Guide overview
The essential idea
Automated comparison can detect that NHANES documentation changed between releases; it cannot determine by itself whether the scientific measurement changed. A trustworthy harmonization record separates what was observed from what is inferred, links the source evidence, and assigns an analysis action proportional to the confidence of the classification.
Practical framework
What to apply
Documentation variation
Did the text change without evidence of a scientific discontinuity?
- Record the exact source documents, release dates, affected sections, and detected difference.
- Check whether wording, formatting, labels, URLs, or explanatory detail changed while the documented method remained stable.
- Classify the event as documentation variation when substantive method change is not demonstrated.
- Keep the event visible so a later source or expert review can refine the classification.
Confirmed method change
Is there direct evidence that collection, instrument, assay, or processing changed?
- Seek explicit statements in component documentation, laboratory methods, analytic notes, manuals, or revision notices.
- Identify the affected cycles, variables, specimens, units, detection limits, calibration, and eligible population.
- Capture any NCHS correction, conversion equation, crossover study, or recommended analytical treatment.
- Assign a concrete action: transform, calibrate, stratify, adjust, sensitivity-test, or avoid pooling.
Uncertain or incomplete evidence
Is a change plausible but not yet confirmed or ruled out?
- State what evidence is available, what is missing, and why the current sources do not resolve the question.
- Retain an explicit uncertainty status rather than forcing a binary stable-or-changed decision.
- Prioritize manual source review, archived documentation, related variables, or methodological consultation.
- Use conservative analysis actions until the uncertainty is resolved.
Stable evidence
Do the reviewed sources support cross-cycle continuity?
- Confirm consistent construct, wording, eligibility, protocol, method, unit, coding, and applicable weight.
- Record the source set and review date even when no discontinuity is identified.
- Distinguish reviewed stability from merely having no automated alert.
- Continue to evaluate empirical cycle patterns and future revisions.
Before you finish
A short quality review
- The observed change and the scientific interpretation are written separately.
- Every classification links to cycle-specific primary documentation.
- Confidence and unresolved evidence are explicit.
- The affected variables, cycles, methods, units, and populations are identified.
- A proportional harmonization action is attached to the classification.
- The record has a version, review date, and path for correction or reclassification.
A text change is a signal, not a verdict. Automated comparison should direct scientific review. The final harmonization classification must remain source-backed, qualified, versioned, and correctable.