Verifying Dose Event Implementation

Why Dose Events Need Their Own Review

A model can have correctly written structural equations and parameter units yet still receive the wrong dose history. An amount may be assigned to the wrong time, an infusion may stop too early, or a repeated-dose record may be interpreted differently from the intended schedule. Each error changes the input to the model and can alter every later prediction. Reviewing dose implementation therefore means tracing the dose records into the model's actual input function, then checking the resulting amount and concentration at selected times.

The calculations below use a one-compartment model with linear elimination. This is a deliberately simple setting for an independent event check: the purpose is to make the relationship between records and input explicit, not to replace review of the model's full structural equations. Assume the amount in the compartment is \(A(t)\), the volume is \(V\), the elimination rate constant is \(k\), and the concentration is \(C(t)=A(t)/V\). Between dose events, elimination follows a first-order process. During an infusion, the input rate is constant over each recorded infusion interval.

For a bolus at time \(T\), the amount just after the event is \(A(T^+)=A(T^-)+D\), where \(T^-\) and \(T^+\) denote the instants immediately before and after the event. For an infusion, \(R_{\text{in}}(t)=R\) while the infusion is active and is zero when it is not, apart from any other simultaneous inputs. During intervals without a bolus jump, the model equation is

The event review assumes that time units are consistent across dose times, infusion durations, and \(k\); that amounts, rates, and volume use compatible units; and that the stated event schedule is the intended schedule. It also assumes the model's stated compartment and linear elimination apply over the interval being checked. If a model has more complex input, time-varying parameters, or nonlinear elimination, the event logic still needs review, but these simple closed-form calculations do not by themselves verify those additional features.

From Event Records to an Amount Trajectory

Between events, the differential equation can be solved with an integrating factor. Multiplying both sides by \(e^{kt}\) gives \(e^{kt}\frac{dA}{dt}+ke^{kt}A=e^{kt}R_{\text{in}}(t)\), whose left side is \(\frac{d}{dt}\left(e^{kt}A(t)\right)\). Integrating from \(t_0\) to \(t\), provided there is no unrepresented bolus jump inside the interval, yields

This expression has a direct audit interpretation: the initial amount decays from \(t_0\), and each amount entering later contributes after being discounted by elimination between its entry time and \(t\). A bolus of amount \(D_j\) at \(T_j\) contributes \(D_j e^{-k(t-T_j)}\) for \(t\geq T_j\). An infusion at constant rate \(R\) from \(a\) to \(b\) contributes \(R\int_a^b e^{-k(t-s)}ds\) when \(t\geq b\).

While the infusion is running, from \(a\) through \(t\leq b\), its accumulated contribution is \(\frac{R}{k}(1-e^{-k(t-a)})\). At the start of an infusion, the amount is continuous; the input rate changes, but the amount does not jump. At its end, the rate returns to zero and the amount then declines through elimination. These distinctions help identify a model that incorrectly treats an infusion as a bolus or gives the infusion an unintended duration.

Repeated Doses, Dose Changes, and Event Semantics

A repeated-dose schedule is not a single event with an average amount; it is a sequence of inputs at specified times. If doses are equally spaced by an interval \(\tau\), one way to check the schedule is to verify that the event times satisfy \(T_j=T_0+j\tau\) for the intended indices \(j\), and that the number of generated events matches the intended count. If a record format stores a dose plus a repeat interval or additional-dose count, expand that shorthand into its actual event times before comparing it with the model trace. Confirm whether the first dose is included in the stated count and whether the terminal event time is within the intended treatment period.

A dose change ordinarily changes the amount or rate of a later event; it does not rewrite drug already administered. For bolus dosing, a new 150 mg event is an additional 150 mg bolus at its recorded time, not a replacement of an earlier 100 mg dose. For an infusion rate change, represent the schedule as adjacent intervals with the relevant rates. Unless the specification explicitly says otherwise, the amount delivered before the change remains part of the compartment and continues to be eliminated.

Practical Review: Diagnostics and Alternatives

An independent event check should compare three things: the source dose records, the event records supplied to the model, and the model's resulting input function or event trace. For each event, verify subject assignment, event time, amount, route or compartment, and—when applicable—rate and duration. Then check selected times immediately before and after boluses, at infusion starts and ends, and during dose-free intervals. The expected behavior is specific: a bolus causes an amount jump of exactly \(D\); an infusion start changes the input rate without an amount jump; an infusion stop ends that input; and an interval with no input produces exponential decline under this model.

Useful warning signs include a negative or implausible duration, a rate that does not equal amount divided by duration, events out of chronological order, a repeated-dose count that produces the wrong number of times, or a discontinuity at an infusion boundary that the specified input does not imply. Check units before interpreting a numerical discrepancy: a rate entered in mg/min but treated as mg/h differs by a factor of 60. Also inspect conventions for events at the same time. A bolus and a new infusion may coincide; the intended model must apply both inputs, while the event implementation must not silently discard one because of ordering or record-handling rules.

For small examples or selected audit times, a hand calculation using the event ledger is transparent and easy to reproduce. A separate script that expands the records and computes the input function can efficiently check many subjects and repeated events, but its event-expansion logic should itself be reviewed rather than treated as independent evidence by default. Inspecting model code can show how records are interpreted, while simulation can reveal the resulting trajectory; neither alone proves the input matches the written schedule. A strong check compares these approaches at selected event boundaries and times between events.

These checks have limits. Agreement at a few times does not prove every record is correct, and a one-compartment analytic calculation cannot validate a more complex absorption, transit, or nonlinear input structure. Analytic checks are especially useful as targeted tests: select events and intervals that could expose timing, unit, repetition, or rate-translation errors, then compare those expectations with the implementation. Broader coverage may require automated checks across all subjects, but unusual cases—such as simultaneous events, a final partial interval, or a dose change during an infusion—still merit explicit review.