Why a Concentration Profile Can Have a Shoulder
A concentration-time profile that rises, flattens, and then rises again—or declines more slowly over an interval than the fitted model predicts—may indicate that drug is entering the systemic circulation over more than one characteristic timescale. A shoulder is not, by itself, proof of two absorption processes. It is a pattern to investigate after checking the disposition model, data quality, and sampling times. The earlier material on reading residual patterns and delayed peaks provides context for locating such discrepancies; here the focus is on alternative shapes for the absorption input.
Two useful alternatives are a mixture of two first-order absorption components and a chain of transit compartments. The first represents two fractions with distinct first-order input rates. The second represents a distribution of transit times generated by successive first-order steps. Both can spread input over time, but they impose different shapes and interpretations. They are structural alternatives, not automatic explanations for a shoulder.
Two First-Order Input Components
First consider a single oral dose \(D\), bioavailability \(F\), a one-compartment disposition model with volume \(V\) and elimination rate \(k_e\), and no absorption lag. Assume linear disposition, complete transfer out of each absorption depot, and a fixed fraction \(q\) assigned to the first depot. The absorption input rate is
Each term integrates to its assigned dose fraction: for example, \(\int_0^\infty qFDk_{a1}e^{-k_{a1}t}\,dt=qFD\). Thus the total systemic amount entering over all time is \(FD\). With linear one-compartment elimination, the concentration is the accumulated input remaining after elimination:
This expression follows by adding the concentration contributions from each depot. For one component with fraction \(w\) and rate \(k_a\), an amount \(wFDk_ae^{-k_at}\) enters per unit time. An amount entering at time \(u\) contributes \(wFDk_ae^{-k_au}e^{-k_e(t-u)}/V\) at time \(t\). Integrating from \(u=0\) to \(t\) gives the corresponding term in the equation. If an absorption rate equals \(k_e\), the apparent zero denominator is handled by its limit: \(k_at e^{-k_at}\) replaces \(k_a(e^{-k_et}-e^{-k_at})/(k_a-k_e)\).
Worked Example: Calculating a Shoulder From Two Input Rates
Suppose a hypothetical dose has \(FD/V=100\) mg/L, \(q=0.70\), \(k_{a1}=2.0\) h\(^{-1}\), \(k_{a2}=0.25\) h\(^{-1}\), and \(k_e=0.10\) h\(^{-1}\). Assume the one-compartment linear disposition and complete depot transfer described above. Calculate the model concentrations at 1 and 4 hours.
At 1 hour, \(e^{-0.1}=0.90484\), \(e^{-2}=0.13534\), and \(e^{-0.25}=0.77880\). Substitution gives
The fast component contributes \(100(0.73684)(0.76950)=56.70\) mg/L, and the slower component contributes \(100(0.5000)(0.12604)=6.302\) mg/L. Thus \(C(1)=63.00\) mg/L, to the shown precision.
At 4 hours, \(e^{-0.4}=0.67032\), \(e^{-8}=0.0003355\), and \(e^{-1}=0.36788\). Therefore,
The similar concentrations at 1 and 4 hours illustrate how a slower component can sustain input while the faster component is waning, producing a broad shoulder rather than a sharp, isolated second peak. These calculations are model predictions, not evidence that real data follow this mechanism.
Transit-Compartment Input
A transit model represents absorption as a sequence of waiting times. Let \(m\) be the number of transit compartments before the final absorption step, and let every step—including transfer from the last transit compartment into the systemic circulation—have the same first-order rate \(k_{\mathrm{tr}}\). Assume the dose begins in the first transit compartment, no drug is lost during transit, and the systemic disposition is linear. There are \(m+1\) exponential waiting steps in total, so the mean time to systemic input is
The time to input is the sum of \(m+1\) independent exponential waiting times with rate \(k_{\mathrm{tr}}\). Its density is an Erlang distribution. Consequently, the fraction of \(FD\) entering the systemic circulation per unit time at time \(t\) is
The density integrates to one, so this construction also transfers a total amount \(FD\). Its mode for \(m>0\) is \(m/k_{\mathrm{tr}}\), while its mean is \((m+1)/k_{\mathrm{tr}}\). These are different quantities: the input rate is most concentrated around the mode, but the mean includes the later portion of the distribution. The concentration is the convolution of this input with elimination:
This formula shows why input shape and observed concentration shape are related but not identical. Elimination smooths and carries forward the input signal. A smooth transit input can therefore yield a broad concentration peak or shoulder, but the exact profile also depends on disposition.
Worked Example: Interpreting a Three-Step Transit Input
Consider a hypothetical transit model with \(m=2\) transit compartments and \(k_{\mathrm{tr}}=0.60\) h\(^{-1}\). The chain contains three waiting steps, including the final absorption step. Its mean transit time is
The input density is \(g(t)=0.60^3t^2e^{-0.60t}/2\). At 2 hours,
At 5 hours,
At 8 hours,
The input rate has its mode at \(m/k_{\mathrm{tr}}=2/0.60=3.33\) hours and has mean 5.00 hours. If \(FD=100\) mg, for example, the input rate at 5 hours is \(100(0.1344)=13.44\) mg/h. This smooth, distributed input differs in shape from the sum of two separate exponential input rates; it does not imply two discrete absorption fractions.
Reading Diagnostic Patterns
A candidate absorption model is useful when it explains repeatable structure in the data without requiring implausible parameter values or creating new systematic residual patterns. Residuals should be inspected against time, predicted concentration, and relevant dosing or study features. For a profile with a shoulder, pay particular attention to measurements before, through, and after the apparent flattening. Sparse observations can miss the transition entirely.
- Evidence compatible with two input rates: a relatively rapid early contribution and a sustained later contribution may appear as early concentration rise followed by flattening or a renewed increase. A two-rate model can represent components with distinct timescales, but the shoulder alone cannot establish that the dose truly divided into two physical fractions.
- Evidence compatible with transit input: a smooth, delayed, broad input pattern may be described with a transit chain, especially when a single input timescale cannot reproduce the gradual onset and spread. The number of transit compartments and their rate control the shape; they are not directly identified just by counting visible shoulders.
- Evidence against either candidate: residuals that remain systematically positive or negative over an interval, or a new wave of residuals after the model is changed, indicate unresolved structure. Also inspect whether the fit is being driven by a small number of influential observations.
A brief shift of the entire input profile is conceptually different from broadening or splitting its shape: a lag time shifts input later, whereas these alternatives change how input is distributed over time. The earlier material on delayed peaks addresses that distinction. A fair comparison should use the same disposition assumptions and comparable estimation conditions; otherwise a change attributed to absorption may instead reflect a change elsewhere in the model.
Worked Example: Using Residuals to Question a Single-Rate Model
Suppose a hypothetical single-rate model predicts concentrations of 18, 42, 55, 49, and 35 mg/L at 1, 2, 3, 4, and 6 hours. Measurements from one hypothetical profile are 20, 44, 50, 53, and 34 mg/L. Define the raw residual as observed minus predicted concentration, in mg/L.
The residuals are \(20-18=+2\), \(44-42=+2\), \(50-55=-5\), \(53-49=+4\), and \(34-35=-1\) mg/L. The sequence is therefore \(+2,+2,-5,+4,-1\) mg/L.
The negative residual at 3 hours followed by a positive residual at 4 hours is a local reversal: the model is high at one time and low at the next. If this pattern recurs across individuals and is concentrated around the shoulder, it motivates testing an absorption shape that can better represent the transition. One profile is weak evidence, however. Residuals are affected by observation error, and the values do not determine whether a two-rate or transit model is preferable. That decision requires fitting plausible candidates and checking the resulting diagnostics, stability, and predictive performance.
Choosing Between the Alternatives
| Feature | Two first-order components | Transit compartments |
|---|---|---|
| Input shape | Sum of two exponential input rates | One smooth Erlang-distributed input rate |
| Main shape controls | Fraction \(q\) and rates \(k_{a1}\), \(k_{a2}\) | Number of compartments \(m\) and rate \(k_{\mathrm{tr}}\) |
| Typical interpretation | Two fractions with different input timescales | Sequential waiting steps that spread input over time |
| Potential difficulty | Fractions and rates can be hard to distinguish with sparse data | Different compartment counts and rates can produce similar profiles |
Neither alternative is universally superior. The two-component model may be more flexible for distinct fast and slow contributions, but its fraction and rate parameters can be weakly identified when the observed time window does not resolve both components. A transit model imposes a structured, smooth input distribution, but the number of steps and transit rate can also trade off. In either model, a visually improved fit is not enough: examine whether parameters are estimable, whether residual structure is reduced, and whether predictions remain sensible over the observed dosing and sampling conditions.
Population data add a further consideration: individual profiles may be sparse, and an apparent shoulder in the population summary may reflect differences among individuals in peak timing rather than a shared within-individual absorption shape. Assess the fit at the individual and population levels, and avoid interpreting a structural absorption component as a biological mechanism unless the design and supporting evidence justify it.
Common Pitfall and Practical Limits
Common mistake: “There is a shoulder, so the dose must have two absorption phases; I will choose the two-rate model.” A shoulder is compatible with several explanations, including a transit-shaped input, a lag or timing issue, disposition misspecification, and sampling variability. The correction is to treat the pattern as a diagnostic clue, compare explicit alternatives under the same disposition assumptions, and check whether each alternative removes the relevant residual structure without causing other problems.
These models also have limits. The equations above assume a single dose, linear one-compartment disposition, fixed bioavailability, and specified input structure. Real data may not support the distinction between alternatives, particularly when early or intermediate samples are missing. When \(k_{a1}\) and \(k_{a2}\) are close, the two-component model approaches a single-rate shape and the separate contributions are difficult to estimate. Transit models with too many steps can be unnecessarily complex, while too few may not represent the observed spread. The appropriate conclusion may be that the data do not distinguish the candidate absorption shapes.
Check Your Understanding
Answer each question, then compare with the short solutions below.
- In the two-component model, what does \(q\) represent, and what is the total amount entering systemically under the stated assumptions?
- A transit model has \(m=3\) transit compartments and \(k_{\mathrm{tr}}=0.50\) h\(^{-1}\). What is its mean transit time?
- Why is a shoulder insufficient to establish that two physical absorption fractions exist?
- How would you distinguish the conceptual role of a lag time from that of a transit chain?
- Name one reason a two-component model may be difficult to estimate from sparse concentration data.
Solutions. (1) \(q\) is the fraction assigned to the first depot; the total systemic input is \(qFD+(1-q)FD=FD\). (2) There are \(m+1=4\) waiting steps, so \(\mathrm{MTT}=4/0.50=8.00\) hours. (3) A shoulder is a profile pattern compatible with multiple mechanisms and can also arise from data variability or other model errors; it does not uniquely identify a physical process. (4) A lag time shifts input later, while a transit chain spreads input over a sequence of waiting times. (5) If sampling does not resolve both timescales, the fraction and the two absorption rates may trade off and be weakly identified.