Concentration-Guided Titration With Sparse Measurements

Why the Sample Time Is Part of the Result

A measured concentration is not a complete description of exposure unless its timing relative to the dose is known. Within a dosing interval, concentration may be falling, rising, or near a peak or trough. Thus, the same numerical result can imply different dose changes depending on when the sample was collected. Concentration-guided titration needs both a target and a rule for comparing a measurement with that target at a defined time.

This tutorial develops a simple rule for a fixed regimen: project a sparse measurement to a specified reference time, compare that projected concentration with the target at that time, and scale the dose in proportion to the discrepancy. The approach is useful only when its pharmacokinetic assumptions are reasonably appropriate. It does not replace clinical judgment or establish a drug-specific target.

Model and Dose-Update Rule

Consider a fixed dosing interval and a one-compartment model with first-order elimination. For a post-dose period in which there is no continuing drug input, concentration declines exponentially. Let \(C(t)\) be concentration at time \(t\) after the dose and \(k\) the elimination rate constant, in inverse hours. Then \(C(t_2)=C(t_1)e^{-k(t_2-t_1)}\). For a bolus dose at steady state, the same exponential decline applies within the interval after the dose, provided the one-compartment assumption is suitable.

Suppose the desired concentration is \(C_{\text{target}}\) at reference time \(t_{\text{ref}}\), and a concentration \(C_{\text{obs}}\) is measured at time \(t_{\text{obs}}\). If the sample and reference time lie in the modeled elimination phase, the projected concentration is:

When \(t_{\text{ref}}\) is later than the sample, this projects forward and generally gives a lower concentration. If the reference time is earlier, the negative time difference projects backward and gives a higher concentration under the same model. The timing convention must be consistent: times are measured from the same dose, and the target is defined at the same point in the dosing interval.

With two samples in the same elimination phase, an estimate of \(k\) follows by taking logarithms of the exponential model. For \(t_2>t_1\):

This calculation requires positive concentrations and a decline consistent with elimination. In practice, the estimate may be noisy, especially when the samples are close together or their measurement error is large. With one sample, \(k\) must come from other suitable information, such as a previously estimated individual value or a justified population estimate.

For a fixed regimen under linear pharmacokinetics, concentration at a particular time scales in proportion to dose. If the current dose is \(D_{\text{old}}\), the projected reference-time concentration is \(\widehat{C}_{\text{ref}}\), and the target is \(C_{\text{target}}\), the unbounded proportional update is:

This rule assumes that the dosing interval, formulation, administration conditions, and other determinants of exposure remain fixed, and that concentration is proportional to dose over the range considered. It calculates a candidate dose; any applicable dose limits or clinical safeguards must still be applied.

Worked Calculations

Why the Rule Works—and When It Does Not

The proportional update follows directly from the assumed dose-concentration relationship. At the reference time, let the concentration under a dose \(D\) be \(C_{\text{ref}}=aD\), where \(a\) is a constant for the fixed regimen and patient conditions. Under the current dose, the model-based estimate is \(\widehat{C}_{\text{ref}}=aD_{\text{old}}\). The dose that would produce the target satisfies \(C_{\text{target}}=aD_{\text{new}}\). Dividing these equations eliminates the unknown \(a\):

The argument is exact only under the assumed proportional relationship. If clearance or bioavailability changes with dose, the relationship may not be linear. If the dose interval changes, or if the patient has not reached the modeled state, the concentration at a given time may not scale in the simple way assumed. A model-informed projection is therefore a conditional calculation, not a guarantee that the next measured value will equal the target.

A sparse rule is most credible when sample times are known precisely, the measurements are from a comparable regimen, and the samples occur in a pharmacokinetic phase described by the model. For example, an early oral sample may still be rising because absorption is continuing. Applying an elimination-only equation to that sample can project the wrong direction or magnitude. Similarly, a changing dose history, missed doses, a recent infusion, or a substantial change in organ function can undermine the assumed relationship.

Useful diagnostics include checking whether the later of two samples is lower than the earlier one in the assumed elimination phase, whether the estimated \(k\) is plausible for the drug and patient, and whether the projected concentration is highly sensitive to a small change in sample time or \(k\). A negative estimate of \(k\) from declining-phase samples, or a dramatic estimate based on nearly equal concentrations, is a warning to review the data and model rather than blindly apply the update. With only one or two noisy measurements, uncertainty in the projection can be substantial even when the arithmetic is correct.

This calculation differs from comparing a measured concentration directly with a target at the same sampling time. Direct comparison is appropriate only when measurement and target refer to the same time and regimen phase. It also differs from model-based estimation using a richer concentration-time data set: a fuller model may represent absorption, distribution, accumulation, covariates, and measurement error, but requires more information and modeling assumptions. The sparse proportional rule is transparent and easy to implement, yet less able to distinguish true pharmacokinetic change from sampling or assay variability.

Common Pitfall and Practical Checks

Before using a calculated update, check the following:

• Is the target explicitly tied to a sampling time and dosing interval?
• Were the sample time, dose time, and recent dosing history recorded reliably?
• Does the assumed elimination model apply between the sample and reference time?
• Is \(k\) estimated from suitable data or supported by an appropriate prior estimate?
• Is dose proportionality plausible, and are regimen features being held fixed?
• Would measurement or model uncertainty make a direct adjustment unsafe or unstable?

If these checks fail, repeating a correctly timed measurement or using a more suitable pharmacokinetic model may be preferable to acting on a fragile projection. The rule’s strength is that it makes timing and assumptions explicit; its limitation is that sparse data cannot reliably correct a misspecified model.