Plausibility Audits for PBPK Parameter Sets

Why Audit a Parameter Set?

A PBPK model can be coded correctly and still give misleading predictions if its inputs are implausible, inconsistently defined, or transcribed with the wrong units. A plausibility audit is a structured review of the parameter record before model evaluation. It asks whether each value has a clear meaning and source, falls within a reasonable domain, and fits with the other inputs that describe the same physiology or compound.

The audit is a screening step, not a method for estimating the “true” value. Physiological quantities vary with age, body size, health, posture, and measurement conditions. A value outside a broad reference range should prompt investigation: it might reflect a special population, a different tissue definition, or an error. It should not be rejected solely because it differs from a familiar reference value.

The audit assumes that the model’s intended population and parameter definitions are specified. For example, “liver volume” might refer to an anatomical volume, a tissue mass converted using a density, or an effective model compartment. Those are not automatically interchangeable. Likewise, blood flow may refer to total hepatic inflow or only hepatic arterial flow. The audit must preserve the definitions used by the model rather than compare superficially similar labels.

A Structured Audit

A useful audit proceeds from simple, objective checks toward more contextual judgments. Begin with data integrity: confirm parameter names, units, species, population, and source. Then check hard mathematical domains, such as positive organ mass and fractions between zero and one. Next, compare magnitudes with broad physiological expectations and examine whether quantities with shared definitions can be reconciled. Finally, record what is flagged, why, and what evidence would resolve it.

Organ Volumes and Tissue Composition

Organ volume and tissue composition checks begin with units and definitions. If the model stores tissue mass but a source reports volume, conversion requires a stated density and compatible definitions. A mass in grams must not be compared directly with a volume in liters. For an adult reference subject, organ masses such as liver around 1.5 kg, brain around 1.4 kg, and total kidneys around 0.3 kg are useful only as rough orientation; values depend on the individual and on what the measurement includes.

For a set of mutually exclusive tissue compartments, their masses should sum to the modeled body mass, apart from explicitly represented residual compartments and rounding. If the model uses fractions of body mass, each fraction is dimensionless and lies between zero and one; fractions for a complete, non-overlapping partition sum to one. These checks are valid only when the listed compartments are truly disjoint. Adding an organ mass to a tissue category that already includes that organ creates an apparent excess even if neither input is individually wrong.

Blood Flows: Check Units and Network Meaning

Blood-flow inputs should be compared in compatible units and with attention to how blood moves through the modeled network. Resting cardiac output in a typical adult is often on the order of 4–7 L/min, but body size, activity, clinical state, and other conditions matter. Approximate values such as renal flow near 1.0–1.3 L/min, cerebral flow around 0.7–0.9 L/min, and total hepatic inflow around 1.2–1.8 L/min are broad screening references, not universal acceptance limits.

A common structural problem is treating serial and parallel flows as though they were all independent branches. Blood passing from the gut through the portal vein to the liver is part of hepatic inflow. If a parameter list includes total hepatic flow and also adds portal-drained gut flow as a separate independent systemic outflow, the same flow may be counted twice. Check the diagram and definitions before using a total as a consistency test.

Drug Inputs: Domains, Units, and Source Context

Drug input audits begin with the same basic questions: what is the measured or predicted quantity, what units are expected, and does the value fall within its mathematical domain? For example, molecular weight must be positive and needs units such as g/mol; a fraction unbound is dimensionless and lies between zero and one; and an acid dissociation constant is often represented as pKa, a logarithmic quantity. A numerical value cannot be judged from its magnitude alone if its unit or encoding is unclear.

Do not infer a universal “plausible” range for every chemical property. A wide range can be legitimate across compounds, and values from different methods may not represent the same conditions. Instead, verify the compound identity, assay or prediction method, experimental conditions, and any conversions. The earlier material on binding and blood-to-plasma partitioning provides definitions for those inputs; in an audit, preserve those definitions and check the recorded value and representation before attempting a mechanistic interpretation.

Why the Checks Work—and Where They Stop

Mass closure follows from the definition of a complete partition. If \(M_i\) is the mass of compartment \(i\), and the compartments are mutually exclusive and exhaustive for a subject of mass \(M\), then their masses must add to that subject’s total mass. Defining \(w_i=M_i/M\), division by \(M\) gives the corresponding fraction constraint.

The argument depends on the assumptions stated: the compartments refer to the same subject, use compatible mass definitions, and neither overlap nor omit mass. A numerical total near one does not establish that each compartment is correct; opposing errors can cancel. Similarly, a flow sum can diagnose a mismatch only after the relevant flows have been defined as independent branches over the same interval.

A plausibility audit is complementary to, rather than a substitute for, other approaches. A reference-range comparison is quick but may overlook unusual populations and definitional differences. A formal mass-balance or network reconciliation is more informative when the model structure is explicit, but it cannot repair inaccurate source data automatically. Expert review can interpret context that a simple rule misses, though it should be documented so the decision is reproducible. For uncertain inputs, retain the evidence and uncertainty rather than changing values merely to make the model look tidy.

Diagnostics and Common Pitfalls

Several patterns signal that an audit is not yet reliable: the same parameter appears with different units in different files; fractions exceed one or are entered as percentages without conversion; organ totals are far above or below the subject’s body mass; flows appear to exceed cardiac output; or a value has no traceable source or population definition. These are reasons to pause, inspect the raw record, and ask whether the discrepancy is a unit, definition, topology, or transcription issue.

• Rejecting a value just because it is outside a familiar range. A reference value may not fit the modeled age, body size, or clinical state. Confirm context before rejecting it.
• Adding overlapping compartments. A tissue category may already include an organ or its blood content. Establish a non-overlapping partition before interpreting its total.
• Comparing unlike flow definitions. Total hepatic inflow and hepatic arterial flow are not interchangeable. Check the network before summing.
• Confusing a percentage with a fraction. Eight percent is 0.08 as a fraction, not 8.0.
• Treating an audit flag as a correction. A flag identifies a question. Any change needs a defensible source or explicit rationale.