Limit of Detection, commonly referred to as LoD, is one of the most important analytical performance characteristics for an in vitro diagnostic assay. It is often described as the lowest concentration of analyte that can be reliably detected, but this definition can lead to an oversimplified interpretation. LoD is not simply the lowest concentration that produces a positive result during testing. It is a statistically derived threshold that reflects both background measurement variability and the variability of measurements close to the detection limit. This distinction is particularly important for high-sensitivity immunoassays, molecular diagnostics and point-of-care systems, where small differences in signal can change whether a result is classified as detected or not detected. A credible LoD claim therefore depends far more on the design and execution of the study than on the final number reported.
Why Limit of Detection Is Not a Simple Threshold
At very low analyte concentrations, the signal generated by an assay is often difficult to distinguish from background variation. Instrument baseline fluctuations, random analytical variation and ordinary assay imprecision all affect measurements close to zero. A blank sample may occasionally produce a measurable signal, while a genuine low-level positive sample may not be detected in every replicate. These two result distributions can overlap, which means a single observed positive result cannot establish a reliable LoD.
Under the CLSI EP17-A2 guideline (a framework for evaluating and verifying detection capabilities), the LoD estimate combines the Limit of Blank with the variability observed in low-concentration samples. The Limit of Blank represents the highest apparent analyte concentration expected when a true blank sample is measured, while the low-level sample data show how consistently the assay can detect concentrations close to the expected threshold. The resulting LoD therefore reflects both background signal and assay imprecision. This produces a much more meaningful estimate than simply identifying the lowest positive result seen during development testing.
The Importance of Selecting the Right Low-Level Samples
The quality of an LoD study depends heavily on the selection of low-concentration samples. Samples chosen too far above the expected detection limit tend to show less variation and may produce an unrealistically low or optimistic LoD estimate. By contrast, samples chosen too close to zero can generate a high number of non-detectable results, making the statistical estimate unstable or difficult to interpret.
A stronger study design includes several low-concentration levels around the anticipated detection region. This allows the manufacturer to observe how detection performance changes across the range where the assay transitions from mostly negative to consistently positive. Repeated measurements should be distributed across multiple days and operators and, where practical, more than one reagent lot. This is not simply a matter of generating a large number of results. The purpose is to ensure that the study captures the real sources of variability the assay will encounter during routine use.
A well-selected sample panel also improves the credibility of the final claim because it shows that the LoD was established through deliberate study design rather than by selecting a convenient concentration after testing had already been completed.
Understanding the Limit of Blank
The Limit of Blank, or LoB, is a critical component of the LoD calculation. It represents the highest apparent analyte concentration expected when testing a sample that contains no analyte. Establishing the LoB requires a substantial number of blank replicates tested across multiple runs so that normal background behaviour can be characterised.
One important consideration is that instrument noise close to zero is not always normally distributed. Blank results may be skewed , particularly when the assay produces truncated, censored or near-zero measurements. Automatically applying a normal distribution can therefore lead to an inappropriate LoB estimate. A nonparametric percentile-based approach may be more suitable where the data do not meet the assumptions required for parametric analysis. Using the 95th percentile of the observed blank distribution allows the LoB to be based on actual blank behaviour rather than on an unsuitable statistical model.
The statistical method should be chosen after examining the data. The objective is not to apply the simplest calculation, but to use an approach that accurately represents the assay and supports a defensible conclusion.
Variability Near the Detection Limit
Once the LoB has been established, the next step is to understand how variable the assay is at low analyte concentrations. Higher-level controls are not necessarily suitable for this purpose because assay variability may behave differently close to the detection threshold. Near the LoD, instrument noise and background effects can have a much greater influence on the result than they do at higher concentrations.
This changing variance is known as heteroscedasticity. It means that one estimate of standard deviation may not apply consistently across the full measuring range. If this behaviour is ignored, the calculated LoD may be biased and may not accurately reflect real assay performance. Low-level variance should therefore be estimated using samples that are genuinely close to the expected detection region.
Manufacturers should examine whether variability increases, decreases or changes unpredictably across low concentration levels. This analysis helps determine whether a standard deviation-based model is appropriate or whether a hit-rate approach would provide a better representation of the data.
Choosing the Right Statistical Method
CLSI EP17-A2 describes both parametric and nonparametric approaches to establishing LoD. A parametric calculation may be appropriate where the blank and low-level data reasonably follow a normal distribution and a standard deviation-based model is scientifically justified. However, normality should be assessed rather than assumed.
Where the data are skewed, non-normal or dominated by detected and non-detected outcomes, a nonparametric hit-rate approach may be more suitable. Under this method, several low concentrations are tested repeatedly and the proportion of positive detections at each level is calculated. The LoD is then defined as the lowest concentration at which the required probability of detection, typically 95 percent, is achieved.
The method should follow the data. Selecting a statistical approach simply because it produces the lowest LoD value undermines the purpose of the study. The objective is to establish a detection limit that can be defended scientifically and reproduced in practice, not to produce the most commercially attractive number.
Why Study Conditions Must Reflect Routine Use
An LoD study can appear statistically precise while still failing to represent routine assay performance. Testing completed on a single day by one operator using one reagent lot leaves important sources of variation unexplored. The resulting value may look convincing on paper but may not hold once the assay moves into broader clinical or commercial use.
A robust design should include sources of variation that reflects the assay’s intended operating environment. Multiple days allow environmental and calibration-related variation to emerge. Multiple operators capture differences in handling and technique. Multiple reagent lots provide information about manufacturing consistency. Where relevant, more than one instrument may also be appropriate.
The study design should be proportionate to the assay and its intended use, but it should be broad enough to demonstrate that the LoD claim is not dependent on ideal laboratory conditions.
The Importance of Specimen Matrix
The specimen matrix can have a major effect on performance close to the detection limit. Artificial or overly simplified samples may not behave like genuine clinical specimens. Differences in protein composition, viscosity and endogenous substances can alter signal response and assay variability.
An LoD study should therefore consider the intended specimen type and relevant matrix effects. Where contrived samples are used, the manufacturer should justify their suitability and demonstrate that they behave comparably to clinical specimens. If multiple specimen types are included in the intended use, the study strategy should address whether one LoD claim can reasonably apply to all of them or whether separate evaluations are needed.
Ignoring matrix effects creates a risk that the analytical claim will not translate into real post-market performance. A result generated under idealised development conditions may overstate what the assay can reliably detect in routine clinical samples.
Common Weaknesses in LoD Submissions
Several recurring weaknesses can reduce the credibility of an LoD claim. One is the use of too few blank or low-level replicates, which limits the reliability of the statistical estimate. Another is selecting low-level samples that are not close enough to the expected detection threshold. Some studies also rely on one operator, one day or one reagent lot, which fails to capture routine variation.
Unsupported assumptions of normality are another common issue. Parametric calculations may be used without demonstrating that the data meet the required distributional assumptions. In other cases, manufacturers may choose a nonparametric method without explaining why it is appropriate.
Poorly defined specimen matrices, inadequate consideration of interference and failure to explain how concentrations were assigned can also weaken the submission. The final LoD value should always be supported by a clear study rationale, transparent statistical analysis and enough contextual information for a reviewer to understand how the claim was derived.
How to Build a Defensible LoD Study
A defensible LoD study begins with a clear protocol that defines the expected detection region, specimen matrix, sample levels, number of replicates, sources of variation and statistical approach. The study should include an adequate blank assessment, several low analyte concentrations and repeated testing under conditions representative of routine use.
The protocol should explain how low-level samples were prepared or selected, how concentration values were assigned and how sample stability was maintained. Acceptance criteria should be defined before testing begins. Statistical methods should be justified based on the expected characteristics of the data and confirmed after reviewing the results.
The final report should present blank results, low-level results, detection rates, variability estimates and the rationale for the selected LoD. Deviations and unexpected results should be discussed rather than omitted. A transparent report is generally more convincing than one that presents only a final number without explaining the study behind it.
Why LoD Must Remain Relevant After Market Entry
The analytical performance claim established during development should remain relevant throughout the assay lifecycle. Changes to reagents, instruments, software, manufacturing processes or specimen handling may affect low-level performance and should be assessed through established change control process.
Post-market information may also reveal that assay behaviour close to the LoD differs from the original validation findings. Complaint trends, external quality assessment results and field performance data can all provide information that supports or challenges the established claim.
LoD should therefore be treated as part of an ongoing analytical performance framework rather than a fixed development milestone. Where significant changes occur, additional verification or revalidation may be required.
A Defensible LoD Is Built, Not Found
The Limit of Detection should never be presented as the lowest positive result observed during testing. It is the outcome of a carefully designed study intended to understand assay behaviour where analytical signal and background noise are most difficult to separate.
A strong LoD claim rests on appropriate blank and low-level samples, adequate replication, realistic sources of variation, representative specimen matrices and a statistical method that fits the data. When these elements are addressed properly, the final value becomes more than a validation output. It becomes a realistic and supportable statement of what the assay can reliably detect in use.
LFH supports IVD manufacturers in planning, reviewing and documenting analytical performance studies, including Limit of Detection, Limit of Blank and low-level performance assessments. Our team helps organisations develop robust protocols, identify study design gaps and prepare clear, regulator-ready evidence that supports defensible analytical performance claims.
FAQs – Limit of Detection
What is the Limit of Detection in an IVD assay?
The Limit of Detection is the lowest analyte concentration that an assay can detect reliably under defined conditions.
Is the LoD the lowest positive result observed during testing?
No. LoD is a statistically derived threshold based on blank behaviour and variability in low-concentration samples.
What is the difference between LoB and LoD?
The Limit of Blank describes the highest apparent analyte concentration expected in a true blank, while the Limit of Detection incorporates both the LoB and low-level assay variability.
Why should multiple low-concentration levels be tested?
Testing several levels around the expected detection region allows the study to show how detection probability and variability change close to the threshold.
When is a nonparametric LoD method appropriate?
A nonparametric approach may be more appropriate where data are skewed, non-normal or best represented through detection rates rather than standard deviation-based models.
Why do specimen matrix effects matter in LoD studies?
Clinical specimen characteristics can affect low-level assay performance, meaning that artificial samples may not accurately represent routine use.
Should LoD studies include multiple days and operators?
Yes. Including realistic sources of variation helps demonstrate that the LoD remains reliable outside ideal development conditions.
