Calibrating Segmental Absorption with Regional Data

Why Regional Data Change the Calibration Problem

A plasma concentration profile combines absorption from multiple intestinal regions with distribution and elimination. As discussed in the earlier material on why plasma profiles rarely identify regional absorption, a good plasma fit does not by itself establish that uptake was assigned to the correct segments. Regional availability measurements or intestinal sampling can provide direct information about where drug is absorbed or how much remains in the lumen.

The calibration task is to choose segment-specific transit and absorption parameters so that the model predicts those regional observations. The data may be segment-level absorbed fractions, measured luminal amounts at sampling times, or both. The meaning of each observation must be specified before fitting: for example, a fraction of the administered dose absorbed in a segment is not the same quantity as a fraction of the amount entering that segment.

Definition: In a sequential segmental model, \(k_{a,i}\) is the first-order absorption rate constant in segment \(i\), and \(k_{t,i}\) is the first-order rate constant for transfer from segment \(i\) to the next segment. Both have units of inverse time. Regional calibration estimates these parameters, or a selected subset of them, by comparing model predictions with observations that contain information about intestinal region.

Model and Assumptions

For clarity, first consider a dissolved drug amount in a single segment. Let \(A_i(t)\) be the luminal amount in segment \(i\), and let \(I_i(t)\) be its input rate. With constant first-order absorption and forward transfer,

\frac{dA_i(t)}{dt} = I_i(t)-(k_{a,i}+k_{t,i})A_i(t), \qquad J_{a,i}(t)=k_{a,i}A_i(t), \qquad J_{t,i}(t)=k_{t,i}A_i(t).

Here \(J_{a,i}(t)\) is the absorption flux and \(J_{t,i}(t)\) is the transfer flux. For a sequential chain, the transfer flux from one segment supplies the next, so \(I_{i+1}(t)=J_{t,i}(t)\), unless a separate input such as a delayed-release input is specified. This is a mass-conserving description when absorption and forward transfer are the only exits from each segment.

The simple regional-fraction formulas below require a cohort entering a segment, constant rates during its passage, and no competing loss other than absorption and transfer. They also treat the segment as well mixed: every molecule in it has the same instantaneous absorption and transfer hazards. These assumptions are useful for calibration, but they are not guaranteed to describe every formulation or intestinal condition. If the data show time-dependent input, use the time-dependent model equations rather than treating all drug as if it entered at one instant.

From Rates to Regional Predictions

Suppose an amount \(L_i\) enters segment \(i\) as a bolus and there is no further input. The luminal amount declines exponentially at rate \(k_{a,i}+k_{t,i}\). Integrating the absorption flux gives the amount absorbed in that segment; integrating the transfer flux gives the amount passed onward.

A_i(t)=L_i e^{-(k_{a,i}+k_{t,i})t}, \qquad \int_0^\infty J_{a,i}(t)\,dt = L_i\frac{k_{a,i}}{k_{a,i}+k_{t,i}}, \qquad \int_0^\infty J_{t,i}(t)\,dt = L_i\frac{k_{t,i}}{k_{a,i}+k_{t,i}}.

The two integrated fractions sum to one, as required by mass conservation. Define \(q_i=k_{a,i}/(k_{a,i}+k_{t,i})\), the fraction of the amount entering segment \(i\) absorbed there under these assumptions. The fraction passed to the next segment is \(1-q_i\). Consequently, the fraction of the initial amount absorbed in segment \(i\) is

g_i = q_i\prod_{j=1}^{i-1}(1-q_j).

For the first segment, the product over preceding segments is defined as one. The fraction remaining after the final modeled segment is \(\prod_{j=1}^{m}(1-q_j)\), where \(m\) is the number of segments. Thus the regional fractions and the remaining fraction sum to one. If only \(q_i\) is known, the regional fraction within a segment identifies the rate ratio, not both rates separately:

\frac{k_{a,i}}{k_{t,i}}=\frac{q_i}{1-q_i}.

Multiplying both rates by the same positive constant leaves \(q_i\) unchanged. Information about the time course, such as intestinal amounts measured at several times, or independent residence-time data, is needed to distinguish their absolute magnitudes.

Worked Example: Estimating Absorption Rates From Regional Fractions

Consider three hypothetical segments. Residence-time information is assumed to have supplied transit rates \(k_{t,1}=0.50\), \(k_{t,2}=0.40\), and \(k_{t,3}=0.30\) per hour. Regional measurements indicate conditional segmental absorption fractions \(q_1=0.50\), \(q_2=0.32\), and \(q_3=0.48\). Estimate the absorption rate constants using the competing-rate model.

Rearranging \(q_i=k_{a,i}/(k_{a,i}+k_{t,i})\) gives \(k_{a,i}=k_{t,i}q_i/(1-q_i)\). Therefore,

k_{a,1}=0.50\frac{0.50}{0.50}=0.500\ {\rm h}^{-1}, \qquad k_{a,2}=0.40\frac{0.32}{0.68}=0.1882\ {\rm h}^{-1}, \qquad k_{a,3}=0.30\frac{0.48}{0.52}=0.2769\ {\rm h}^{-1}.

If 100 mg enters segment 1, the predicted regional absorbed amounts are 50.0 mg in segment 1, \(100(1-0.50)(0.32)=16.0\) mg in segment 2, and \(100(1-0.50)(1-0.32)(0.48)=16.32\) mg in segment 3. The remaining amount is \(100(0.50)(0.68)(0.52)=17.68\) mg. These values sum to 100 mg. The calculation relies on the supplied transit rates; regional fractions alone would not have determined these absorption rates.

Fitting to Regional Availability or Sampling Data

For regional availability data, let \(y_i\) be the observed fraction of the administered amount absorbed in segment \(i\). The model prediction is \(g_i\), not \(q_i\), unless the data are explicitly expressed as a fraction of the amount entering that segment. For intestinal sampling, the observation model instead predicts the sampled luminal amount at each time, using the differential equations and the sampling times. The appropriate observation equation depends on whether the assay measures total drug, dissolved drug, or another defined quantity.

A common calibration method is weighted nonlinear least squares. If observation \(y_j\) has a known measurement standard deviation \(\sigma_j\), and the model prediction at candidate parameters is \(f_j\), minimize the sum of squared standardized residuals. Parameters must be positive; one practical method is to optimize their logarithms and transform back to rate constants. The assumed error model should reflect how the data were measured. Independent errors with known standard deviations are a convenient starting assumption, not a universal property of regional data.

Q(\boldsymbol{\theta}) = \sum_{j=1}^{n} \left(\frac{y_j-f_j(\boldsymbol{\theta})}{\sigma_j}\right)^2, \qquad k_{a,i}>0,\quad k_{t,i}>0.

Here \(\boldsymbol{\theta}\) is the vector of estimated parameters and \(n\) is the number of observations. If the observations are correlated, the objective should account for their covariance rather than treating each standardized residual as independent. If transit rates are already informed by residence-time measurements, they can be fixed or estimated jointly with those measurements; treating uncertain transit rates as exact can understate uncertainty in absorption rates.

Worked Example: Checking Regional Residuals

Use the three-segment predictions from the preceding example: 50.0, 16.0, and 16.32 mg absorbed in the respective segments. Suppose hypothetical regional measurements are 50.0, 12.0, and 20.0 mg, and each measurement has an independently known standard deviation of 2.0 mg. Calculate the standardized residuals and weighted sum of squares.

For each segment, the standardized residual is observed minus predicted, divided by its standard deviation. The results are \(0/2.0=0\), \((12.0-16.0)/2.0=-2.00\), and \((20.0-16.32)/2.0=1.84\). Therefore,

$$ Q=0^2+(-2.00)^2+(1.84)^2=7.39. $$

The residuals show that the model predicts too much absorption in segment 2 and too little in segment 3 for these observations. The near-opposite discrepancies could indicate that the model places uptake in the wrong region even though the total absorbed amount is similar. \(Q=7.39\) is a discrepancy summary at the stated parameter values; by itself, it is not a formal goodness-of-fit conclusion. Such a conclusion also requires an appropriate error model and consideration of how many parameters were fitted.

Residual Patterns and Parameter Uncertainty

Residuals should be examined by segment, sampling time, and predicted amount, not only summarized in one objective value. A consistent run of positive residuals in later segments and negative residuals in earlier segments suggests a regional allocation mismatch. Time-ordered residuals that change systematically can indicate that the predicted transit or depletion time course is wrong. A large isolated residual may instead reflect measurement error, an incorrectly recorded sample, or an observation-model mismatch. These patterns are clues for investigation, not proof of a particular mechanism.

Parameter uncertainty can be assessed locally from how sensitive the predictions are to the fitted parameters. Let \(J\) be the matrix of derivatives of model predictions with respect to parameters, evaluated at the fitted values. With independent errors of known standard deviations, a common approximate covariance matrix is \((J^\mathsf{T}WJ)^{-1}\), where \(W\) has diagonal entries \(1/\sigma_j^2\). This approximation relies on a locally linear model, a suitable error model, and enough information to distinguish the parameters. Strong parameter correlation or a nearly singular matrix indicates weak identification; a numerical optimizer returning a value does not establish that the value is precise.

For a regional fraction \(q_i\) and a known \(k_{t,i}\), uncertainty in \(q_i\) propagates to \(k_{a,i}=k_{t,i}q_i/(1-q_i)\). The sensitivity is \(d k_{a,i}/dq_i=k_{t,i}/(1-q_i)^2\). Thus, uncertainty in \(q_i\) has an especially large effect as \(q_i\) approaches one. If \(k_{t,i}\) is also uncertain, its uncertainty and its association with \(q_i\) must also be included; the calculation below deliberately treats it as known.

Worked Example: Approximate Uncertainty in an Absorption Rate

Suppose a hypothetical regional study estimates \(q_2=0.32\) with standard error 0.04, while \(k_{t,2}=0.40\ {\rm h}^{-1}\) is treated as known. Estimate the absorption rate and a local approximate 95% interval, using a normal approximation for \(q_2\) and the first-order delta method.

The point estimate is \(k_{a,2}=0.40(0.32)/(1-0.32)=0.1882\ {\rm h}^{-1}\). The derivative at \(q_2=0.32\) is

\frac{d k_{a,2}}{d q_2} = \frac{0.40}{(1-0.32)^2} = 0.8651\ {\rm h}^{-1}.

The approximate standard error is \(0.8651(0.04)=0.0346\ {\rm h}^{-1}\). The corresponding normal-approximation interval is \(0.1882\pm1.96(0.0346)\), or approximately \(0.120\) to \(0.256\ {\rm h}^{-1}\). This interval is conditional on treating the transit rate as known and on the local approximation being adequate. It is not a statement that there is a 95% probability that the fixed rate lies in this particular interval.

Choosing Data and Recognizing Limitations

Regional measurements provide more direct location information than plasma profiles, but they do not automatically identify every segmental rate. Endpoint regional fractions mainly constrain competing-rate ratios. Timed luminal samples can add information about the rate at which amounts disappear, while independent residence-time data can help constrain transit. Combining data types can therefore be more informative than fitting many segment-specific rates to one type of observation.

Data source	What it can constrain	Key limitation
Regional absorbed fractions	Where uptake is allocated and, with transit information, absorption rates	Fractions alone generally identify rate ratios rather than absolute rates
Timed intestinal samples	Amounts remaining in sampled regions over time	Requires a clear sampling and assay observation model
Plasma profiles	Systemic concentration-time behavior	Do not by themselves establish the intestinal region of absorption

A useful calibration strategy is to start with the smallest set of parameters supported by the data, compare predicted and observed regional quantities, and then assess whether additional parameters are actually informed. Uncertainty should reflect measurement error and any estimated transit parameters. More detailed segment-specific estimates are not necessarily more reliable: if the observations cannot distinguish them, the estimates may be unstable or strongly correlated.

Common Mistake

Wrong reasoning: “The measured fraction absorbed in segment 2 is \(q_2\), so I can use it directly as the fraction of the original dose absorbed there.”
Correction: \(q_2\) is conditional on the amount entering segment 2. The fraction of the original dose absorbed there is \(g_2=q_2(1-q_1)\) in a three-segment sequential model. Track the amount passed forward from each preceding segment before comparing with dose-normalized regional measurements.
Another warning: A close plasma fit is not a substitute for checking regional predictions when the calibration target is regional absorption.

Key takeaway: Regional data can anchor where absorption occurs, but the information they provide depends on how the observations are defined. Check the predicted quantity against that definition, use transit or time-course information to separate rate magnitudes, and inspect both residual patterns and parameter uncertainty.

Check Your Understanding

Answer each question, then compare with the brief solutions below.

For one segment, \(k_a=0.30\ {\rm h}^{-1}\) and \(k_t=0.20\ {\rm h}^{-1}\). What fraction of the entering amount is absorbed there, and what fraction passes onward?
Why can a regional endpoint fraction alone fail to determine both \(k_a\) and \(k_t\)?
If 60 mg enters a segment and its conditional absorbed fraction is 0.25, how much is absorbed there?
Name one residual pattern that could suggest regional misallocation, and explain why it is a diagnostic clue rather than proof.

Solutions. (1) The absorbed fraction is \(0.30/(0.30+0.20)=0.60\); the onward fraction is \(0.20/0.50=0.40\). (2) The endpoint fraction identifies \(k_a/k_t=q/(1-q)\), which is unchanged if both rates are multiplied by the same positive constant. (3) The absorbed amount is \(60(0.25)=15\) mg. (4) For example, negative residuals in an earlier segment followed by positive residuals in a later one may indicate that predicted uptake is shifted upstream; other model or measurement problems can create a similar pattern, so the residuals prompt investigation rather than establish its cause.