Why a Plasma Fit Does Not Locate Absorption
A plasma concentration profile is an indirect record of drug reaching the systemic circulation. It reflects both the timing and amount of systemic input and the body's disposition of that input. It does not, by itself, label the gut segment from which each absorbed molecule originated. Consequently, two different regional absorption patterns can produce the same plasma profile, or profiles so similar that ordinary sampling and measurement error cannot distinguish them.
This is an identifiability issue, not necessarily a defect in the concentration measurements or fitting algorithm. If the observed quantity depends on regional absorption only through the sum of regional systemic inputs, plasma data may constrain that sum while leaving its regional components unresolved. A small fitting error or a precise estimate of total input does not automatically establish how that input was divided among segments.
How Regional Inputs Become a Plasma Profile
Consider a deliberately simple model to expose the issue. Let \(R_i(t)\) be the rate, in mg/h, at which drug absorbed from gut segment \(i\) reaches the systemic circulation. Define the total systemic input rate as \(R_{\mathrm{sys}}(t)=\sum_i R_i(t)\). Assume a well-mixed central compartment of fixed volume \(V\), first-order elimination with rate constant \(k\), and linear, time-invariant disposition. Assume also that the regional inputs have already been expressed as systemic inputs, so differences in presystemic availability are not being separately modelled.
Under these assumptions, the central amount \(A(t)\) satisfies \(dA/dt=R_{\mathrm{sys}}(t)-kA(t)\), with \(A(0)=0\), and plasma concentration is \(C(t)=A(t)/V\). Multiplying the differential equation by the integrating factor \(e^{kt}\), integrating from zero to \(t\), and dividing by \(V\) gives the input-response relation:
The exponential term accounts for elimination between the time \(s\) of systemic entry and the observation time \(t\). The equation shows the central point: plasma is a weighted response to the sum of the regional systemic input rates. If two sets of regional rates have the same sum at every time, they yield the same concentration at every time in this model, even when their segment-by-segment rates differ substantially.
Worked Example: Different Regional Shares, Exactly the Same Profile
Suppose a hypothetical two-segment model has \(V=20\) L and \(k=0.20\) h\(^{-1}\). Compare two regional patterns, with time \(t\) measured in hours:
Each rate is in mg/h. Pattern A assigns 6 mg of total systemic input to segment 1 and 4 mg to segment 2, since \(\int_0^\infty ae^{-t}\,dt=a\) mg. Pattern B assigns 2 mg and 8 mg, respectively. Both patterns conserve the same total systemic amount of 10 mg, but their regional contributions differ.
At every time, however, \(R_{\mathrm{sys}}(t)=10e^{-t}\) mg/h under either pattern. Substituting this into the input-response equation gives, for \(t>0\):
For example, at \(t=1\) h, \(C(1)=0.625(e^{-0.20}-e^{-1})=0.625(0.818730753-0.367879441)=0.281782\) mg/L, or 0.2818 mg/L to four significant figures. This value is exactly the same for Patterns A and B. Even a perfectly measured, densely sampled plasma curve cannot distinguish these patterns in the stated model; the ambiguity is structural.
Similar Profiles Under Sparse or Noisy Observation
Exact equality of total input is not required for practical ambiguity. If regional patterns produce slightly different input timing, their concentration curves may differ by less than the measurement error, or their differences may occur between sampling times. A visually close fit can therefore support several plausible regional explanations. This matters especially when concentration sampling is sparse relative to absorption and transit times.
Worked Example: Two Timing Patterns That Are Hard to Separate
Consider hypothetical instantaneous systemic inputs in the same \(V=20\) L, \(k=0.20\) h\(^{-1}\) model. Pattern A supplies 5 mg at time zero and 5 mg at 2 h. Pattern B supplies 4 mg at time zero and 6 mg at 2 h. These may represent different allocations of absorption across regions and times. After an instantaneous input of amount \(D\) at time \(\tau\), its contribution at \(t\geq\tau\) is \(D e^{-k(t-\tau)}/V\).
At 2 h, including the input occurring at that time, Pattern A predicts \(C_A(2)=\frac{5}{20}e^{-0.20(2)}+\frac{5}{20}=0.25(0.670320046)+0.25=0.417580\) mg/L. Pattern B predicts \(C_B(2)=\frac{4}{20}e^{-0.20(2)}+\frac{6}{20}=0.20(0.670320046)+0.30=0.434064\) mg/L. The difference is \(0.434064-0.417580=0.016484\) mg/L, about 3.95% of Pattern A's prediction.
At 4 h, Pattern A predicts \(C_A(4)=0.25(e^{-0.8}+e^{-0.4})=0.25(0.449328964+0.670320046)=0.279912\) mg/L. Pattern B predicts \(C_B(4)=0.20e^{-0.8}+0.30e^{-0.4}=0.20(0.449328964)+0.30(0.670320046)=0.290962\) mg/L. The difference is \(0.290962-0.279912=0.011050\) mg/L, about 3.95% of Pattern A's prediction. Thus the patterns are not exactly equivalent, but observations with errors of comparable size—or sampling only at a few times—may not reliably distinguish them. The calculation illustrates the possibility, not a universal threshold for declaring two profiles indistinguishable.
What Can Resolve the Ambiguity?
The most direct way to learn about regional absorption is to observe something that depends on location, rather than only on the final systemic mixture. Depending on the experimental setting, informative evidence may include regional sampling of luminal contents or tissue, segment-targeted dosing, a validated regional mass-balance measurement, or an independently measured marker of transit or dosage-form location. These measurements have their own assumptions and errors: for example, an estimate based on drug remaining in a segment must account for drug entering and leaving that segment as well as the sampling procedure.
Independent information can also reduce ambiguity without directly sampling absorption in every segment. Examples include externally measured transit times, independently supported regional permeability or availability, and formulation data that constrain where and when drug is released. Such information helps only to the extent that it is relevant to the uncertain quantities and sufficiently reliable. A constraint imported from another source can narrow possible patterns, but it should not be mistaken for information extracted from plasma concentrations alone.
Worked Example: A Regional Measurement Separates Plasma-Equivalent Models
Return to the first example. Its two hypothetical models predict the same plasma profile, but assign different systemic amounts to segment 1: 6 mg for Pattern A and 2 mg for Pattern B. Suppose an independent, unbiased regional mass-balance method estimates that segment 1 contributed 5.8 mg to systemic absorption, with measurement standard deviation 0.5 mg. Assume the estimate and its uncertainty are appropriate for comparing these predictions.
For Pattern A, the standardized residual is \((5.8-6.0)/0.5=-0.4\). For Pattern B, it is \((5.8-2.0)/0.5=7.6\). The corresponding squared standardized residuals are \(0.16\) and \(57.76\), respectively. The regional measurement is therefore much more compatible with Pattern A than Pattern B under the stated measurement model, even though plasma data alone give identical predictions. This is an illustration of discrimination, not a complete estimation procedure: real analysis would need to represent regional measurement error, bias, and the operational definition of regional absorption.
Diagnostics, Alternatives, and Limits
A good plasma fit is not a diagnostic of regional identifiability. Instead, compare plausible regional patterns while holding the model's other assumptions fixed and examine whether their predicted plasma profiles meaningfully differ over the actual sampling times. If large changes in regional allocations leave predictions almost unchanged, the regional estimates are weakly informed by those data. Examine parameter correlations and uncertainty as supporting diagnostics, but do not rely on one numerical optimizer's standard errors alone: those summaries can be misleading when the model has a flat or nearly flat direction.
The main alternatives differ in what information they use. Plasma-only fitting estimates systemic input or model parameters from central concentrations and is efficient when the question concerns systemic exposure, but may not resolve segmental contributions. Adding regional observations or independent transit, release, or permeability information can distinguish regional explanations, provided the added data are valid and correctly linked to the model. Imposing mechanistic constraints can make estimates determinate, but the result then depends on those assumptions; it is not evidence that plasma alone identified the regions.
The simple derivation also has limits. Nonlinear disposition, time-varying clearance, region-dependent presystemic loss, or differences in the mapping from regional absorption to systemic entry may alter the input-response relationship. More complex models should represent such processes rather than applying the linear equivalence without checking its assumptions. Even where the model is structurally identifiable, sparse sampling, assay error, uncertain dose release, or correlated physiological parameters can make regional estimates practically uninformative. These issues concern the information content of the full model and data, not just the visual closeness of plotted curves.
Decide whether the question concerns total systemic input, regional absorption, or both; these are not interchangeable quantities.
Write how regional inputs enter disposition and determine whether plasma depends on them only through their sum.
Evaluate whether distinct regional patterns change predicted concentrations at the actual sampling times by more than plausible observation error.
Use regional observations or independently supported constraints when the scientific conclusion requires location-specific absorption.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- In the linear model, what condition on two sets of regional input rates guarantees identical plasma profiles at every time?
- Can a perfect, densely sampled plasma curve distinguish the two patterns in Worked Example 1? Why or why not?
- In Worked Example 2, are the timing patterns exactly equivalent? Give the calculated difference at 4 h and explain its practical meaning.
- Name one kind of independent information that could help distinguish regional absorption patterns, and state one assumption or limitation relevant to it.
- Why does a close plasma fit not prove that the estimated regional fractions are correct?
Solutions. (1) Under the stated shared linear disposition model, the regional rates must have the same sum at every time: \(\sum_i R_i(t)\) must be equal for the two patterns. (2) No. Their total input rates are identical at every time, so the convolution with the same disposition is identical at every time. (3) No. At 4 h the calculated difference is approximately 0.011050 mg/L; it may be difficult to distinguish if observation error or sampling limitations are comparable, but it is not exact equality. (4) A regional mass-balance estimate could help, provided its uncertainty and relevant transfers or losses are properly represented. Transit measurements or a validated regional marker are other possibilities, each with its own measurement assumptions. (5) Plasma can constrain the combined systemic input while leaving its regional components unresolved; multiple regional patterns may therefore fit the same observed concentrations.