Why Audit a Parameter Dossier?
A PBPK parameter table can look complete while hiding gaps that change the meaning of its values. A clearance number may be copied from a secondary source without its assay conditions, scaling steps, or units. A tissue-to-plasma ratio may use a different species or denominator than the model requires. An audit makes those gaps visible and records whether each input is traceable, interpretable, and suitable for its intended use.
The earlier material on parameter records, evidence ranking, assay reconstruction, measurement comparability, extraction, transformations, uncertainty, conflicting evidence, and versioning supplies the vocabulary for this audit. Here the emphasis is on applying that vocabulary consistently to a compact dossier and correcting representative records. An audit does not make weak evidence strong: it distinguishes what is known, what was assumed, and what remains unresolved.
A Practical Audit Sequence
Audit the chain from model need to selected input, not just the final numeric cell. First state what the model needs: for example, hepatic intrinsic clearance in a specified unit or an equilibrium tissue-to-plasma partition coefficient for a named tissue and species. Then check whether the source actually measured that quantity or whether the value is derived from a different endpoint. A citation alone is not a complete provenance record; the cited item must lead to the original measurement or clearly identify the derivation being relied on.
Record the parameter name, biological entity, target population or species, units, and intended model role. Clarify whether the required input is directly measured or derived.
Follow the citation to the original source where possible. Preserve the source identifier, page, table, figure, or other location, and note any secondary-source link in the chain.
Record the assay, biological material, species, conditions, analytical method when available, and reported qualifiers. Check that these details make the measurement interpretable for the target.
Write the equation, values, units, and assumptions that convert the reported result to the model input. Check dimensional consistency and retain the untransformed value.
Separate reported measurement uncertainty from uncertainty introduced by scaling or assumptions. State the situations to which the evidence does and does not apply.
Compare the evidence on the dimensions relevant to the model. Document the selection rationale, alternatives, unresolved limitations, and versioned change rather than silently replacing a value.
What Makes a Record Auditable?
A useful record allows a reviewer to answer a few concrete questions without guessing: What exactly was measured? In what material, species, and conditions? Where did the number appear in the original evidence? Was it reported directly or calculated? If calculated, can the calculation be reproduced? What uncertainty and applicability limits remain? These questions apply to both experimental and model-dependent inputs.
For assay detail, record what is available rather than filling gaps with plausible-sounding assumptions. Depending on the measurement, useful details may include the biological preparation, incubation or sampling conditions, analyte and matrix, assay method, and whether a value is total, unbound, apparent, or otherwise qualified. Missing information should be labelled as unavailable, not inferred from a familiar protocol. Retain qualifiers such as “approximately,” “less than,” or “apparent”; replacing them with an exact point value changes the evidence.
For a derived input, keep the reported measurement and the model-ready value as separate fields. The transformation record should include the equation, conversion factors, their sources, and any assumptions. This is especially important when a conversion depends on biological scaling factors: a numerically correct calculation can still be inappropriate if those factors do not match the intended species or preparation.
Worked Example: Correcting a Scaled Clearance Record
Hypothetical dossier entry: “Human hepatic clearance = 480 mL/min; source: review article.” The model developer says the number came from an in vitro microsomal assay, but the table contains no assay conditions, calculation, or source location. Audit and correct the entry for a hypothetical 70 kg adult.
Audit finding. The label is misleading: the available evidence describes a scaled estimate from microsomal intrinsic clearance, not a directly measured human hepatic clearance. For this example, suppose the original assay reports apparent intrinsic clearance of \(8.0\ \mu\text{L}/\text{min}/\text{mg}\) microsomal protein. Suppose the documented scaling factors are \(40\ \text{mg}\) microsomal protein per gram liver and \(1500\ \text{g}\) liver. Assume these factors are appropriate for the intended calculation, and that no additional correction for binding or hepatic blood-flow effects is being applied.
The unbound-independent scaling calculation for this record is:
Substitute the hypothetical values and retain the units:
The units cancel to volume per time. For the stated body mass, the mass-normalized estimate is \(480\ \text{mL/min} \div 70\ \text{kg}=6.86\ \text{mL/min/kg}\). This arithmetic reproduces the reported 480 mL/min, but it does not establish that this is whole-body systemic clearance. The record should call it a scaled in vitro intrinsic clearance estimate, retain the assay value and factors, and state that further corrections or a separate model calculation would be needed for a different clearance quantity.
Corrected record: “Apparent microsomal intrinsic clearance: \(8.0\ \mu\text{L}/\text{min}/\text{mg}\) protein; hypothetical assay conditions and source location to be recovered from the original report; scaled estimate: \(480\ \text{mL/min}\) using \(40\ \text{mg/g}\) and \(1500\ \text{g}\); intended use: specified hepatic intrinsic-clearance input; limitations: scaling factors are assumed applicable, and the value is not labelled as observed systemic clearance.” The audit should mark source and assay metadata as unresolved until verified.
Showing Uncertainty Without Inventing Precision
An input record should distinguish uncertainty in the measured value from uncertainty in the steps that make it model-ready. For the scaled-clearance example, suppose the dossier reports a plausible assay range of \(6\) to \(10\ \mu\text{L}/\text{min}/\text{mg}\), while the scaling factors are treated as fixed for a simple sensitivity check. The corresponding scaled values are \(360\) to \(600\ \text{mL/min}\). If scaling factors are also uncertain, that contribution should be described or explored separately; it should not be hidden inside an unsupported number of decimal places.
A range calculated by combining the smallest and largest plausible values is a bounding scenario, not automatically a confidence interval or probability statement. For example, if the protein scaling factor is also considered to range from \(35\) to \(45\ \text{mg/g}\), and liver mass from \(1400\) to \(1600\ \text{g}\), the extreme-product bounds are:
This \(294\)–\(720\ \text{mL/min}\) range is an extreme-combination scenario under the stated bounds. It is not a formal 95% interval, and it does not account for dependence among factors. The record should name the range type and the values varied. The earlier material on input uncertainty provides the broader framework; the audit requirement here is to preserve enough detail that the stated uncertainty can be interpreted and reproduced.
Auditing Tissue Partitioning Inputs
A tissue-to-plasma partition coefficient, commonly written \(K_p\), describes a ratio of tissue concentration to plasma concentration under a specified convention and condition. A dossier must state the tissue, species, concentration basis, denominator matrix, and whether the measurement is intended to represent equilibrium partitioning. “Brain \(K_p=3.15\)” alone is not sufficient: a reviewer cannot tell whether it is a tissue-to-plasma or tissue-to-blood ratio, which species was studied, or whether it is an observed ratio at one sampling time.
For a direct ratio, numerator and denominator must be expressed on compatible concentration bases. If tissue concentration is reported per unit tissue mass and plasma concentration per unit volume, conversion requires an explicit tissue-density assumption. If the evidence instead provides concentrations on compatible volume bases, no such mass-to-volume conversion is needed. The chosen convention must match the model input definition; a ratio with the wrong denominator cannot be repaired by rounding.
Worked Example: Reconstructing a Tissue-to-Plasma Ratio
Hypothetical dossier entry: “Brain \(K_p=3.15\), literature value.” Suppose the original hypothetical study reports a steady-state brain concentration of \(6.30\ \mu\text{g/g}\) tissue and a plasma concentration of \(2.10\ \mu\text{g/mL}\) in rats. The model needs a total tissue-to-plasma concentration ratio for the same species. For this calculation, the dossier supplies a tissue density of \(1.05\ \text{g/mL}\), and the study conditions are assumed to support the stated steady-state comparison.
Convert the tissue concentration to a volume basis:
Then divide by plasma concentration:
The units cancel, so the ratio is dimensionless under this concentration convention. The corrected record retains both measured concentrations, the tissue-density conversion, and the assumption that the study conditions support an equilibrium interpretation. If the measurements were only taken at a single non-equilibrium time point, the defensible label would instead be a time-specific observed tissue-to-plasma ratio; it should not be relabelled as an equilibrium \(K_p\) merely because the arithmetic gives a ratio.
Making a Defensible Conflict Decision
When two values differ, first determine whether they estimate the same target. Compare species, tissue or biological preparation, assay conditions, concentration basis, sampling condition, reported qualifiers, and transformation methods. A difference may reflect a genuine biological or methodological distinction rather than an error. Evidence ranking matters, but a highly ranked source is not automatically applicable if its measurement does not match the model’s target.
The main alternatives are to select one directly measured input, derive an input from another measurement, or use a model-dependent estimate. A direct measurement can be preferable when it matches the intended species and endpoint and is well documented. A derived value may be necessary when the model needs a quantity not reported directly, but its transformations and assumptions must be explicit. A model-dependent estimate can be useful when direct evidence is unavailable, but it should be labelled as derived and its dependence on other inputs recorded. None of these approaches justifies averaging incompatible values without a scientific reason.
Worked Example: Resolving Conflicting Partitioning Records
Consider two hypothetical entries for a rat brain tissue-to-plasma ratio. Record A is the steady-state result just audited: \(K_p=3.15\), based on a tissue measurement converted using a documented density. Record B reports \(K_p=4.2\) for mouse brain from a secondary review, with no original source location or concentration-basis details. The PBPK model being populated is a rat model and needs a rat brain input.
Check comparability. The values are not equivalent evidence for the same target: Record B is from a different species, is traceable only to a secondary review, and lacks the basis needed to check its ratio. It therefore does not directly contradict Record A for the rat input. Do not average the values to obtain \(3.68\); that number would combine distinct and incompletely documented evidence without a defensible interpretation.
Decision. Retain \(3.15\) provisionally for the rat model, with the assumption and source-location gap stated in the record. Mark Record B as a non-comparable contextual value pending verification of its original source and method. If the model target were mouse instead, Record A would not establish the mouse input; additional suitable evidence or a clearly labelled model-dependent estimate would be needed.
Decision note: “Selected rat brain \(K_p=3.15\) from the documented hypothetical steady-state rat measurement and explicit density conversion. Mouse review value \(4.2\) excluded from the rat input because species differs and original measurement details are unavailable. Verify Record A’s original source location before final release; preserve both entries and this rationale in the dossier change history.”
Common Audit Failures and Checks
A final audit should look for errors that can survive an apparently tidy table. Check units by dimensional analysis, not by visual plausibility. Recalculate transformations from retained source values rather than trusting a copied result. Confirm that qualifiers and measured endpoints have not been silently changed. For ratios, verify the tissue and reference matrix. For conflicts, confirm that the records are comparable before deciding that one is wrong.
- Wrong reasoning: “The clearance value has a published citation, so it is traceable.” Correction: follow the citation to the underlying measurement and retain its location, assay context, and any secondary-source steps.
- Wrong reasoning: “The conversion reproduces 480 mL/min, so the entry is human systemic clearance.” Correction: correct arithmetic does not change the measured quantity. Label the scaled intrinsic-clearance estimate accurately and document what it does not represent.
- Wrong reasoning: “Both tissue ratios are published, so their average is the best input.” Correction: compare species, measurement basis, and conditions first. If they do not estimate the same target, averaging has no clear interpretation.
- Wrong reasoning: “The assay detail is missing, but a standard protocol probably applies.” Correction: record the detail as unresolved and seek the original source; do not convert a guess into provenance.
- Wrong reasoning: “A broad numerical range is a confidence interval.” Correction: state whether a range is reported, calculated as a bound, or produced by a formal uncertainty analysis.
These checks have limits. An audit can identify missing information, inconsistent units, and unsupported decisions, but cannot recover unreported assay conditions or prove that a study population represents the intended model population. A well-documented record can still be uncertain or inapplicable. Its value is that those limitations remain visible and can be revisited when better evidence becomes available.
Check Your Understanding
Answer each question, then compare with the concise solutions below.
- A dossier gives an intrinsic clearance of \(5.0\ \mu\text{L}/\text{min}/\text{mg}\), \(50\ \text{mg/g}\) protein per gram liver, and \(1200\ \text{g}\) liver. What scaled value follows, and what quantity should the record avoid claiming without further steps?
- A tissue concentration is \(4.0\ \mu\text{g/g}\), plasma concentration is \(2.0\ \mu\text{g/mL}\), and tissue density is \(1.0\ \text{g/mL}\). Calculate the dimensionless tissue-to-plasma ratio.
- Why is it not defensible to average a rat tissue ratio and a mouse tissue ratio automatically?
- A published value is described only as “about 10.” What should the audit preserve, and what should it avoid doing?
- Two plausible parameter bounds are multiplied to produce a minimum-to-maximum range. Can that range be called a 95% interval? Explain.
Solutions. (1) \(5.0\times50\times1200=300{,}000\ \mu\text{L/min}=300\ \text{mL/min}\). Label it as a scaled in vitro intrinsic-clearance estimate; do not call it observed systemic clearance without evidence and additional steps supporting that interpretation. (2) Convert tissue concentration: \(4.0\ \mu\text{g/g}\times1.0\ \text{g/mL}=4.0\ \mu\text{g/mL}\). The ratio is \(4.0/2.0=2.0\), dimensionless. (3) The species differ, so the values may not estimate the same target; compare applicability and evidence before selecting or deriving an input. (4) Preserve the qualifier “about,” the reported value, units if available, and its source location. Do not silently turn it into an exact value or infer missing assay details. (5) No. It is a bound-based scenario unless a formal method establishes interval coverage; record the bounds and how the range was constructed.