Laboratory Units: Preserve Original Results Before Standardizing

A central laboratory and several local laboratories deliver the same planned trial measurement in different units, with different methods and reference ranges. Which records can be converted to a common reporting unit, and which need clarification? Start by preserving what each laboratory actually reported. Then create a separate mapping that records the measurement identity, proposed target unit, applicable rule and unresolved questions.

The useful output for clinical data managers, endpoint leads and statistical programmers is a result-by-result decision record. It should allow a reviewer to reconstruct both the original observation and the derived representation. Keep the source report identifier and version alongside the received result and unit text; retain available laboratory, method, specimen and range metadata rather than overwriting them with a single study-wide label.

The publicly readable Study Data Tabulation Model Implementation Guide (SDTMIG) v3.3 separates original results from standardized character and numeric results. Its submission unit labels can use controlled terminology, so an original-unit variable is not a substitute for retaining every raw spelling. The workflow below is a proposed staging aid, not a required dataset or a claim of regulatory acceptance.

Before looking up a factor, classify the requested transformation. The 2022 PharmaSUG dictionary paper distinguishes test-name mapping, normalization of unit names and numeric conversion. That implementation provides a useful baseline, although its historical submission approach is not the current FDA LB/LC advice. Four separate questions belong in the review:

• Unit-label normalization: Does the laboratory confirm that a received spelling means the same unit as the chosen terminology label? Retain that spelling and the reviewed mapping. Do not treat arbitrary capitalization changes as harmless; unit symbols can carry case-dependent meanings.

• Dimensional scaling: Is this the same quantity expressed at a different magnitude? For the fictional mg/dL-to-mg/L examples below, one deciliter is one tenth of a liter, so multiply the result and each available continuous range bound by ten. This changes representation, not the assay.

• Analyte-dependent conversion: A mass concentration such as mg/L cannot be converted to mmol/L from the unit labels alone. The measured entity and an appropriate analyte-specific factor must be established. The 2012 PharmaSUG paper describes this dependence and the ambiguity of local unit labels. It supplies historical workflow context, not a factor dictionary for this study.

• Assay comparability: Can observations from different methods support the intended combined interpretation? A factor of one does not answer that question. Record the method review separately from the arithmetic disposition; a common unit supplies no numerical bridge between assays.

The UCUM v2.2 specification supplies unit semantics and prefixes. It also explains why procedure-defined arbitrary units have no generic SI relationship: their meaning depends on the assay. A valid-looking unit string therefore cannot supply an assay-specific conversion. Keep arbitrary-unit cases out of automatic mass-concentration scaling until appropriate evidence supports the requested transformation.

The LOINC structure FAQ distinguishes component, property, time, system, scale and an optional method axis. Method detail may also be carried outside the code. Consequently, a shared test name or code does not replace local method and specimen metadata. Check what was measured before treating two observations as the same measurement in different units.

Reference-range conversion is another distinct task. Pair an observation with the range applicable to its laboratory, method and population, retaining the source and version of that range. Multiplying the result by ten while leaving its limits in the old unit creates an inconsistent comparison. Conversely, converting both does not normalize one laboratory to another or establish a clinical threshold.

A missing range need not prevent arithmetic on a result whose measurement and unit are known. It does prevent pretending that the applicable limits are known. In T06 below, retain the converted value and an unresolved range status; ask the originating laboratory for the applicable limits. Do not fill the gap using a neighboring row or another laboratory merely because its test name matches.

All seven rows are invented teaching examples for fictional markers. The numbers and reference ranges are nonclinical arithmetic inputs, not patient results, medical thresholds or validation evidence. The target mg/L is illustrative, not a universal laboratory standard. The two linked tables form one proposed crosswalk: the first preserves inputs, while the second records decisions using the same row IDs. No required study or participant identifiers are supplied, so this is not a submission-ready LB/LC dataset.

For T01, the reproducible arithmetic is 3.2 × 10 = 32, 1.0 × 10 = 10 and 4.0 × 10 = 40. T02 applies an identity rule; T06 uses the same dimensional rule for its value without inventing limits. These outcomes illustrate why conversion success, range completeness and comparability need separate fields.

T03–T05 deliberately have no proposed common-unit output. Those staging labels do not instruct a sponsor to submit missing results: each source observation still exists. T07 demonstrates the opposite boundary. Its number and unit can remain unchanged while the question of combining M-D with M-A remains open. The different toy ranges alone prove neither interchangeability nor a particular magnitude of bias.

The following compact T01 correspondence uses SDTMIG v3.3 sections 4.5.1 and 6.3.6. It illustrates successful numeric mapping; source-report and rule versions remain in the proposed lineage record. This example does not replace the applicable implementation guide, terminology or full dataset specification.

LBORRESU uses submission terminology; retain the raw unit string separately when its spelling changes. LBNRIND describes position relative to a reference range, with its original or standardized basis specified in metadata. It does not express clinical significance. The fictional method and laboratory labels above are explanatory placeholders, not a controlled-terminology package.

Assign an owner and retain the evidence needed to close each exception. The following actions are proposed workflow choices; they are not a regulator-prescribed escalation chart. A response should resolve the specific missing input, rather than merely acknowledge that a query was sent.

• For T05, ask the laboratory to identify the original unit for that observation and provide the supporting report or corrected transfer. Preserve the relationship between versions. A unit copied from T01 would conceal the uncertainty rather than resolve it.

• For T04, ask the laboratory lead to establish the measured chemical entity, reporting basis and appropriate conversion factor. Record the factor source and review before introducing a new rule. Do not derive a molecular weight for an unnamed fictional marker.

• For T03, ask the method owner what evidence, if any, supports expressing the procedure-defined output in the requested target. A generic unit dictionary cannot establish that relationship. Preserve the native observation while its eventual tabulation is decided under the applicable standard.

• For T06, obtain the range applicable to the observation or document why it is unavailable. Keep range completeness separate from the successful value calculation. For T07, route intended cross-method use to laboratory and statistical review; a completed identity mapping does not close that review.

A performed test with a mapping problem is not a test that was not done. Keep its observation visible and distinguish the staging disposition from final tabulation and analysis decisions. Nor does a mapping hold by itself justify exclusion. Service-triggered review has a different trigger and evidence chain, covered in Assessing Endpoint Data After Analyzer Service.

Keep inequality results separate from the continuous numeric examples here. In SDTMIG v3.3, inequality qualifiers remain in the standardized character result and the numeric result is null; analysis imputation belongs in analysis datasets under the analysis plan. Do not insert a substituted analysis value into LBSTRESN. This boundary note does not supply a below-quantification analysis strategy.

The FDA Study Data Technical Conformance Guide v6.3, October 2026, LB/LC subsection on printed pages 13–14, recommends both International System of Units (SI) results in LB and conventional-unit results in LC for clinical studies. Special situations call for clinical review-division agreement. It asks sponsors to obtain both representations from laboratories where feasible, explain conversion handling in the Reviewer’s Guide, provide applicable ranges and keep corresponding domain dimensions aligned.

The guide also addresses unchanged values where a conversion factor is unavailable. This is not permission to label an arbitrary-unit number as a mass concentration. Keep source units, conversion decisions and submission handling distinct. The document contains nonbinding recommendations for FDA submissions; it does not establish assay comparability or a worldwide laboratory rule.

CDISC’s Standardized Lab Units knowledge-base article uses a less specific description of conventional-unit submission as a possible FDA request. Its public page shows no date establishing the chronology of that wording. Use the current FDA document for the FDA clinical-study advice here; do not reduce its dual-domain recommendation to an optional LC domain.

A versioned dictionary is useful only if a reviewer can connect a rule to the observation it transformed. The 2022 PharmaSUG paper describes a maintained dictionary and separates stages of mapping. For this proposed crosswalk, use that distinction to make the handoff inspectable:

• Trace source to output: retain the report/version, received result and unit, laboratory/method, range version and resulting record. If a source correction arrives, identify the affected outputs and the rules that must be rerun rather than silently replacing their history.

• Version the decisions: record the test-name map, unit-label map, numeric rule and applicable range source separately. A change to a synonym does not automatically justify changing a factor; a new laboratory method does not inherit comparability from an old rule.

• Recompute the examples independently: check T01’s result and both bounds, T02’s identity mapping and T06’s value-only scaling. Check the refusal cases as well: T03–T05 must remain visible without invented target values. These checks verify the teaching arithmetic, not a production system.

• Review residual uncertainty: record who resolved missing metadata and who assessed cross-method use. Keep open items and documented absences visible at handoff. A completed conversion count alone cannot show that the range and method questions were answered.

The detailed field reference used here is the readable v3.3 edition, paired with SDTM v1.7. SDTMIG v3.4 was published on 29 November 2021, supersedes prior editions and is paired with SDTM v2.0. Its public page includes errata, while the full files require sign-in; this article does not claim a field-by-field verification against that full specification. Obtain the applicable authorized standard, model, errata and terminology for implementation. Publication alone does not determine agency support for a study.

As checked on 24 October 2026, CDISC’s development page lists SDTMIG v4.0 and SDTM v3.0 as resolving public-review comments, with projected publication in 2027. Those projections are provisional and do not make the developing editions final requirements.

Connect this mapping handoff to the study’s other decisions. The SAP guide addresses endpoint derivations and analysis rules; the database-lock checklist addresses completeness and sign-off. Use the eCOA/EDC reconciliation guide for participant, visit and timestamp conflicts and the computerized-systems validation guide for the broader validation evidence. The synthetic crosswalk supplies neither assay validation nor production-system approval.