
  4. Explain why modeling has not started yet.

Reuse the code blocks above.

Focus on interpretation and readiness—not additional cleaning.

Summary

  • You reviewed theo_sd.
  • You explored event records.
  • You visualized concentration profiles.
  • You created a modeling object.
  • You are ready for structural PK modeling.
  • Understand first.
  • Model second.
  • Keep datasets stable.
  • Use plots to ask questions.