Tutorials › Pharmacometrics › Log-Linear Analysis of the Terminal Phase
Pharmacokinetics · Noncompartmental Analysis

Log-Linear Analysis of the Terminal Phase

Learn how the terminal portion of a concentration-time profile becomes approximately linear on a logarithmic scale, how to estimate the terminal elimination rate constant and half-life, and why terminal-phase selection matters for pharmacokinetic interpretation.

Intermediate NCA Terminal Phase Half-Life
01 · The big picture

1. What Is Log-Linear Analysis?

Log-linear analysis is a method used to characterize the terminal decline of a drug concentration-time profile. The basic idea is simple: if concentration declines approximately exponentially during the terminal phase, taking the natural logarithm of concentration transforms that exponential decline into an approximately straight line.

For a first-order terminal process:

$$C(t)=C_{\mathrm{term}}e^{-\lambda_z t}$$

Taking the natural logarithm gives:

$$\ln C(t)=\ln C_{\mathrm{term}}-\lambda_z t$$

This has the form of a straight-line regression, \(y=a+bx\), with time as the predictor and \(\ln C\) as the response.

Core idea: exponential decline becomes linear after taking the logarithm. The negative slope of the terminal log-linear regression estimates the terminal elimination rate constant, \(\lambda_z\).
02 · Why the logarithm?

2. Why Analyze Concentration on a Log Scale?

Many pharmacokinetic concentration declines are naturally described by exponential functions. A linear plot of concentration against time therefore produces a curved terminal profile.

On a semi-logarithmic plot, the same exponential decline appears approximately linear:

terminal phase approximately linear on log scale Time ln(C)

When the terminal decline follows an approximately first-order process, plotting ln(concentration) against time produces an approximately straight terminal segment.

The transformation is useful because the slope directly represents a kinetic rate constant:

$$\text{slope}=-\lambda_z$$

Thus, the terminal half-life can be obtained from the fitted slope rather than by visually estimating how long it takes concentration to halve.

03 · Terminal phase

3. What Is the Terminal Phase?

The terminal phase is the portion of the concentration-time profile in which the final observed decline is approximately governed by the terminal disposition process being characterized.

For a simple one-compartment model, there may be one dominant exponential decline. In a multi-compartment model, however, the concentration profile can contain multiple exponential components.

PhaseTypical interpretationLog-linear appearance
Early declineMay reflect distribution, absorption, or other transient processesMay not follow the terminal slope
Intermediate declineMay represent another disposition componentCan have a different slope
Terminal phaseFinal identifiable disposition component used to estimate \(\lambda_z\)Approximately linear if the terminal model is appropriate
Important: the terminal phase is not simply “the last few observations.” The observations selected should support an approximately linear log-concentration versus time relationship and should represent the terminal process of interest.
04 · The terminal regression

4. The Log-Linear Regression Equation

Suppose the terminal observations are \(C_1,\ldots,C_n\), measured at times \(t_1,\ldots,t_n\). Define:

$$y_i=\ln(C_i)$$

The terminal phase can then be modeled as:

$$y_i=\beta_0+\beta_1t_i+\varepsilon_i$$

Under the exponential terminal model:

$$\beta_1=-\lambda_z$$

Therefore:

$$\lambda_z=-\beta_1$$

Because the terminal rate constant must be positive, the estimated slope is normally negative when concentration is declining over time.

The terminal concentration intercept is:

$$C_{\mathrm{term}}=e^{\beta_0}$$

where \(C_{\mathrm{term}}\) is the extrapolated concentration at time zero of the terminal component. It should not automatically be interpreted as the actual observed initial concentration.

05 · Half-life

5. Estimating Terminal Half-Life

Once \(\lambda_z\) has been estimated, the terminal half-life follows directly:

$$t_{1/2,z}=\frac{\ln(2)}{\lambda_z}=\frac{0.693}{\lambda_z}$$

For example, if the terminal regression gives:

$$\lambda_z=0.10\ \mathrm{h}^{-1}$$

then:

$$t_{1/2,z}=\frac{0.693}{0.10}=6.93\ \mathrm{h}$$

The subscript \(z\) emphasizes that this is the terminal-phase half-life, rather than necessarily the half-life of every process occurring earlier in the profile.

06 · Selecting points

6. How Should Terminal-Phase Points Be Selected?

Choosing the observations used in the terminal regression is one of the most important steps in log-linear analysis.

  1. Start with the concentration-time profile. Examine the data on both arithmetic and semi-log scales.
  2. Identify the apparent terminal decline. Look for the final portion that behaves approximately linearly on the log scale.
  3. Use an adequate number of observations. A terminal slope should not be based on an arbitrarily small number of points.
  4. Check the regression. Examine linearity, residual behavior, and the stability of the estimated slope.
  5. Consider sensitivity to point selection. If adding or removing a terminal observation substantially changes \(\lambda_z\), the terminal estimate may be poorly determined.
Do not select points solely because they are the last observations. The final observation can be influential, and a late sample may be below the quantification limit or may not provide reliable information about the terminal process.
07 · Assessing the fit

7. How Do We Know the Terminal Phase Is Approximately Linear?

Several diagnostics can help evaluate whether the selected observations support a terminal log-linear model.

DiagnosticWhat to examine
Visual semi-log plotDoes the selected terminal segment appear approximately straight?
Regression residualsAre there systematic patterns suggesting curvature or model inadequacy?
\(R^2\)Does the selected segment explain most of the variation in log concentration?
Estimated slopeIs \(\lambda_z\) biologically and pharmacokinetically plausible?
Point-selection sensitivityDoes the estimate remain reasonably stable when the terminal window changes?

A high \(R^2\) can be useful, but it should not be treated as the sole criterion for identifying the terminal phase. A small number of points can produce a high \(R^2\) even when the terminal slope is poorly characterized.

Key principle: terminal-phase identification is a pharmacokinetic interpretation problem, not merely a search for the regression with the largest \(R^2\).
08 · Worked example

8. Worked Example: Estimating \(\lambda_z\) and Half-Life

Suppose a drug has the following late concentration measurements after an IV dose:

Time (h)Concentration (mg/L)\(\ln(C)\)
810.002.3026
126.701.9021
164.491.5019
203.011.1019
242.020.7031

Step 1: Transform concentration

Take the natural logarithm of each terminal concentration. The resulting values are approximately linear with time.

Step 2: Estimate the slope

Using the first and last points as an illustration:

$$\text{slope}\approx\frac{0.7031-2.3026}{24-8}=-0.09997\ \mathrm{h}^{-1}$$

Therefore:

$$\lambda_z\approx0.100\ \mathrm{h}^{-1}$$

Step 3: Calculate terminal half-life

$$t_{1/2,z}=\frac{0.693}{0.100}=6.93\ \mathrm{h}$$

Step 4: Interpret the result

The terminal concentration is declining at a rate corresponding to an estimated terminal half-life of approximately 6.9 hours.

Interpretation: after each additional terminal half-life, the concentration is expected to decrease by approximately one-half, assuming the terminal first-order process remains applicable.
09 · Multi-compartment PK

9. Why Terminal Analysis Is Especially Important in Multi-Compartment Models

After an IV bolus in a two-compartment model, the concentration may be represented as the sum of two exponential components:

$$C(t)=Ae^{-\alpha t}+Be^{-\beta t}$$

Typically, the early portion is dominated by the faster component, while the later portion becomes increasingly dominated by the slower component.

At sufficiently late times, the slower component can dominate:

$$C(t)\approx Be^{-\beta t}$$

Taking logarithms then gives an approximately straight terminal line:

$$\ln C(t)\approx\ln B-\beta t$$

Thus, the terminal regression estimates the slower terminal disposition rate constant when the late data adequately represent that component.

Important distinction: in a multi-compartment model, the terminal half-life is not necessarily the same thing as the initial distribution half-life. The concentration profile can contain multiple kinetic phases.
10 · AUC extrapolation

10. Why \(\lambda_z\) Matters for AUC

Terminal-phase estimation is especially important in noncompartmental analysis because \(\lambda_z\) is used to estimate the portion of exposure beyond the last measured concentration.

If the last quantifiable concentration is \(C_{\mathrm{last}}\), a common terminal extrapolation is:

$$AUC_{\mathrm{extra}}=\frac{C_{\mathrm{last}}}{\lambda_z}$$

Thus:

$$AUC_{0-\infty}=AUC_{0-t_{\mathrm{last}}}+\frac{C_{\mathrm{last}}}{\lambda_z}$$

This relationship shows why an unreliable terminal slope can affect more than the reported half-life. It can also influence the estimated total exposure.

Practical consequence: an inaccurate \(\lambda_z\) can lead to an inaccurate terminal AUC extrapolation, particularly when a substantial fraction of total AUC lies beyond the last observed concentration.
11 · Sampling design

11. Why Sampling Matters

The quality of terminal-phase analysis depends strongly on the sampling schedule.

  • Too few late samples: the terminal slope may be poorly estimated.
  • Late samples too close together: concentrations may not change enough to provide strong information about the slope.
  • No sufficiently late samples: the terminal process may not yet be observable.
  • Samples below quantification: unreliable concentrations should not simply be treated as ordinary quantitative observations.
  • Short follow-up: AUC extrapolation and terminal half-life may be particularly sensitive to the assumed terminal behavior.

Terminal-phase sampling should therefore be considered during study design rather than treated solely as an analytical decision after data collection.

12 · Common mistakes

12. Common Mistakes in Log-Linear Terminal Analysis

MistakeWhy it matters
Using the last two points automaticallyA two-point slope is highly sensitive to measurement error and does not establish a reliable terminal phase.
Choosing the largest \(R^2\) without pharmacokinetic reviewStatistical linearity alone does not establish that the selected points represent the true terminal process.
Including distribution-phase pointsThis can produce a slope that mixes distinct kinetic processes.
Including concentrations near or below the quantification limit without appropriate handlingUnreliable measurements can distort the terminal slope.
Assuming every drug has a single exponential terminal declineComplex disposition, nonlinear kinetics, absorption phenomena, and other processes can violate the simple model.
Interpreting \(t_{1/2,z}\) as the half-life of every PK processThe terminal half-life describes the estimated terminal phase, not necessarily earlier distribution or absorption processes.
13 · Practical workflow

13. A Practical Terminal-Phase Workflow

  1. Plot concentration versus time. Inspect the overall PK profile.
  2. Plot the data on a semi-log scale. Look for the terminal portion that appears approximately linear.
  3. Identify candidate terminal observations. Avoid points clearly belonging to earlier distribution or absorption phases.
  4. Fit a linear regression to \(\ln(C)\) versus time.
  5. Estimate the slope. Calculate \(\lambda_z=-\text{slope}\).
  6. Calculate terminal half-life. Use \(t_{1/2,z}=0.693/\lambda_z\).
  7. Evaluate diagnostics. Review the plot, residuals, fit, and point-selection sensitivity.
  8. Assess the impact on AUC extrapolation. Determine whether the terminal estimate materially affects \(AUC_{0-\infty}\).
  9. Document the terminal-point selection. The rationale should be reproducible and scientifically defensible.
14 · Interpretation

14. What the Terminal Slope Does—and Does Not—Tell You

The estimated \(\lambda_z\) summarizes the rate of the terminal log-linear decline under the selected analysis assumptions.

It can be used to calculate:

  • terminal half-life,
  • terminal extrapolation of AUC,
  • the expected fractional decline during the terminal phase, and
  • other NCA quantities that depend on \(\lambda_z\).

However, \(\lambda_z\) does not by itself prove that a particular biological elimination pathway is responsible for the observed terminal decline. It is a parameter describing the selected terminal kinetic behavior.

Modeling principle: a terminal slope is an empirical summary of the late concentration-time profile. Its interpretation depends on whether the selected observations genuinely represent the terminal process.

15. Key Takeaways

  • Log-linear analysis transforms an approximately exponential concentration decline into an approximately straight line.
  • The terminal regression models \(\ln(C)\) as a function of time.
  • The negative regression slope estimates the terminal elimination rate constant, \(\lambda_z\).
  • Terminal half-life is calculated as \(t_{1/2,z}=0.693/\lambda_z\).
  • The terminal phase should not be defined simply as the last few observations.
  • Semi-log plots, residuals, regression diagnostics, and point-selection sensitivity all help evaluate the terminal fit.
  • A high \(R^2\) alone does not establish that the selected observations represent the correct terminal phase.
  • In multi-compartment PK, the terminal phase can represent the slowest observable disposition component after earlier distribution processes.
  • \(\lambda_z\) is used to estimate the extrapolated portion of AUC beyond the last observed concentration.
  • Terminal-phase sampling is an important part of study design because adequate late observations are needed to characterize the decline.
Next step

Where to Go Next

A natural progression is to study Noncompartmental Analysis: AUC and AUMC, followed by Noncompartmental Analysis: MRT and Terminal Half-Life, and then more advanced topics such as partial AUC, sparse sampling, and PK parameter estimation from concentration-time data.

The next tutorial can build directly on the terminal-phase concepts introduced here by showing how AUC, AUMC, MRT, and terminal extrapolation are calculated from observed concentration-time data.

← Back to Pharmacokinetics Tutorials