Additional Capabilities for Combine Assays Bioassay

· 5 min read

On this page
  1. Bioassay — Combine Assays
  2. What Was Added
  3. Regulatory and Statistical Background
  4. The Mathematics
  5. Reasoning: Why Two Conventions, and When to Use Each
  6. What Changes in the Output, and What Does Not
    1. Affected:
    2. Not affected:
  7. Where to Find This Capability
  8. Verification Against Published Reference Data
  9. References

Bioassay — Combine Assays

Capability added: New Combination Method: "EP 5.3, 4th–9th Edition Legacy (Heterogeneously Weighted)"

 

What Was Added

The Combine Assays tool — which pools potency estimates and confidence intervals from several independent assay runs into a single combined result — gains a third combination method, alongside its existing two. The new method reproduces an older, specific formula from a widely-used pharmacopoeial framework, added so that combined results can be reconciled with software and legacy records that still use that formula.

Regulatory and Statistical Background

Combining several independent assay runs into one overall potency figure requires an estimate of two separate sources of uncertainty: the precision of each individual assay run (intra-assay variance, already known from each run's own confidence interval), and the run-to-run variability between assays (inter-assay variance). European Pharmacopoeia (Ph. Eur.), Chapter 5.3 — "Statistical Analysis of Results of Biological Assays and Tests", across several editions, has specified more than one formula for estimating that inter-assay variance, and different combination software packages in the market have implemented different editions of it.

Independent bioassay-statistics literature has documented that the formula used in the 4th through 9th editions of Ph. Eur. Chapter 5.3 for this purpose has a known limitation: it does not subtract out the average intra-assay variance when estimating the inter-assay component. As summarised by an independent bioassay CRO's technical documentation reviewing this formula: "The formula in the PhEur is incorrect: it does not account for the intra-assay variation, and is smaller by a factor of N (the number of assays) than it should be... Note this has been fixed in the latest version of the guidance, but the error still exists in older versions." The same source notes the consequence is that the formula "can lead to a severe underestimate of the inter-assay variation, which will lead to a CI for the combined RP that is too narrow."

Despite this documented limitation, the older formula remains in active use because several established, EP 5.3-compliant commercial software packages implement it by default and have not been updated to the corrected version, and many legacy assay records and regulatory submissions were built using it. The new method reproduces that specific, older formula — clearly labelled as a legacy option — precisely so that a combined result can be reconciled with those existing records rather than silently disagreeing with them for an undocumented reason.

The Mathematics

All three combination methods available in this tool work on the natural logarithm of each assay's potency estimate, and all three build a weighted average of those log-potencies, differing only in how the weights and the inter-assay variance are computed.

The existing "USP <111>" and "EP 5.3" methods use a heterogeneity-adjusted (DerSimonian-Laird style) inter-assay variance, which explicitly subtracts the pooled intra-assay variance before attributing any remaining spread to genuine inter-assay variability, and floors the result at zero.

The new "EP 5.3, 4th–9th edition legacy (heterogeneously weighted)" method instead estimates the inter-assay variance directly from the spread of the individual assays' log-potencies alone, without subtracting intra-assay variance:

inter-assay variance = sample variance of the assays' log-potencies / number of assays

weight (per assay) = 1 / (that assay's own log-potency variance + inter-assay variance)

combined log-potency = (sum of weight x log-potency, across assays) / (sum of weights)

This weighting is applied unconditionally under the legacy method — unlike the "EP 5.3" method, which first tests whether the assays are statistically homogeneous and switches its weighting scheme depending on the result. The combined confidence interval's degrees of freedom are also computed differently under the legacy method: the sum of every individual assay's own residual degrees of freedom, rather than (number of assays - 1).

Reasoning: Why Two Conventions, and When to Use Each

The legacy method is not being introduced as a recommended default, and the tool continues to default to the existing (corrected) USP method. It is offered specifically for situations where matching an existing, already-accepted result matters more than using the statistically preferable current formula — for example, reproducing a combined potency figure from a validated legacy assay record, or cross-checking against commercial software that has not adopted the corrected guidance.

Because the underlying formula is documented to understate inter-assay variability, any confidence interval produced by the legacy method should be interpreted with that limitation in mind — it is expected to run narrower than the interval the same data would produce under the "USP" or "EP 5.3" methods, precisely because it omits the intra-assay variance subtraction those methods perform.

What Changes in the Output, and What Does Not

Affected:

  • The combined potency confidence interval and its degrees of freedom, when the legacy method is selected
  • The reported inter-assay variance figure
  • The homogeneity-test result wording (the legacy method applies its weighting unconditionally, so the wording notes this rather than describing a homogeneity-based switch)
  • Two new comparison columns: "Legacy (PLA/EP 5.3) Mean Potency" and "Inter-Assay Variance (legacy)", shown alongside the existing weighted/semi-weighted/unweighted mean potency comparison for context

Not affected:

  • The combined potency and confidence interval produced by the existing "USP <111>" and "EP 5.3" methods
  • The per-assay input table, and the forest plot of individual assay results
  • The homogeneity chi-square and Bartlett's test statistics themselves (only how the wording describes their consequence changes)

Where to Find This Capability

On the Combine Assays tool's input form, the "Combination method" field gains a third choice: "EP 5.3, 4th-9th edition legacy (heterogeneously weighted)", alongside the existing "USP <111> (semi-weighted throughout)" and "EP 5.3 (weighted / semi-weighted / unweighted, auto-selected by homogeneity)".

ca-add-option-bioassay


Verification Against Published Reference Data

This method was not implemented directly from a formula written down in the pharmacopoeia text. It was reverse-engineered by testing candidate formulas numerically against a published worked example until an exact match was found, then implemented and re-verified. The example combines five independent assay runs into a group-level potency for two separate groups of test material.

Source: European Pharmacopoeia, 4th–9th edition, Chapter 5.3, "Combination type: Weighted — CI based on inter- and intra-assay variation" — as reproduced in Stegmann Systems' PLA 3.0 published example report (Document-3591, "PLA_3.0.7_Example_Combination_Of_Assay_Results_Standard_Report.pdf")

Group

Metric

Published (PLA/EP 5.3 legacy)

Computed (legacy method)

1

Combined potency

1.02417

1.0242

1

95% CI

1.00231 – 1.04650

1.0023 – 1.0465

1

Inter-assay variance

6.109E-05

6.1E-05

2

Combined potency

1.00738

1.0074

2

95% CI

0.98277 – 1.03261

0.9828 – 1.0326

2

Inter-assay variance

2.537E-04

2.54E-04

 

Both groups' combined potency, confidence interval, and inter-assay variance matched the published report to the precision shown. This match is a stronger check than usual, since the formula was derived by fitting to this exact data — for that reason, the result should be read as confirming the formula was correctly reverse-engineered and implemented, rather than as an independent validation on unseen data. The existing "USP" and "EP 5.3" methods were confirmed unaffected on the same input, continuing to produce their own (different, corrected-formula) combined confidence intervals as before.

References