Chapter 1: Statistical Interpretation of the 1% Failure Rule

The 1% failure rate recommended by the FDA defines a probabilistic design condition between the Screening Cut Point (SCP) and the Low Positive Control (LPC).
This relationship can be formalized and interpreted in terms of two independent random variables.

1. Probabilistic Definition

Let the assay signal for negative samples be \(Y_{SCP}\) and for the low-positive control be \(Y_{LPC}\). The cut point \(CP\) represents the common decision boundary between them.

\[ P(Y_{SCP} \le CP) = 0.99, \quad P(Y_{LPC} \le CP) = 0.01 \]

These two statements describe the same threshold from opposite perspectives:

  • From the SCP side, 99% of negative samples are below the cut point.

  • From the LPC side, only 1% of positive samples fall below it — defining the 1% LPC failure probability.

2. Standardized Representation

When both distributions are assumed approximately normal,

\[ Z_{SCP} = \frac{CP - \mu_{SCP}}{\sigma_{SCP}} = z_{0.99} = 2.326 \] \[ Z_{LPC} = \frac{CP - \mu_{LPC}}{\sigma_{LPC}} = z_{0.01} = -2.326 \]

Thus, the expected LPC mean is roughly \(z_{0.99}\) standard deviations above the cut point when both distributions share similar variability.

3. Variance Combination Principle

In practice, SCP and LPC are independent sources of variability.
When evaluating the probability that an LPC measurement falls below CP, we consider the difference:

\[ X = Y_{LPC} - CP \]

Then:

\[ Var(X) = Var(Y_{LPC}) + Var(CP) - 2\,Cov(Y_{LPC}, CP) \]

Since \(Cov(Y_{LPC}, CP) = 0\), it follows that:

\[ Var(X) = \sigma_{LPC}^2 + \sigma_{SCP}^2 \]

Therefore, the 1% failure rate is defined under joint uncertainty from both distributions, not from SCP or LPC alone.

4. Combined Expression for LPC Mean

To ensure a 1% failure rate:

\[ P(X \le 0) = 0.01 \Rightarrow \frac{0 - (\mu_{LPC} - CP)}{\sqrt{\sigma_{LPC}^2 + \sigma_{SCP}^2}} = z_{0.01} \]

Rearranging yields:

\[ \boxed{ \mu_{LPC} = CP + z_{0.99}\,\sqrt{\sigma_{SCP}^2 + \sigma_{LPC}^2} } \]

This is the fundamental LPC setting equation, showing that the LPC mean must exceed the cut point by a margin proportional to the combined analytical variance.

5. Simplified Practical Form

In most validation settings, \(\sigma_{LPC}\) is unknown because the LPC material does not yet exist when the cut point is determined.
Hence, it is set to zero for practical estimation:

\[ \mu_{LPC} \approx \mu_{SCP} + k\,\sigma_{SCP} \]

where k depends on available data and conservativeness:

k value Statistical basis Typical context Approx. effect
\(z_{0.99}=2.326\) Normal, n ≥ 6 Standard practice ~1% fail
\(t_{0.99,df=n-1}\) Limited data Conservative Slightly higher LPC
\(t_{0.95,df=n-1}\) Small n or critical assays Very conservative <1% fail

6. Interpretation and Limitations

The theoretical 1% failure rate assumes that:

  1. SCP and LPC are normally distributed and independent,
  2. The SCP estimate (\(\mu_{SCP}, \sigma_{SCP}\)) remains stable over time, and
  3. Analytical precision at the LPC level is comparable to that at the SCP level.

In real assay runs, deviations from these assumptions are common, i.e., \(\sigma_{SCP} \ne \sigma_{LPC}\).
Therefore,

\[ \begin{aligned} P(\mathrm{LPC} \le CP) &= \Phi\!\left( \frac{CP - \mu_{\mathrm{LPC}}}{\sigma_{\mathrm{LPC}}} \right) \\[6pt] &= \Phi\!\left( \frac{ \mu_{\mathrm{SCP}} + z_{0.99}\sigma_{\mathrm{SCP}} - (\mu_{\mathrm{SCP}} + z_{0.99}\sigma_{\mathrm{SCP}}) - z_{0.99}\sigma_{\mathrm{SCP}} }{ \sigma_{\mathrm{LPC}} } \right) \\[6pt] &= \Phi\!\left( -\; z_{0.99} \frac{\sigma_{\mathrm{SCP}}}{\sigma_{\mathrm{LPC}}} \right) \end{aligned} \]

It indicates that

  • If LPC variance is higher than that of SCP, actual failure rates may exceed 1%.

  • If SCP drifts upward over time, the empirical failure rate may drop below 1%.

Therefore, the 1% rule should be viewed as a statistical design target, not as an empirical constant.

7. Bridge to Practical Evaluation

In many titer-based ADA assays, the SCP is converted from signal to concentration units using a linear calibration model.
Then, a one-sided \(t_{0.99}\) is applied on the concentration scale to set the LPC concentration.

While convenient, this practice implicitly assumes equal variance across the calibration domain —
an assumption that does not always hold, particularly when dilutional nonlinearity or background noise affects low-concentration regions.

In the next chapter, we will analyze how this conversion impacts the effective failure probability,
and how to interpret or adjust LPC design when the signal–concentration relationship is not strictly linear.