Tutorials › Bioavailability and Bioequivalence › Bioequivalence Analysis With BLQ Concentrations

Home  /  Tutorials  /  Worked Bioequivalence Analyses

Worked Bioequivalence Analyses · Tutorial 372 of 399

Bioequivalence Analysis With BLQ Concentrations

Apply a prespecified rule for BLQ concentrations, quantify its impact on AUC, and assess how that impact can alter a crossover bioequivalence conclusion.

Intermediate 12 min read

What You'll Learn

  • Distinguish BLQ observations from zero concentrations and ordinary missing values.
  • Apply a stated BLQ handling rule to a concentration-time profile.
  • Calculate how alternative handling changes a trapezoidal AUC estimate.
  • Analyze paired log exposure ratios and interpret a 90% confidence interval.
  • Identify diagnostics and limitations when BLQ results are concentrated in one treatment.

Why BLQ Handling Matters

A concentration reported as below the lower limit of quantification (LLOQ) is not a measured zero. It indicates that the assay did not quantify the concentration reliably below a stated threshold. Nevertheless, exposure measures such as AUC require a numerical concentration-time profile. The rule used to handle BLQ observations can therefore affect AUC, the test-to-reference comparison, and ultimately the bioequivalence (BE) decision.

BLQ handling is not a cosmetic data-cleaning choice. The rule should be prespecified, applied consistently to both treatments, and interpreted in light of the assay and sampling schedule. Different rules can yield different exposure estimates, especially when BLQ observations occur between quantifiable samples or occur more often for one treatment.

Definition: A BLQ concentration is a sample concentration below the assay's lower limit of quantification (LLOQ). It is an interval-censored observation: the concentration is below the LLOQ, but its exact value is not known. A BLQ result is not, by itself, evidence that the true concentration is exactly zero.

This tutorial uses one explicit noncompartmental handling rule for illustration: set predose BLQ values and BLQ values before the first quantifiable postdose concentration to zero; treat an internal BLQ value, occurring between quantifiable concentrations, as missing and calculate the area across that time using the surrounding quantifiable observations; and set BLQ values after the last quantifiable concentration to zero if those times are included in the defined integration interval. The examples focus on AUC from time zero to the last quantifiable sample, so no post-last-quantifiable area is added.

This is a stated analysis rule, not a universal instruction for every assay, drug, or regulatory setting. The protocol and applicable guidance determine the primary rule. In particular, internal BLQ observations should not be silently replaced by zero or by a fraction of the LLOQ without justification.

From Concentrations to AUC Under a Stated Rule

For observed concentrations \(C_i\) and \(C_{i+1}\) at times \(t_i\) and \(t_{i+1}\), the linear trapezoidal area is \( (t_{i+1}-t_i)(C_i+C_{i+1})/2 \). When an internal BLQ is treated as missing, the neighboring quantifiable observations form the trapezoid across the gap. This calculation does not assert that the actual concentration followed a straight line; it is the specified numerical integration convention.

Worked Example: Integrating a Profile With an Internal BLQ

Suppose a hypothetical participant receives treatment T. The assay LLOQ is 2 mg/L. Samples are collected at 0, 1, 2, 4, and 6 hours, with concentrations BLQ, 8, 12, BLQ, and 6 mg/L, respectively. Estimate AUC from time zero to the last quantifiable sample using the stated rule.

The predose BLQ is set to zero, and the internal BLQ at 4 hours is omitted as a concentration value. The last quantifiable concentration is at 6 hours. The trapezoids are therefore calculated over 0–1, 1–2, and 2–6 hours:

$$ \begin{aligned} AUC_{0-6} &= \frac{1-0}{2}(0+8) + \frac{2-1}{2}(8+12) + \frac{6-2}{2}(12+6)\\ &=4+10+36\\ &=50\ \text{mg}\cdot\text{h/L}. \end{aligned} $$

The 2–6 hour trapezoid spans the internal BLQ time at 4 hours. The AUC is 50 mg·h/L under this rule; the BLQ sample is not interpreted as a concentration of zero.

A practical consequence is that the handling rule must be applied to the same time interval and integration definition for every participant and treatment. If the primary endpoint is AUC to the last quantifiable time, that endpoint should not be confused with an AUC that extends to a scheduled final time after the last quantifiable sample.

How Alternative Rules Change Exposure

Replacing an internal BLQ with zero or with LLOQ/2 can change the trapezoids on both sides of that time. The direction and size of the change depend on the neighboring concentrations and the sampling times. There is no general guarantee that one substitution will move AUC in a particular direction relative to every other possible underlying profile.

Worked Example: Comparing Internal-BLQ Conventions

Use the profile from the previous example. Its main-rule estimate is 50 mg·h/L. Compare that estimate with two illustrative alternatives: substitute zero at 4 hours, or substitute LLOQ/2 = 1 mg/L at 4 hours. The other observations and the integration interval remain unchanged.

Internal BLQ set to zero. The area from 2 to 4 hours is \(2(12+0)/2=12\), and the area from 4 to 6 hours is \(2(0+6)/2=6\). Including the first two areas, which sum to 14 mg·h/L, gives:

$$ AUC_{0-6}=14+12+6=32\ \text{mg}\cdot\text{h/L}. $$

Internal BLQ set to LLOQ/2. The corresponding areas are \(2(12+1)/2=13\) and \(2(1+6)/2=7\), so:

$$ AUC_{0-6}=14+13+7=34\ \text{mg}\cdot\text{h/L}. $$

For this profile, the three estimates are 50, 32, and 34 mg·h/L. The difference is consequential: the substitutions produce estimates 36% and 32% below the main-rule estimate, respectively. Those percentages describe this hypothetical calculation only; they are not a general correction factor for BLQ data.

Setting an internal BLQ to zero makes the profile dip to zero at that time, while LLOQ/2 imposes a particular value that the assay did not measure. Treating it as missing instead uses the neighboring quantifiable samples to span the interval. Each convention encodes a different assumption about the unobserved concentration, so the resulting AUC should be presented as conditional on the chosen rule.

From AUCs to a Crossover BE Decision

For a crossover comparison, exposure is commonly analyzed on the logarithmic scale. For participant \(i\), define the paired treatment contrast \(d_i=\log(AUC_{T,i})-\log(AUC_{R,i})=\log(AUC_{T,i}/AUC_{R,i})\). The mean contrast estimates the log test-to-reference geometric mean ratio. A confidence interval on this scale is exponentiated to obtain an interval for the ratio.

The transparent calculation below uses a paired \(t\) interval. Its assumptions are that participant contrasts are independent across participants, the contrasts are approximately normally distributed, and all AUCs were computed using the same prespecified rule. For the paired calculation to represent the treatment comparison in a simple balanced crossover, period effects and carryover must not distort the contrast. A real study should use its prespecified crossover model when it accounts for sequence, period, or other design terms; the corresponding model-based estimate, standard error, and degrees of freedom take precedence over a simplified paired calculation.

Result: For \(n\) independent paired log contrasts with sample mean \(\bar{x}_d\), sample standard deviation \(s_d\), and \(df=n-1\), the two-sided 90% confidence interval for the mean log contrast is \(\bar{x}_d \pm t_{0.95,df}s_d/\sqrt{n}\). Exponentiating both endpoints gives the 90% confidence interval for the test-to-reference geometric mean ratio. Under the conventional 80%–125% limits, BE is concluded for the endpoint only when the entire interval lies within those limits.

Worked Example: Primary-Rule Crossover Analysis

In a hypothetical balanced crossover study of 12 participants, AUCs were calculated with the stated BLQ rule. The resulting participant-level log AUC ratios \(d_i=\log(AUC_{T,i}/AUC_{R,i})\) are:

−0.10, −0.06, −0.04, −0.02, 0.00, 0.01, 0.02, 0.03, 0.04, 0.06, 0.08, 0.16.

The sum is 0.18, so \(\bar{x}_d=0.18/12=0.0150\). The sum of squared deviations from 0.0150 is 0.0515; thus \(s_d=\sqrt{0.0515/11}=0.06842\), with \(df=11\). Using \(t_{0.95,11}=1.796\), the standard error and 90% interval on the log scale are:

$$ \begin{aligned} SE&=\frac{0.06842}{\sqrt{12}}=0.01975,\\ 0.0150\pm 1.796(0.01975) &=0.0150\pm 0.03547\\ &=[-0.02047,\ 0.05047]. \end{aligned} $$

Exponentiating the endpoints and the mean gives:

$$ \begin{aligned} \exp(0.0150)&=1.015,\\ [\exp(-0.02047),\exp(0.05047)] &=[0.9797,\ 1.0518]. \end{aligned} $$

The estimated test-to-reference geometric mean AUC ratio is 101.5%, with a 90% confidence interval of approximately 97.97% to 105.18%. Because the complete interval lies inside 80%–125%, this hypothetical analysis concludes BE for AUC under the stated model and handling rule. This conclusion does not establish BE for another endpoint such as \(C_{\max}\).

Sensitivity, Diagnostics, and Limitations

The main analysis answers the question under the prespecified rule. A sensitivity analysis asks whether a different plausible handling convention would materially change the exposure comparison or its decision. It should be clearly labelled as a sensitivity analysis, not substituted after seeing which result is preferable.

Worked Example: Does an Alternative Rule Change the Decision?

Suppose the same hypothetical crossover dataset is reprocessed by setting internal BLQ observations to zero. For illustration, assume this recomputation gives a mean paired log ratio of −0.2050 and an SD of 0.06842 across the same 12 participants. These are stipulated summary results from the hypothetical reanalysis, not quantities inferred from the preceding single-participant profile.

The standard error and 90% interval have the same width as before:

$$ \begin{aligned} SE&=\frac{0.06842}{\sqrt{12}}=0.01975,\\ -0.2050\pm 1.796(0.01975) &=[-0.24047,\ -0.16953]. \end{aligned} $$

Exponentiation gives a geometric mean ratio of \(\exp(-0.2050)=0.8147\), or 81.47%, and a 90% confidence interval of approximately \([78.63\%,84.41\%]\). The lower endpoint is below 80%, so this sensitivity analysis does not meet the conventional BE criterion. Unlike the primary-rule result, it would not conclude BE for AUC.

The contrast between the two decisions shows why BLQ handling deserves scrutiny before model fitting. It does not justify selecting whichever rule produces a favorable decision. The primary rule must be justified in advance, and material sensitivity to alternatives should be reported.

Useful checks include confirming the assay LLOQ and units; verifying that BLQ flags are not confused with zero values or data-entry errors; and tabulating BLQ counts by treatment, period, and sampling time. Review whether BLQs cluster before the first quantifiable sample, within the quantifiable profile, or after the last quantifiable sample. A pronounced treatment imbalance in internal BLQ frequency can be especially consequential because it may affect the treatment contrast rather than merely add similar uncertainty to both arms.

Also inspect whether the calculated AUC changes are concentrated in a few participants, whether the same integration interval was used throughout, and whether conclusions are sensitive to a defensible alternative rule. These are diagnostics of the analysis and data; they do not prove that a particular rule is correct. If many internal values are BLQ, or a treatment profile is largely below the LLOQ, the available data may not support a stable exposure comparison under simple substitution or trapezoidal conventions. The assay sensitivity and sampling design may then limit what can be concluded.

Model-based approaches that account for concentrations as censored observations are alternatives in some settings, but they require an appropriate concentration-time model and assumptions about its error structure. They are not interchangeable with ordinary noncompartmental AUC calculations and are not automatically preferable. The primary analysis should follow the applicable protocol and regulatory expectations; a sensitivity analysis can show how much the decision depends on the BLQ convention.

Common Pitfalls

  • Equating BLQ with zero. A BLQ result means the assay did not quantify the sample below its LLOQ; it does not establish a true concentration of zero.
  • Changing the rule by treatment or after viewing results. A handling convention should be applied consistently and prespecified. Post hoc selection can make the BE conclusion depend on outcome-driven choices.
  • Reporting only the decision. State the handling rule, its effect on exposure estimates, and the confidence interval. A pass/fail label alone hides important sensitivity.
  • Calling a sensitivity result the primary result. Alternative substitutions answer a robustness question. They do not replace the justified primary analysis simply because they yield a different conclusion.
Key takeaway: A BE conclusion involving BLQ samples is conditional on the prespecified handling rule. Calculate exposure consistently, examine whether BLQs are imbalanced across treatments, and report sensitivity when plausible alternatives materially change the estimate or decision.

Check Your Understanding

Answer each question using the stated rule and distinguish the primary calculation from sensitivity analyses.

  1. A predose sample is reported BLQ. Does this establish that the participant's true concentration was zero?
  2. For an internal BLQ between two quantifiable samples, explain what “treat as missing and span the gap” means in the trapezoidal calculation.
  3. In the sensitivity example, why is the zero-substitution interval not the primary BE result?
  4. What does a 90% confidence interval entirely within 80%–125% support in the paired AUC analysis?
  5. Name two diagnostics that could reveal a concern with BLQ handling in a crossover dataset.

Solutions. (1) No. It means the concentration was below the assay's LLOQ, not that it was measured as zero. (2) Omit the internal BLQ as a numerical concentration and calculate the trapezoid using the quantifiable observations on either side, with their actual time separation. (3) It is an alternative-rule sensitivity analysis; the prespecified and justified rule defines the primary result, and the alternative cannot be selected merely because of its outcome. (4) Under the stated analysis assumptions, it supports BE for that endpoint because the entire 90% interval for the test-to-reference geometric mean ratio is within the conventional limits. (5) Examples include tabulating BLQ counts by treatment and time, checking the LLOQ and BLQ flags, and evaluating whether alternative defensible handling materially changes AUC or the decision.

Previous

Next

Back to Tutorials