Resolving Conflicting Preclinical and Clinical Evidence

Why Conflicting Evidence Needs an Audit Before a Reconciliation

Preclinical and clinical dose–response data can appear inconsistent even when each source is internally coherent. One source may report nominal concentration while another reports total plasma concentration; the sources may differ in unbound fraction, endpoint definition, sampling time, or assay scale. A fitted curve can absorb such differences into a potency parameter, but that does not establish that the biological potency actually changed.

The practical task is to make the comparison explicit: identify what concentration and response each observation represents, state the assumptions that connect sources, and calculate what each plausible reconciliation predicts. The aim is not to force agreement. If unit conversion and measurement checks do not account for the discrepancy, the remaining difference is evidence that the assumed cross-scale relationship may be inadequate.

Set Up the Comparison on a Common Scale

Consider a hypothetical preclinical assay and clinical biomarker analysis for the same drug. Both report inhibition as a percentage of a common reference response, with larger values meaning greater inhibition. For illustration, suppose the response follows a simple saturating concentration–response relationship:

Here \(R(C)\) is inhibition in percentage points at the concentration \(C\), \(E_{\max}\) is the maximum inhibition in percentage points, and \(EC_{50}\) is the concentration producing half of \(E_{\max}\). This is a descriptive model for the examples, not a claim that every QSP response follows this curve.

A direct comparison requires more than matching the printed concentration units. It assumes the concentration has the same meaning in both sources, the response is on a comparable scale and direction, and the concentration is relevant to the response measurement at the stated time. It also assumes the chosen response model is adequate over the compared range. If any of these assumptions is doubtful, test it before attributing the discrepancy to a potency shift.

For the examples below, assume the clinical endpoint has been mapped to the same percentage-inhibition scale as the assay, and that its observations are paired with the relevant concentration measurements. These are hypothetical data constructed to illustrate the calculations; they are not results from a real study.

Compare the Clinical Evidence and Candidate Explanations

Suppose the clinical data report total plasma concentrations of 20, 50, and 100 nM, with inhibition of 40, 57.1, and 66.7 percentage points. To compare these with the preclinical unbound-scale potency, suppose the clinical unbound fraction is estimated as \(f_{u,\mathrm{clinical}}=0.50\). If that estimate is correct and potency transfers unchanged, clinical unbound concentrations are 10, 25, and 50 nM. Applying the preclinical \(EC_{50,u}=20\) nM gives predictions of 26.7, 44.4, and 57.1 percentage points, respectively.

Those predictions are below the clinical observations at all three concentrations. This is a systematic discrepancy, not a unit conversion by itself. Several explanations remain possible: the clinical concentration is not represented by the assumed total-to-unbound mapping; clinical potency differs from preclinical potency; or the response measurements are not actually on the same scale. Each explanation implies different evidence to check.

One useful way to write the concentration mapping is \(C_u=qC_{\mathrm{total}}\), where \(q\) is the effective fraction of the reported clinical total concentration represented as relevant unbound concentration. Under the unchanged-response model,

Consequently, a curve observed against total concentration identifies the ratio \(EC_{50,u}/q\), not \(EC_{50,u}\) and \(q\) separately. A better curve fit cannot, on its own, decide whether exposure mapping or potency is responsible.

Two Reconciliations Can Fit but Mean Different Things

The observed clinical curve has an apparent \(EC_{50}\) of 20 nM when plotted against total plasma concentration: at 20 nM, the response is 40 percentage points, half of the assumed \(E_{\max}=80\). There are at least two ways to reproduce that curve while retaining the same saturating response form.

• Exposure-mapping reconciliation: set \(q=1.00\), with \(EC_{50,u}=20\) nM. This treats each reported total concentration as the relevant concentration scale.
• Potency-shift reconciliation: retain \(q=0.50\), but set the clinical \(EC_{50,u}=10\) nM. This retains the assumed unbound fraction while allowing a twofold lower clinical unbound \(EC_{50}\).

Both choices give the same total-concentration \(EC_{50}\): the first gives \(20/1=20\) nM, and the second gives \(10/0.50=20\) nM. They therefore make identical predictions at every clinical total concentration under this model. The first changes the exposure mapping; the second changes the potency parameter. This exact equivalence means these clinical concentration–response observations alone cannot distinguish the explanations.

Check the Measurements That Could Break the Tie

When alternatives fit equally well, the next step is not to choose the more appealing explanation. It is to look for measurements that inform the assumptions on which they differ. In this example, a direct assessment of the clinical unbound fraction under relevant conditions bears on \(q\). An independent preclinical or clinical potency measurement on a matched unbound scale bears on \(EC_{50,u}\). Repeated or orthogonal assessment of the response mapping can reveal whether a shared percentage-inhibition scale is defensible.

Audit the data and model for practical failure modes before interpreting either reconciliation:

• Unit errors: confirm concentration units and prefixes. For example, \(0.050\) µM is 50 nM; treating it as 0.050 nM creates a thousandfold error.
• Different concentration meanings: record whether each value is nominal, total, or unbound, and which matrix or assay conditions inform the conversion. A fraction measured in one matrix is not automatically valid in another.
• Response-scale mismatch: check baseline, direction, denominator, and whether the endpoint is a percentage, a change from baseline, or a transformed measure. Matching labels do not guarantee matching scales.
• Timing mismatch: establish whether the concentration measurement represents the exposure relevant to the response. A single concentration may not summarize a response driven by a preceding time course.
• Systematic residual patterns: inspect whether discrepancies change with concentration, time, study, or protocol. A constant direction may signal a scale offset; a concentration-dependent pattern may indicate that the chosen curve or conversion is inadequate.

These checks help distinguish a measurement conflict from model misfit, but they cannot guarantee that the assumed Emax form or cross-scale mapping is correct. A good fit to the available points is not proof of a unique biological explanation. Calibration conclusions should remain conditional on the data, endpoint mapping, and exposure assumptions used.

Comparison and Limitations

A limitation of this comparison is structural: because \(EC_{50,u}\) and \(q\) enter through their ratio on the total-concentration scale, these data cannot estimate both independently without additional information or constraints. More flexible reconciliation choices may improve apparent agreement while adding assumptions that the observations do not identify. The appropriate conclusion is therefore a set of conditional predictions and the evidence needed to discriminate among them, not an unqualified declaration that one source is correct.