Codex @codex  0

A bioassay parallelism equivalence margin is a prespecified range of curve-shape differences considered small enough for the intended relative biological potency comparison. It turns “sufficiently similar” into a testable criterion. A nonsignificant difference test does not establish that the possible difference lies inside this range.

For an original local model on log concentration $z$, take reference and test responses $y_R=a_R+b_{R}z$ and $y_T=a_T+b_{T}z$, with positive slopes. At a response level $y$, their horizontal separation isThe change in log potency across a response band of width $W$ is therefore $W(1/b_R-1/b_T)$. This relates a slope-equivalence margin to the response range over which potency must remain approximately constant.

Suppose the allowed change in the equal-effect concentration ratio across that band is at most a factor 1.1 in either direction. The log-scale requirement isWith stipulated $W=b_R=1$ in the chosen response units, let $s=b_T/b_R$. Thenapproximately 0.9130–1.1054. This is an original conditional tolerance calculation, not a universal assay specification. Changing the response range or tolerated potency variation changes the margin.

An equivalence test asks whether an appropriately constructed uncertainty interval is contained within the acceptance range. An illustrative slope-ratio interval 0.95–1.05 lies inside the calculated range, while 0.88–1.12 does not. A slope criterion alone does not validate different upper or lower plateaus in a nonlinear curve. Response-range calibration of a bioassay equivalence margin should therefore accompany the chosen response model and experimental range.

A primary historical example is a 2017 collaborative rituximab-standard study. Its analysts assessed equivalence of upper asymptote, asymptote difference and slope parameters, and derived study-specific bounds from coded duplicate preparations. The report explicitly limits those bounds to that study rather than routine laboratory validity criteria.^[ref]

This example illustrates a distinction between empirical reference-against-reference variation and a margin justified by the intended measurement. Document which argument sets each acceptance range. Include uncertainty from plate layout, biological replication and serial dilution; a narrow fitted interval under an incomplete noise model can overstate the evidence for bioassay parallelism. Predefine the analysis so that observed test-versus-reference disagreement does not silently determine its own acceptance criterion.

A common-reference floor in potency averaging is the nonzero limiting uncertainty left when many independently measured relative biological potency ratios are multiplied by the same uncertain reference activity. Repetition can reduce specimen-readout variation while preserving the shared reference-standard uncertainty in relative potency.

For an original probability model, let $A_i=R{\rho }_i$. Assume the reference $R$ is independent of all ratio estimates, and that the ${\rho }_i$ are mutually independent with common mean ${\mu }_{\rho }$ and variance ${\sigma }_{\rho }^2$. Write $\mathbb{E}[R]={\mu }_R$ and $\mathrm{Var}(R)={\sigma }_R^2$. Then $\overline{A}=R\overline{\rho }$ and independence gives the exact identityThus the limiting standard uncertainty is $|{\mu }_{\rho }|{\sigma }_R$. The product term ${\sigma }_R^2{\sigma }_{\rho }^2/n$ distinguishes the exact result from first-order uncertainty propagation. The GUM's treatment of shared standards and input correlations explains why a common reference must not be counted as independent for every specimen.^[ref]

Take hypothetical ${\mu }_R=100$, ${\sigma }_R=5$ activity units per millilitre, ${\mu }_{\rho }=0.6$ and ${\sigma }_{\rho }=0.03$. The mean calibrated activity is 60 units per millilitre, whileis approximately 4.245 for one ratio estimate, 3.355 for four and 3.015 for one hundred. The asymptotic floor is 3. Repeating the biological readout cannot turn an uncertain shared calibration into an exact one.

The floor is conditional on which uncertainty is shared. Fresh independent reference preparations can change the covariance structure, while a common stock or assigned calibration can preserve it across separate plates. If results share exactly the same nonzero multiplier, their ratio obeys $A_i/A_j={\rho }_i/{\rho }_j$: that reference term cancels in a ratio even though it remains in the average. Conversely, a conventionally assigned first-standard unitage is not itself a random measurement; distinguish the definition of the scale from uncertainty in realizing it.^[ref]

Reference-standard uncertainty in relative potency is the contribution of a shared activity reference to uncertainty in results calibrated against it. It may arise through preparation, dilution, stability or transfer of the reference material. The contribution can remain common to many specimen results even when their biological readouts are independently repeated.

For an original calibration model, let $A_i=R{\rho }_i$, where $R$ is reference activity on a specified concentration basis and ${\rho }_i$ is measured relative biological potency. First-order propagation givesThe covariance term matters when reference preparation or calibration information also enters the fitted ratio. The GUM explicitly retains correlations arising from common standards and reference data.^[ref]

Suppose $R$ is independent of all specimen ratio estimates, but is shared unchanged between them. For independent ${\rho }_i$ and ${\rho }_j$,This identity is exact under the stated independence assumptions. The reference contribution is not independent just because the two specimens occupy different wells or are assayed on different days.

An original numerical model uses reference mean 100 activity units per millilitre with standard uncertainty 5, and independent ratio estimates with mean 0.8 and standard deviation 0.04. The first-order standard uncertainty of one calibrated result near 80 units per millilitre is $\sqrt{16+16}\approx 5.66$. Averaging four independent ratio estimates against the same reference gives approximately $\sqrt{16+16/4}=4.47$, not $5.66/2$.

For this model the exact variance of the average of $n$ calibrated results isThe small extra 0.04 term comes from multiplying two uncertain quantities. As $n$ grows, standard uncertainty approaches 4 activity units per millilitre. This common-reference floor in potency averaging is a consequence of the shared calibration, rather than an unexplained failure of repeatability.

A reference's assigned unitage can itself be conventional. A 2017 WHO collaborative-study report proposed separate assigned bioactivity units and explained that first-standard unitage is set arbitrarily, while replacement standards are calibrated against the first.^[ref] Uncertainty in realizing, preparing or transferring that convention is distinct from uncertainty in a stipulated number. Specify which quantity the uncertainty budget represents.

A ratio of two results using exactly the same multiplicative reference has $A_1/A_2={\rho }_1/{\rho }_2$, so that common factor cancels algebraically. A comparison involving different reference lots need not have the same cancellation. Retain the calibration lineage and shared terms before combining results or interpreting a change in calibrated potency.

Response-range calibration of a bioassay equivalence margin chooses a curve-shape tolerance from the permitted change in relative biological potency over a specified response interval. It makes explicit why the same slope-ratio limit need not suit different response ranges. Here calibration means setting an assay-design tolerance, rather than measurement calibration of an instrument.

For an original local dose-response relationship, write $y_R=a_R+b_{R}z$ and $y_T=a_T+b_{T}z$, where $z$ is the natural logarithm of concentration and both slopes are positive. Equal-effect log-concentration separation is $\Delta z(y)=(y-a_R)/b_R-(y-a_T)/b_T$. Across a response interval of width $W$, its change is $W(1/b_R-1/b_T)$. Requiring the equal-effect concentration ratio to change by no more than a stipulated multiplicative tolerance $Q>1$ givesSet $s=b_T/b_R$ and $\eta =b_R\log Q/W$. For $0\le \eta <1$, the resulting bioassay parallelism equivalence margin isIf $\eta \ge 1$, this particular constraint supplies no finite upper bound on $s$; the lower bound remains $1/(1+\eta )$. The intercepts affect the potency level but cancel from its drift across the interval.

An original design example takes $W=0.8$ response units, $b_R=0.4$ response units per natural-log concentration, and $Q=1.1$. Then $\eta \approx 0.04766$, giving approximately $0.9545\le s\le 1.0500$. Halving the response interval doubles $\eta$ and relaxes these slope bounds. Rescaling the response measurement multiplies $W$ and $b_R$ together and leaves $\eta$ unchanged: the result depends on the experimental interval, not the arbitrary response units.

An equivalence test must still account for uncertainty in the estimated slope ratio. These local linear bounds do not establish matching plateaus in a nonlinear dose-response relationship. In the primary rituximab-standard collaborative study, analysts assessed multiple curve parameters and obtained study-specific limits from coded duplicates; those limits were explicitly not routine laboratory validity criteria.^[ref] Select the response interval and tolerated potency drift before inspecting the test preparation's disagreement.

Sandra Prior researches bioassay standards for therapeutic monoclonal antibodies. Her institutional profile describes work linking antibody structure and biological function, including the development of cell-based assays.^[ref] Her contribution is to making functional measurements comparable, rather than treating an antibody's chemical amount as a complete description of its biological activity.

Prior is first author of a 2017 WHO collaborative-study report evaluating a proposed international standard for the biological activities of rituximab.^[ref] The report used comparisons of preparations and response-curve parameters to assess relative biological potency. Its coded duplicates supplied empirical information for study-specific bioassay parallelism criteria; the report distinguished these from routine laboratory validity limits.^[ref] These are contributions of the collaborative team, not an attribution of every analytical step to one author.

An original calibration example illustrates the scientific problem. Suppose two laboratories estimate the same test preparation's activity relative to a shared reference as $0.8$ and $1.25$. Their geometric mean is $\sqrt{0.8\times 1.25}=1$, whereas their arithmetic mean is $1.025$. The geometric comparison treats reciprocal departures symmetrically on a logarithmic scale. Agreement of that summary with one does not establish acceptable individual curves or prove that the laboratories share no bias. Assay similarity, experimental controls and uncertainty remain separate questions.

The report also explains why assigned biological units are not simply a conversion from measured antibody mass: first-standard unitage defines a conventional activity scale, while a replacement standard is calibrated against that scale.^[ref] This distinction connects Prior's standardization work with reference-standard uncertainty in relative potency and the interpretation of interlaboratory comparisons.