Additional Capabilities for 4 Parameter Logistic Bioassay

· 4 min read

On this page
  1. Bioassay — Four-Parameter Logistic Method
  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 — Four-Parameter Logistic Method

Capability added: Optional ANOVA / Confidence Interval Error Term ("Sequential" vs. "Pure Error")

 

What Was Added

The Four-Parameter Logistic Method gains the same kind of optional error-term convention already added to the Parallel Line and Slope Ratio methods, adapted to this method's own model structure. It controls which variance estimate feeds all three validity F-tests and the potency confidence interval together (rather than only the confidence interval, as in the two linear methods) — reflecting how this method's ANOVA is structured. The fitted curve parameters and the potency point estimate are unaffected either way.

Regulatory and Statistical Background

A four-parameter logistic (4PL) assay fits a sigmoidal curve — governed by an upper asymptote, a lower asymptote, a slope factor, and an inflection point — to each preparation's response against log-dose. Three nested models are compared to build the validity tests: a null (flat-line) model, a model where all preparations share one common curve shape (only the inflection point may differ, which is what generates the potency estimate), and a fully separate curve fitted independently to each preparation.

As with the other bioassay methods, European Pharmacopoeia (Ph. Eur.), Chapter 5.3 — "Statistical Analysis of Results of Biological Assays and Tests"-compliant commercial software conventionally builds all three validity F-tests, and the potency confidence interval, from a single shared error variance term — its "Pure-error ANOVA" convention. This tool's existing default instead gives each F-test its own, more locally appropriate error term.

The Mathematics

Convention A — "Sequential" (this tool's existing default). Each of the three F-tests is built from the specific pair of nested models it is testing between, using that comparison's own residual variance as the denominator:

F(Regression) = MS(explained by shared curve vs. flat) / MS(residual of the shared-curve fit)

F(Non-parallelism) = MS(explained by separate curves vs. shared curve) / MS(residual of the separate-curves fit)

F(Lack of Fit) = MS(explained by saturated fit vs. separate curves) / MS(residual of the saturated fit, i.e. pure error)

and the potency confidence interval likewise uses the shared-curve model's own residual — pooling in any small, non-significant amount of non-parallelism.

Convention B — "Pure Error" (the new option; EP 5.3's "Pure-error ANOVA"). All three F-tests, and the potency confidence interval, instead share one single denominator throughout — the residual of the fully saturated, one-curve-per-preparation/dose-group fit:

F(any test) = MS(explained by that test's comparison) / MS(pure error, from the saturated fit) — same denominator every time

Reasoning: Why Two Conventions, and When to Use Each

The two conventions answer slightly different questions, exactly as for the two linear methods: the sequential convention lets each test's significance be judged against the specific variability relevant to that comparison, while the pure-error convention asks every test the same underlying question — "how does this deviation compare to pure replicate noise, with nothing else pooled in?" — and additionally ties the potency confidence interval to that same, narrowest possible error estimate.

For a well-fitting, valid assay, the difference between the two is usually small. It becomes more noticeable when there is a modest, non-significant amount of non-parallelism or lack-of-fit, since the sequential convention folds that extra variability into the corresponding test's (and the potency interval's) error term, while the pure-error convention deliberately excludes it everywhere.

As with the other methods, selecting "Pure Error" is primarily intended for reconciling results with EP 5.3-compliant commercial software or a legacy validated method built on its Pure-error ANOVA convention. The tool's existing sequential default remains a fully valid choice where no such external comparison is required.

What Changes in the Output, and What Does Not

Affected:

  • F-value and P-value for all three validity tests (Regression, Non-parallelism, Lack of Fit)
  • A new "Error Term" column in the ANOVA table identifying which convention and degrees of freedom were used
  • Lower CL and Upper CL of the potency confidence interval, and a new "CI Error Term" column

Not affected:

  • The fitted curve parameters (upper/lower asymptote, slope factor, inflection point) for every preparation
  • The estimated relative potency ratio
  • The fitted model equation shown for each preparation

Where to Find This Capability

On the Four-Parameter Logistic Method tool's input form, a field labelled "ANOVA / CI error term" offers a choice between "Sequential — each F-test/CI uses its own nested residual (this tool's default)" and "Pure error — single saturated-model residual for all F-tests and the CI (EP 5.3 default)". The pure-error option requires the saturated (one-fit-per-preparation/dose-group) model to have at least one residual degree of freedom, i.e. some replication in the data.

fpl-add-option-bioassay


Verification Against Published Reference Data

This option was checked against a published serological assay of tetanus antitoxin, comparing a standard and one test serum preparation across ten two-fold dilution steps, two replicates per dose.

Source: European Pharmacopoeia, 5th Edition (2005), Chapter 5.3, worked example 5.4.1.a — as reproduced in Stegmann Systems' PLA 3.0 published example report (Document-3567, "PLA_3.0.7_Example_Parallel-Logistic_Quantitative_Response_Assay_Standard_Report.pdf")

Metric

Published (PLA/EP 5.3)

Computed (Pure Error option)

Regression F (df=3)

9785.4821

9788.5262

Non-parallelism F (df=3)

0.48912

0.4884

Lack of Fit F (df=12)

0.62266

0.6216

Potency ratio

1.45886

1.45880 (rounded: 1.4588)

Estimated sample potency (IU/ml)

0.58354

0.5835

95% CI (sample potency, IU/ml)

0.55511 – 0.61343

0.5544 – 0.6127

 

The curve-shape parameters (upper/lower asymptote, slope factor, inflection points) and the potency point estimate matched the published report almost exactly under both conventions, confirming the fit itself is unaffected by this option, as intended. The F-values and confidence interval computed with the pure-error option are close to the published figures, with a small residual difference (visible mainly in the confidence interval) attributable to the published report rounding its intermediate values to three decimal places before final calculation. The tool's existing sequential default was, as expected, not intended to match this "Pure-error ANOVA" report and was verified separately to remain internally consistent.

References