From a Sampled Subject to a Virtual Patient
A simulated profile depends on more than a parameter vector. The patient’s sampled covariates and individual variability effects determine subject-specific model parameters, while a dosing history determines when and how drug enters the model. If either component is assigned incorrectly, the resulting profile can correspond to no plausible patient or regimen.
Here, a virtual patient means one simulated subject represented by a linked record of covariates, individual effects, model parameters derived from them, and dosing events. Its dosing history is an explicit, time-ordered record of administrations or infusions. The record may be shared by a group of subjects, sampled from observed histories, or generated from subject-specific rules. The choice must match the simulation question.
This construction relies on several assumptions that should be made explicit. Covariates and individual effects must be sampled according to the intended population or conditioning scheme. The parameter relationships must use the same units and definitions as the model. Each dosing event must describe a clinically meaningful administration, and the event record must state whether times and doses are planned or observed. Unless adherence or dose uncertainty is explicitly modeled, the simulation assumes that each listed dose is administered as recorded.
An individual effect is generally attached to a subject, not redrawn at every dose. For example, if \(\eta_i\) represents subject-specific clearance variability, the same \(\eta_i\) applies across that subject’s dosing history unless the model explicitly defines a time-varying effect. This preserves the distinction between between-subject variability and changes in regimen.
Representing Doses as Events
A useful dosing record has one row per event, rather than one row per subject with a compressed description such as “twice daily for five days.” At minimum, each event should identify the subject, event time, amount, dose unit, and route. Depending on the model, it may also require formulation, infusion rate or duration, treatment identifier, and a flag indicating whether the event is a dose, a regimen change, or another event. State the time origin and units—for example, hours since the start of simulation.
| Field | Purpose | Example |
|---|---|---|
| Subject identifier | Links the event to one virtual patient | VP-014 |
| Event time | Places the event on the patient’s time scale | 24 h |
| Amount and unit | Specifies the administered quantity | 150 mg |
| Route or input type | Determines how the model receives the dose | Oral, bolus, infusion |
| Regimen or treatment identifier | Records the applicable treatment period | Regimen B |
Repeated doses are represented by repeated dose events. The dosing interval is a rule for generating event times, not a substitute for those times once the history has been constructed. This distinction matters when subjects have different start times, interruptions, or treatment changes. A regimen change should be recorded explicitly or be unambiguously recoverable from the event records; it should not silently rewrite earlier doses.
For an idealized instantaneous input, the amount entering a model compartment can be written as a sum of dose inputs at their event times. For a constant-rate infusion, the input is represented by a rate over the infusion interval. These expressions describe the input to the model, not the concentration itself.
Here \(A_{ij}\) is the amount for event \(j\) of subject \(i\), \(t_{ij}\) is its time, \(R_{ij}\) is its infusion rate, \(D_{ij}\) is its duration, \(\delta\) denotes an instantaneous input, and \(\mathbf{1}\) is one during the stated interval and zero otherwise. The equations make an important construction rule explicit: two events for the same subject both contribute to the input. A dose record must therefore preserve both events rather than retain only the most recent dose.
Worked Example: Linking Covariates and Individual Effects to a Dose History
Consider a hypothetical virtual patient weighing 70 kg. Suppose a one-compartment model defines clearance and volume as \(CL_i=5( WT_i/70)^{0.75}\exp(\eta_{CL,i})\) L/h and \(V_i=50(WT_i/70)\exp(\eta_{V,i})\) L. For this illustration, the sampled effects are \(\eta_{CL,i}=\ln(1.10)\) and \(\eta_{V,i}=\ln(0.90)\). The patient is assigned oral 100 mg doses at 0, 12, and 24 h.
The weight ratio is \(70/70=1\). Thus \(CL_i=5(1)^{0.75}(1.10)=5.50\) L/h, and \(V_i=50(1)(0.90)=45.0\) L. The same individual effects are used for all three doses because they are subject-level effects in this model.
The dose ledger contains three distinct records: (VP-001, 0 h, 100 mg, oral), (VP-001, 12 h, 100 mg, oral), and (VP-001, 24 h, 100 mg, oral). The total amount administered by 24 h, counting the event at 24 h, is \(3(100)=300\) mg. The virtual patient is therefore not merely the parameter pair \((5.50\text{ L/h},45.0\text{ L})\); it is that pair, the sampled characteristics and effects, and the linked three-event history.
Repeated Doses and Treatment Changes
For a regular schedule beginning at time \(t_0\), with interval \(\tau\), the \(k\)th scheduled dose time is \(t_k=t_0+k\tau\), for integer \(k\) starting at zero. This rule is valid only over the period for which the regimen remains unchanged. If the interval or amount changes, create a new regimen period and generate its events from the applicable rule. Do not apply the original interval beyond the change merely because it was used to construct earlier events.
The timing convention at a treatment change must also be specified. A change at 30 h could take effect immediately, at the next scheduled administration, or after a stated transition period. The change itself does not imply that a dose is administered at 30 h. The event table should contain a dose at that time only if the intended regimen actually gives one then.
Worked Example: Recording a Change in Dose and Interval
Suppose a hypothetical patient receives 100 mg orally every 12 h, starting at 0 h, through the dose at 24 h. A treatment change is effective at 30 h. Under the stated plan, the new regimen is 150 mg orally every 24 h, beginning with the next dose at 48 h. There is no dose at the time of the change.
Before the change, the times are \(0,12,24\) h, since \(t_k=0+12k\) for \(k=0,1,2\). The first dose under the new regimen is explicitly set at 48 h, and the following event is at \(48+24=72\) h. The event ledger is therefore 100 mg at 0, 12, and 24 h, followed by 150 mg at 48 and 72 h.
By 72 h, the listed amount is \(3(100)+2(150)=300+300=600\) mg. The change at 30 h does not add to this total because it is a treatment-change time, not an administration. Keeping the change time and the actual dose times distinct prevents a common accidental extra dose.
When the same schedule applies to many virtual patients, it can be stored as a regimen template and linked to each subject. When schedules differ, each subject needs a separate event history or a rule that generates one. Subject identifiers are essential in either case: sorting or joining event records must not attach one patient’s dose history to another patient’s covariates or effects.
Generating Subject-Specific Schedules
A patient-specific schedule may be generated from covariates, treatment assignment, or a sampled history. For a rule-based schedule, specify the inputs, units, rounding convention, start time, and change rules before generating events. For a sampled history, preserve the intended relationship among events: resampling one patient’s complete history keeps that patient’s dose timing and changes together, whereas independently sampling dose amounts and intervals can create combinations that were never observed and may not represent the target population.
The schedule is part of the simulation design, not a consequence of the pharmacokinetic model. In particular, a model that predicts concentration after administration does not by itself determine which regimen a virtual patient receives. The assigned history must come from the stated study question or an explicit generation rule.
Worked Example: Applying a Weight-Based Rule to Different Patients
Suppose two hypothetical patients receive an oral dose of 2 mg/kg at time 0, with another dose at 24 h. At 36 h, treatment changes to 1.5 mg/kg every 12 h, beginning at 48 h. Patient A weighs 60 kg; Patient B weighs 80 kg. Assume weights remain fixed and the calculated doses are administered without rounding.
For Patient A, the first-regimen amount is \(2\text{ mg/kg}\times60\text{ kg}=120\) mg. The change-regimen amount is \(1.5\text{ mg/kg}\times60\text{ kg}=90\) mg. Patient A’s events are 120 mg at 0 and 24 h, then 90 mg at 48 and 60 h.
For Patient B, the corresponding amounts are \(2\times80=160\) mg and \(1.5\times80=120\) mg. Patient B’s events are 160 mg at 0 and 24 h, then 120 mg at 48 and 60 h. The treatment change at 36 h creates no dose event because the specified new schedule begins at 48 h.
By 60 h, Patient A has received \(2(120)+2(90)=240+180=420\) mg, and Patient B has received \(2(160)+2(120)=320+240=560\) mg. Although the timing pattern is shared, the histories are not interchangeable: their amounts depend on subject-specific weight. Each event must remain linked to the correct patient.
Construction Checks and Failure Modes
Before running a simulation, inspect both the patient-level records and the event-level records. These checks do not prove that a regimen is clinically appropriate, but they can identify common construction errors before they propagate into simulated profiles.
- Check units. Confirm that dose units, weight units, time units, infusion rates, and model parameter units agree. A dose computed in mg from a weight in kg must use a dose rule expressed per kg.
- Check event ordering and duplicates. Sort events within subject by time and investigate duplicate records. Duplicate times may be intentional, but should not arise from generating the same regimen period twice.
- Check regimen boundaries. Confirm that no event follows an obsolete schedule after a change, and that a change time has not been treated as a dose without a stated reason.
- Check linkage and completeness. Every event should refer to a defined subject, and every subject should have the intended history. Verify that covariates and individual effects are not inadvertently joined by row order alone.
- Check plausibility against the design. Compare event counts, amounts, intervals, and routes with the intended regimen or sampled source records. Unusual values require investigation, not automatic deletion.
A common mistake is to treat a regimen label such as “100 mg twice daily” as sufficient history and to assume that every patient starts at time zero and follows it indefinitely. That reasoning fails when start times, treatment changes, or subject-specific schedules differ. The correction is to define the event times and amounts for each patient—or to specify a rule that generates them—and to record the assumptions that make the generated history valid.
Another choice is whether to assign one common schedule, generate schedules from rules, or resample complete observed histories. A common template is simple and appropriate when the study question fixes the regimen for everyone. Rule-based generation accommodates covariate-dependent amounts or planned changes. Complete-history resampling can retain observed patterns, but represents the source population only to the extent that the sampled histories are relevant to the target. No approach automatically captures missed doses, uncertain adherence, or unrecorded treatment changes; those require additional data or explicit modeling.
A dosing history also has limits. A correctly constructed record does not validate the structural model, establish that a dose was absorbed, or account for unmodeled formulation differences. It specifies model input according to the chosen assumptions. If actual administration times are uncertain, a deterministic event ledger may overstate precision. If the simulation needs that uncertainty, represent it rather than silently treating planned doses as actual doses.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A virtual patient has an individual clearance effect drawn once. Should that effect ordinarily be redrawn at each scheduled dose? Explain.
- A patient receives 80 mg at 0 and 8 h. A regimen change occurs at 10 h, and the new regimen begins at 20 h. How many dose events occur through 20 h, and what amounts are listed if the new dose is 120 mg?
- A dose rule is 1.25 mg/kg for a patient weighing 64 kg. Calculate the dose and state its unit.
- Why can resampling each dose time independently from a collection of observed histories fail to preserve the intended dosing-history structure?
- Name two checks that can reveal a broken link between a virtual patient and that patient’s dosing events.
Solutions. (1) No. If it represents a subject-level effect, the same value applies across that patient’s doses unless the model defines a time-varying effect. (2) There are three events: 80 mg at 0 h, 80 mg at 8 h, and 120 mg at 20 h. The change at 10 h is not an extra dose. (3) \(1.25\text{ mg/kg}\times64\text{ kg}=80\) mg. (4) Independent resampling can combine times from different patients or regimens, breaking their within-history pattern and creating schedules not represented in the source histories. Resampling a complete history preserves its event structure. (5) Check that every event identifier maps to a defined virtual patient, and verify that covariates, effects, and events are joined using subject identifiers rather than row order. Sorting and inspecting event times by subject can also expose histories assigned to the wrong patient.