Below-Quantification Observations

Why a Below-Quantification Result Needs Its Own Representation

A laboratory result reported as “below quantification” is not a measured concentration of zero, and it is not simply an absent observation. It says that the assay did not provide a quantitative result below a specified threshold. That threshold is part of the information in the record: without it, “below quantification” does not describe how much is known about the observation.

For example, “<0.10 mg/L” and “<0.50 mg/L” both indicate that the result was not quantified, but they constrain the possible concentration to different ranges. Replacing either with zero or a fixed fraction of the limit discards that distinction and creates a numerical value the assay did not report.

Definition: A below-quantification observation is a result for which the assay indicates that the concentration is below its applicable lower limit of quantification, or LLOQ, rather than providing a quantitative result. The observation is left-censored at the LLOQ: the concentration is known to be below that limit, but its value within the below-limit range is not quantified.

The LLOQ is the lowest concentration at which the assay can quantify results with its specified performance under validated conditions. It is not necessarily the same as a limit of detection. Detection concerns whether an analyte can be distinguished from background; quantification concerns whether a concentration can be reported with acceptable performance. Use the laboratory’s result and assay documentation to determine what a flag means, rather than inferring it from the numeric field alone.

Represent the Result and the Censoring Information

A useful dataset keeps the source report and the fields needed to interpret it. In a general-purpose record, these might include a raw result field, a numeric result field, a censoring-status field, and the applicable assay limit. The names of the fields depend on the data system; their meanings should be explicit and consistent.

Field	Meaning for a below-quantification record
Raw result	Preserves the laboratory report, such as “<0.10”.
Numeric result	Contains a quantitative value only if one was reported. Do not enter zero or a substituted fraction as though measured.
Censoring status	Identifies the result as below the quantification limit, using a documented code.
Assay limit	Stores the applicable LLOQ as a numeric value, with its units.
Units and assay context	Identifies the concentration units and, where needed, which assay or sample processing condition determines the limit.

Some analysis formats require a numeric dependent-variable field even when the result is censored. In that case, follow the specified format and document precisely what the numeric field represents; retain the censoring indicator and limit as separate information. A value placed in a numeric field solely to satisfy a file format must not silently become an assertion that the concentration was measured at that value.

Worked Example: Encoding a Mixed Set of Results

In a hypothetical assay, concentrations are reported in mg/L and the LLOQ is 0.10 mg/L. Three laboratory reports are “0.24”, “<0.10”, and “0.13”. Show how to preserve what each report says.

For “0.24”, record the raw result as “0.24”, the numeric result as 0.24 mg/L, and the status as quantified. The assay limit can also be stored on the record or linked through documented assay metadata.

For “<0.10”, preserve that raw text, mark the status as below quantification, and record the LLOQ as 0.10 mg/L. Do not put 0 or 0.05 mg/L in the numeric result field as if either were measured. If the data specification requires the numeric field to be blank for censored values, leave it blank; the status and limit carry the censoring information.

For “0.13”, record 0.13 mg/L as a quantified result because the report provides a quantitative value above the stated LLOQ. The three records can therefore be distinguished as quantified, below quantification, and quantified, respectively, without inventing a concentration for the censored record.

Limits Can Differ Across Records

The applicable LLOQ may differ across samples because of assay changes, dilution, sample volume, matrix, or other documented laboratory conditions. Do not assume a single threshold applies to every observation unless the assay metadata establish that it does. The limit must be expressed in the same units as the concentration, and any conversion must be traceable.

Worked Example: Preserving Record-Specific Limits

Suppose two hypothetical samples from one subject are reported as “<0.05 mg/L” and “<0.20 mg/L”. The laboratory confirms that different validated assay conditions produced the respective LLOQs. How should the records be represented, and what is known about each concentration?

For the first record, preserve “<0.05”, assign the below-quantification status, and store an LLOQ of 0.05 mg/L. The result indicates a concentration below 0.05 mg/L. For the second, preserve “<0.20”, assign the same status, and store an LLOQ of 0.20 mg/L. It indicates a concentration below 0.20 mg/L.

The statements are not interchangeable: the second result leaves a wider range of possible concentrations. Replacing both limits with a single study-wide value would misstate at least one record unless a documented conversion or assay rule justified that value. Keep the condition that identifies which limit applies, such as an assay version or dilution factor, when it is needed to interpret the record.

This representation also helps distinguish censoring from missingness. If a laboratory result is absent because no sample was collected, the result is missing; there is no basis for claiming that the concentration fell below an LLOQ. If the laboratory reports a value below quantification, there is an observation with a known censoring threshold. The earlier material on missing values versus true zeros makes the related distinction: neither missingness nor a below-limit report should be converted into a true zero without evidence that zero was actually measured.

Why the Limit Is Not a Measured Value

The difference matters when results are summarized or modeled. If a result is “<L”, the data establish an inequality, not equality to \(L\), zero, or \(L/2\). Treating the result as \(L\) puts every censored observation exactly at the limit. Treating it as \(L/2\) assumes a particular value within the unobserved range. Neither conclusion follows from the report.

A model that explicitly accounts for censoring uses the probability of the reported event, \(C<L\), rather than pretending that a concentration was measured at the boundary. The following derivation illustrates the distinction. It is not a substitute for specifying the assay data correctly: a model cannot recover a missing limit or repair an incorrectly coded status.

Model assumption for the following result: Let \(C\) be a positive concentration, and suppose \(Y=\ln(C)\sim N(\eta,\omega^2)\), where \(\eta\) and \(\omega>0\) are specified model parameters. Assume the observation is censored solely because \(C\) is below a known, positive LLOQ \(L\). The model’s contribution for this record is the probability of the event \(C<L\).

Because the logarithm is increasing for positive concentrations, \(C<L\) is equivalent to \(Y<\ln(L)\). Standardizing \(Y\) gives the probability of the observed event:

P(C<L)=P(Y<\ln L) =\Phi\left(\frac{\ln L-\eta}{\omega}\right)

Here \(\Phi\) is the cumulative distribution function of a standard normal random variable. This formula follows by integrating the model density over all concentrations consistent with the report. In concentration units, the same argument is:

P(C<L)=\int_0^L f_C(c\mid\eta,\omega)\,dc

Thus a censored record contributes a cumulative probability, not a density evaluated at \(L\). By contrast, an exact quantified measurement contributes a density at its reported value under a continuous model. This distinction is why substituting a boundary value changes the statistical information represented by the record.

Worked Example: Calculating a Below-Limit Probability

Consider a hypothetical model with \(Y=\ln(C)\sim N(\ln(0.50),\omega^2)\), where concentration is measured in mg/L and \(\omega=\ln(2.5)/2.682\), approximately 0.3416. For a record reported as “<0.20 mg/L”, calculate the model probability of that event.

The standardized limit is:

z=\frac{\ln(0.20)-\ln(0.50)}{\omega} =\frac{\ln(0.20/0.50)}{\ln(2.5)/2.682} =\frac{-\ln(2.5)}{\ln(2.5)/2.682} =-2.682

Therefore:

P(C<0.20)=\Phi(-2.682)\approx 0.003659

Under the stated hypothetical model, the probability of a concentration below 0.20 mg/L is about 0.3659%. This is a model-based probability for the censored event, not an imputed concentration for the individual record. The calculation depends on the stated lognormal model and parameters; it does not establish that every assay or dataset should use this model.

Checks, Alternatives, and Common Failure Modes

Before using the prepared data, check that each below-quantification status agrees with the source report and that its LLOQ is present, positive, and in the concentration’s units. Review the data by assay, time period, and relevant sample-processing condition to detect changes in reporting limits. Investigate records whose status conflicts with the reported result or whose limit is missing. Do not infer a censoring status merely because a numeric value appears small; use the laboratory’s reporting convention.

Preserve the source.
Keep the original reported text or an auditable equivalent so that a coded record can be checked against what the laboratory supplied.

Assign status and threshold.
For a below-quantification report, record the documented status and the applicable LLOQ, with units and necessary assay context.

Check consistency.
Compare source text, status, numeric result, limit, and units; resolve conflicts rather than silently changing a value.

Common alternatives have different implications. Replacing every censored result with zero asserts a true zero. Replacing it with half the LLOQ assumes an unreported value of \(L/2\). Replacing it with the LLOQ treats a strict below-limit report as an exact boundary measurement. These substitutions may be used as explicit approximations in a particular analysis, but they are not faithful data representations and can affect summaries and model estimates, especially when many records are censored or limits vary.

A model-based censored-data method is a different approach: it retains the inequality information and uses the probability of the censored event under a specified model. Its validity depends on the model assumptions and correct coding of the limit and censoring status. It can be inappropriate if the reported flag has another meaning, if the threshold is wrong, or if the assumed distribution does not describe the data. Neither a data flag nor a model can determine the unknown concentration below the limit from that record alone.

Common pitfall: “The laboratory did not quantify this sample, so I will enter zero; zero is a conservative value.” This reasoning confuses an inability to quantify with evidence of no analyte. The correction is to preserve the below-quantification status and assay limit, and to leave the numeric result as unmeasured unless the data specification explicitly defines a separate, documented encoding.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions below.

A report reads “<0.08 mg/L”. Name the information that should be retained and state what should not be claimed about the numeric concentration.
One assay reports “<0.05 mg/L” and another reports “<0.25 mg/L”. Why is it incorrect to treat these as identical censoring thresholds?
Under a specified continuous concentration model, why does a below-limit result contribute a probability rather than a density evaluated at the LLOQ?
A numeric field contains 0 for a record whose raw report is “<0.10”. What should a data reviewer investigate before analysis?

Solutions. (1) Retain the raw report, below-quantification status, LLOQ of 0.08 mg/L, and units; do not claim that the concentration equals zero, 0.08 mg/L, or another specific value. (2) The reports constrain concentrations to different ranges, so each record needs its applicable limit and assay context. (3) The report establishes the event \(C<L\), not an exact value at \(L\); its model contribution is the probability over values consistent with that event. (4) Check whether zero is a documented placeholder required by the data format or an erroneous substitution, and confirm that the censoring status and limit are separately retained. Do not interpret the zero as a measured concentration without supporting evidence.