Biomarker Assay Performance and Sampling Windows

Why assay performance and sampling time belong together

A biomarker measurement can be analytically precise yet uninformative if samples are collected only after a transient response has passed. Conversely, a well-chosen sampling schedule cannot rescue measurements that are unstable during handling, below the assay’s quantification limit, or compressed by detector saturation. Early-phase planning therefore needs two linked judgments: whether the assay can measure the expected samples reliably, and whether the sampling times can reveal the response features relevant to the stated biomarker estimand.

The aim is not to find a universally best assay or schedule. Performance requirements depend on the anticipated concentration range, sample matrix, handling conditions, and decision the measurements are intended to inform. Acceptance criteria should be set before examining study results and justified for the intended use. A criterion suitable for exploratory description may not be adequate for a decision that depends on distinguishing a small biomarker change.

Precision, accuracy, and replicate measurements

Precision is often summarized by the standard deviation \(s\) of repeated measurements and, when the mean is positive and meaningfully away from zero, by the coefficient of variation (CV):

Here, \(\bar{x}\) is the mean of the repeated measurements, and \(s\) is their sample standard deviation. A lower CV indicates less relative dispersion in those measurements. Precision is not the same as accuracy: measurements can cluster tightly while all being shifted away from the true or reference value. Accuracy or bias must be assessed separately using appropriate reference material or comparison procedures. A CV is also a poor summary when the mean is near zero; in that setting, report the absolute variability in the assay’s measurement units.

Precision conditions matter. Repeatability concerns measurements made under similar short-term conditions, such as within one run. Variation across runs, days, operators, or reagent lots can be larger. An assay evaluation should match the conditions likely to affect the study samples and should examine the concentration range expected in those samples. A single precision estimate at one concentration does not establish precision throughout the assay range.

Technical replicates can help characterize measurement variability, but they do not create additional independent participants. If \(m\) independent replicate measurements each contain random error with standard deviation \(\sigma\), their average has random-error standard deviation \(\sigma/\sqrt{m}\). This follows because the variance of the average is:

The result assumes independent errors with the same variance. Replication does not remove a shared systematic shift, such as an incorrectly calibrated assay, and it does not reduce variability between participants. Averaging replicates can also conceal a failed replicate or a handling problem if raw results are not reviewed.

Stability, quantification limits, and saturation

Stability must be evaluated under conditions relevant to the planned samples: for example, the time from collection to processing, refrigerated or frozen storage, and the number of freeze-thaw cycles. Compare measurements after the specified condition with an appropriate reference measurement, using a tolerance chosen in advance. A common descriptive calculation is percent recovery:

Recovery summarizes relative change; it does not itself determine acceptability. The allowable range must be justified for the assay’s intended use. Stability under one tested condition does not imply stability under a different duration, temperature, matrix, or handling procedure. Samples outside validated conditions should not be treated as though their results were known to be unaffected.

The LLOQ is the lowest concentration at which the assay has demonstrated adequate quantitative performance under its validation conditions; the ULOQ is the highest. A result below the LLOQ is not a precise numerical concentration merely because the instrument returns a signal. Similarly, a result above the ULOQ should not be reported as a reliable extrapolation from the validated range. The appropriate reporting convention depends on the assay and analysis plan, but the distinction between a measured value and a result outside the quantitative range must be preserved.

At the high end, saturation can arise from a detector’s finite response or from another assay component becoming limiting. When detector response flattens, meaningfully different analyte concentrations may produce nearly indistinguishable signals. This is a measurement limitation, not evidence that the biological response itself has reached a plateau. If dilution is considered to bring a sample into range, dilution performance and any required dilution correction must be validated; an unvalidated correction does not restore quantitative reliability.

Choosing sampling windows for the expected response

A sampling schedule should reflect the response features the biomarker estimand is intended to describe. Depending on prior biological and translational evidence, those features may include the earliest plausible onset, a period near the expected maximum response, persistence, and recovery. Samples collected too sparsely can miss a brief change or make its timing unclear. A dense schedule may improve temporal description, but it increases burden and can add little information if the response is expected to change slowly.

The schedule must also be feasible for the assay. Collection and processing should occur within validated stability conditions, and the expected sample concentrations should lie within the quantitative range as far as can reasonably be anticipated. If concentrations may be close to the LLOQ or ULOQ, this uncertainty should inform assay preparation and the plan for handling out-of-range results. Sampling time is not a substitute for analytical validation, and a sample collected at the nominally ideal time remains uninformative if it cannot be measured reliably.

Diagnostics, alternatives, and limitations

A useful review examines raw replicate measurements and performance across relevant concentrations and runs, not only a single overall summary. Warning signs include unexpectedly high CVs, consistent shifts between runs, recovery outside prespecified stability criteria, an accumulation of results below the LLOQ or above the ULOQ, and signals that stop increasing as expected at the high end. Such patterns can indicate assay limitations, handling problems, or a mismatch between the anticipated and observed concentration range. They should be investigated rather than resolved by selectively averaging, discarding, or numerically extrapolating results.

Possible responses depend on the cause. If within-run measurement variability is the concern, replicate testing may help characterize it; if run-to-run shifts are the concern, assessment across runs and relevant operating conditions is needed. If samples are unstable, validated handling conditions or a different collection-to-processing plan may be required. If results fall outside the quantification range, a validated assay range or dilution procedure may be necessary. If a transient response may have been missed, future sampling windows may need adjustment. These options address different problems and are not interchangeable.

The approach has important limits. Assay performance established with quality-control material may not fully describe every participant sample or matrix condition. A small precision or stability assessment cannot prove performance under all circumstances. Sampling windows are chosen using uncertain expectations, so no finite schedule guarantees that an unanticipated brief response will be observed. Finally, an assay’s measurement limitations should not be confused with biological properties: a flat instrument signal can mask concentration differences, while a sparse schedule can obscure timing. Neither alone establishes the underlying response shape.