Why PTA Curves Shift
A probability-of-target-attainment (PTA) curve summarizes, across candidate doses, the fraction of a population expected to meet a defined exposure target. A curve can move because the target changed, because the simulated exposure distribution changed, or because both changed. These explanations are not interchangeable: the first changes what counts as success, while the second changes the exposures being evaluated.
To interpret a shift, hold one of those components fixed while comparing the other. For a target-driven comparison, use the same simulated exposures and change only the target rule. For an exposure-driven comparison, keep the target rule fixed and compare exposures generated under different simulation choices. This separation makes the direction and meaning of a curve difference easier to establish.
The curve estimates the population probability of meeting the target under the model and simulation assumptions; it is not itself a guarantee about any particular patient. Its interpretation depends on using a consistent exposure definition, regimen, population, and simulation procedure across doses. As with other Monte Carlo estimates, finite replicate counts also introduce sampling error.
Changing a Lower-Bound Threshold
For a fixed dose and a fixed set of simulated exposures, increasing a lower-bound threshold cannot increase PTA. The reason is that the subjects who meet a higher threshold must also meet every lower threshold, while some subjects who meet the lower threshold may fail the higher one.
This is a vertical comparison at each dose: the higher threshold gives the same or a lower PTA. If the aim is to identify the dose at which a prespecified PTA criterion is attained, the higher threshold will generally require a higher dose, provided PTA increases with dose over the relevant range. That qualification matters. The set-inclusion result establishes the direction of the PTA difference at a fixed dose; it does not establish that every dose-response curve is monotone.
When each simulated subject’s exposure is nondecreasing with dose, the fraction meeting a lower-bound target is nondecreasing with dose as well. With nonlinear pharmacokinetics, dose-dependent covariate or regimen choices, or other features that break this subject-level ordering, monotonicity should not be assumed. Inspect the simulated values and the curve rather than inferring monotonicity from the word “dose.”
Worked Example: Raising the Threshold With Exposures Held Fixed
Suppose five hypothetical subjects have simulated AUC values, in mg·h/L, of 8, 10, 11, 13, and 15 at 100 mg every 24 hours. At 200 mg every 24 hours, their values are 11, 13, 15, 17, and 19 mg·h/L. Compare the PTA curves for thresholds of 10 and 12 mg·h/L. These small sets illustrate the calculation; a real simulation would usually use many more replicates.
At 100 mg every 24 hours and \(T=10\) mg·h/L, four of the five values meet the target:
At the same dose and \(T=12\) mg·h/L, only 13 and 15 meet the target:
At 200 mg every 24 hours, all five subjects meet the 10 mg·h/L threshold, while four meet the 12 mg·h/L threshold:
The exposure values, subjects, and dose levels are identical in both comparisons; only the threshold changed. The higher threshold lowers PTA at each dose in this example. If the decision criterion were at least 80% PTA, the 10 mg·h/L target is met at 100 mg every 24 hours, whereas the 12 mg·h/L target is first met at 200 mg every 24 hours among the two doses shown.
Distinguishing Target Changes From Exposure Changes
A change in simulated exposures can also move a curve. For example, a different virtual cohort, covariate distribution, dose history, or model-based exposure prediction can change the distribution of \(E_i(D)\). If the target rule stays the same, any resulting PTA difference is exposure-driven, subject to Monte Carlo variation. To attribute that difference, document which simulation inputs changed and keep the target and calculation rule fixed.
A useful comparison begins with a two-part question: were the target definition and the simulated exposure values identical? If the target changed but the exposures did not, the difference reflects reclassification of the same exposures. If the target stayed fixed but exposures changed, the difference reflects a changed exposure distribution. If both changed, the curve alone cannot isolate their contributions; make separate comparisons in which one component is held fixed.
Worked Example: A Shift Caused by Changed Simulated Exposures
Consider two hypothetical simulations at 100 mg every 24 hours, both evaluated against the same target, AUC at least 10 mg·h/L. Simulation A gives values 8, 10, 11, 13, and 15 mg·h/L. Simulation B gives 7, 9, 10, 12, and 14 mg·h/L. The target rule is unchanged.
For Simulation A, four of five values are at least 10 mg·h/L. For Simulation B, three of five values are at least 10 mg·h/L:
Here the threshold, exposure measure, dose, and counting rule are held constant. The difference is therefore attributable to the changed simulated values, not to a changed target definition. To interpret it further, one would need to know what changed in the simulation and whether that change is justified. The five values alone do not identify the cause of the exposure shift.
If the difference arose only because independent Monte Carlo samples were used, it could instead reflect simulation noise. A convincing comparison should therefore use adequate replicate counts and a reproducible simulation design, and should consider Monte Carlo error when interpreting modest differences.
Moving a Target Window
A target can also be an interval rather than a lower bound. For a window \([L,U]\), a subject meets the target when \(L\leq E_i(D)\leq U\). Unlike raising a lower-bound threshold, shifting the entire window upward has no universally fixed effect on PTA: it depends on where the simulated exposures lie relative to both window boundaries. The upward shift can include more subjects, exclude more subjects, or leave the proportion unchanged.
The same separation principle applies. To compare two windows, hold exposures fixed and change the bounds. To compare two exposure scenarios, hold the window fixed. This tutorial uses windows to explain curve interpretation; the detailed calculation of PTA within a therapeutic window is a separate topic.
Worked Example: A Window Shift Depends on Exposure Locations
At a hypothetical dose of 100 mg every 24 hours, five simulated AUC values are 7, 9, 11, 13, and 15 mg·h/L. Compare the window from 8 to 12 mg·h/L with the upward-shifted window from 11 to 15 mg·h/L. Both endpoints are included.
For the first window, 9 and 11 mg·h/L qualify, so the estimated PTA is \(2/5=40.0\%\). For the shifted window, 11, 13, and 15 mg·h/L qualify, so it is \(3/5=60.0\%\):
The increase occurs because more of these particular values lie in the shifted window. It is not a general consequence of moving a window upward. For a different exposure distribution, the shifted window could have lower PTA. The simulated values are unchanged here, so the difference is target-driven.
How to Compare Curves Reliably
A curve shift is easiest to interpret when comparisons are designed deliberately. First state the exposure measure and its units, the dose and regimen, and the target rule. Then identify whether the comparison changes the target, the exposure-generating simulation, or both. When possible, compute both target definitions on the same simulated values; this makes a target-only comparison direct.
Record the exposure measure, units, threshold or bounds, and whether endpoints are included.
For a target comparison, use identical exposures. For an exposure comparison, use the same target and success rule.
Evaluate the estimated PTA at the same dose levels before interpreting a dose needed to reach a criterion.
Check for unexpected reversals, inconsistent denominators, and differences in exposure values or target coding.
At a fixed target and dose, a difference in the number of qualifying simulated subjects changes estimated PTA by increments of \(1/n\). Consequently, small apparent changes can be caused by finite simulation size. A curve that is jagged or reverses direction may warrant checks of simulation precision, subject alignment across doses, and dose-specific simulation choices. Such a pattern is not proof of an error: the simulated exposure distribution and Monte Carlo variability both matter.
A common alternative is to compare only the dose at which each curve reaches a chosen PTA criterion. That is useful for dose decisions, but it can hide how the curves differ at other doses and whether the crossing is stable. Conversely, comparing PTA at a single dose does not describe the whole curve. Use the summary that matches the question, and report the target alongside it.
Limits of Curve-Shift Interpretation
A PTA curve is conditional on the exposure measure, target, simulated population, dose regimen, and model-based simulation assumptions used to construct it. It does not show that the target is clinically appropriate, nor does it establish that a dose will produce a clinical outcome. A threshold may be uncertain or context-dependent; the curve describes attainment under the specified definition rather than resolving that uncertainty.
Likewise, a visually large shift should not automatically be treated as precise. Finite Monte Carlo samples yield estimated proportions, and the uncertainty in those proportions can matter, particularly near a decision criterion. Comparisons are also difficult to interpret when different denominators, target inclusivity rules, exposure metrics, or regimens are used without being made explicit. The central diagnostic is to verify that the comparison changes only the intended component.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- With the same simulated values at a fixed dose, can raising a lower-bound threshold increase PTA? Explain.
- At one dose, six of ten simulated subjects meet a threshold of 8 mg·h/L and four meet a threshold of 10 mg·h/L. What is the estimated PTA under each target?
- A PTA curve changes after both the target and the virtual cohort are changed. Can the curve difference alone identify which change caused it? What comparison would help?
- Why does shifting an exposure window upward have no guaranteed direction of effect on PTA?
- A curve has lower PTA at every shown dose after the threshold is raised, with the same simulated exposures. Is this evidence that the exposures fell?
Solutions. (1) No. The subjects meeting the higher threshold are a subset of those meeting the lower threshold, so the qualifying count cannot increase. (2) The estimates are \(6/10=60.0\%\) at 8 mg·h/L and \(4/10=40.0\%\) at 10 mg·h/L. (3) No. The changes are confounded in that comparison. Apply both target definitions to the same simulated exposures to isolate the target change, and compare exposure scenarios under one fixed target to assess the simulation change. (4) A shifted window may include more exposures or exclude more exposures, depending on their locations relative to both boundaries. (5) No. The threshold alone can explain the lower PTA. The exposures were held fixed, so the difference reflects changed classification against the target.