Starting Values for Transit and Turnover Models

Why Timing Provides Useful Starting Values

Transit and turnover models are often introduced because a response does not occur immediately after a dose or change in drug concentration. Their parameters therefore have a direct connection to observable timing: a transit rate controls how quickly material moves through a chain, while a turnover rate controls how quickly a response approaches or returns toward baseline. These connections provide useful initial estimates, but they do not make the parameters directly observable without assumptions.

The key distinction is between timing of the model’s input and timing of the measured concentration or response. A delayed absorption input can produce a later plasma concentration peak because absorption is spread over time and drug is simultaneously eliminated. Likewise, a delayed pharmacodynamic response can reflect turnover, drug-effect delay, or both. Initial values should be calculated from timing summaries that correspond to the model quantity being initialized, or adjusted for the process between that quantity and the observation.

Transit Chains: Rate, Number of Stages, and Delay

This explicit convention matters. Some software or publications count the source compartment or the total number of transitions differently. Before translating a delay into \(k_{tr}\), check what the stated compartment count means in the model equations.

If each transition has an exponential waiting time with rate \(k_{tr}\), its mean waiting time is \(1/k_{tr}\). The mean total transit time, often called the mean transit time (MTT), is therefore

For \(n\geq 1\), the absorption-input rate over time has a gamma-shaped profile with shape \(n+1\) and rate \(k_{tr}\). Its mode—the time at which that input rate is largest—is

The mean and the mode are not interchangeable. For example, a model with \(n=3\) has a mean transit time of \(4/k_{tr}\), but its input mode occurs at \(3/k_{tr}\). With \(n=0\), there are no intermediate compartments: the input begins at time zero and the mode is at zero.

What Timing Summaries Can Identify

A single timing summary usually cannot determine both the number of transit compartments and their common rate. If the mean transit time is known, each selected \(n\) implies a different rate through \(k_{tr}=(n+1)/\mathrm{MTT}\). The compartment count controls the shape of the input profile as well as its timing. It therefore needs to be specified from prior model structure or informed by features such as the spread and shape of the estimated input.

When both the input mean and input mode are available and \(n\geq1\), the two equations can be solved together. Their ratio is \(t_{\mathrm{mode,input}}/\mathrm{MTT}=n/(n+1)\). This can suggest a starting count, but in real data the summaries are uncertain; a non-integer implied count is not a literal compartment count.

Turnover Models: Baseline and Recovery Timing

A basic turnover model describes a response \(R(t)\) through a zero-order production rate \(k_{in}\) and a first-order loss rate \(k_{out}\):

Here, \(R\) has response units, \(k_{in}\) has response units per time, and \(k_{out}\) has units of inverse time. In the absence of drug effect or another perturbation, a stable baseline \(R_0\) satisfies \(dR/dt=0\). Thus \(R_0=k_{in}/k_{out}\). Baseline observations can initialize \(R_0\), but they identify only this ratio unless additional information is available.

After a perturbation is removed and the system follows the unmodified turnover equation, the difference from baseline decays exponentially. If \(R(t_0)\) is the response when recovery begins, then \(R(t)-R_0=[R(t_0)-R_0]e^{-k_{out}(t-t_0)}\). The half-life \(t_{1/2}\), the time for this difference to halve, gives \(k_{out}=\ln(2)/t_{1/2}\). This inference assumes recovery is governed by this single turnover process and that the baseline remains stable.

Observed Response Timing Requires a Model Check

The same caution applies to response onset. For a simple step change in production, with a constant new input and the same \(k_{out}\), the response approaches its new equilibrium exponentially. The fraction of the total change completed by time \(t\) is \(1-e^{-k_{out}t}\). Reaching 63.2% of the total change therefore takes \(1/k_{out}\). This provides a possible timing-based start only when the step-input and single-turnover assumptions are plausible. A gradual concentration change or an effect compartment can add delay that this calculation would mistakenly assign to turnover.

Likewise, a plasma concentration peak is not generally the absorption-input mode. Concentration reflects the input filtered through distribution and elimination. A concentration peak can initialize a transit rate only after accounting for those processes, for example through an appropriate model-based calculation or an input estimate. Simply inserting the observed concentration \(t_{\max}\) into \(k_{tr}=n/t_{\max}\) is not justified in general.

Diagnostics, Alternatives, and Limits

Use timing-derived values as informed starts, not as proof that a structure or parameter is correct. Before using them, check whether the timing quantity is an input mean, input mode, response recovery half-life, or raw observed peak. Confirm units, the count convention, and the process assumed to generate the timing. Then compare the initial model’s predicted profiles with the observed delay, peak, and recovery.

• Transit-chain warning signs: a fitted rate at a parameter bound, a large mismatch between predicted and observed delay, or a pronounced residual pattern around the absorption peak can indicate a poor count, a wrong count convention, or inadequate timing information.
• Turnover warning signs: non-exponential recovery, a changing apparent baseline, or different recovery rates across time ranges challenges the single-turnover calculation. Sparse observations may also make a half-life estimate unstable.
• Confounding: absorption delay can be difficult to distinguish from a lag time or slow first-order absorption. Turnover delay can be difficult to distinguish from a delayed drug effect or delayed input. Timing alone may not separate these explanations.

Alternatives include specifying an explicit lag when the data support a delay without a distributed transit profile, using a simpler first-order absorption description when a chain is not warranted, or using a richer response model when recovery is not single-exponential. These are competing structural descriptions, not interchangeable parameterizations in every data set. The appropriate choice depends on the sampling schedule, model assumptions, and whether the data contain information about the relevant timing and shape.

Initial estimates are especially uncertain when observations are sparse near onset or peak, the baseline is poorly characterized, or the response is still changing for reasons not represented in the equations. A mathematically valid calculation can still be a poor start if its assumptions do not match the process. Record the timing definition and compartment-count convention alongside the numerical value so that the initialization can be checked and reproduced.