How to Replace Weak Secondary Citations with Original Evidence

A secondary citation is a claim you are carrying on someone else's authority: a review, guideline, editorial, or textbook says a study found something, and you cite the study without reading it. That shortcut is fine for orientation and risky for scholarship. The fix is a repeatable trace: decide whether the secondary source is enough, locate the original work, read what it actually reported, and rewrite your sentence so its strength matches the evidence.

This guide gives clinicians and researchers a complete sequence for that task, using a focused population and outcome example. It covers diagnostic checkpoints, a worked example, a reusable checklist, and clear stop or escalation rules. The workflow stays focused on evidence appraisal and citation practice.

A secondary source is usually sufficient when you need background, a definition, or a map of a field, and when you attribute the claim to that source rather than to the underlying study. It becomes insufficient when your sentence makes a specific factual claim that depends on the design, population, or result of a primary study you have not examined.

Use these diagnostic checkpoints to decide whether to trace:

• Specificity. Does your sentence state a direction, magnitude, population, or outcome that only the primary study can support? If yes, trace it.
• Attribution. Are you citing the primary study, or the review that mentioned it? If you cite the primary study, you are asserting you know what it found.
• Stakes. Will the sentence anchor a guideline recommendation, a grant rationale, or a teaching point? Higher stakes justify reading the original.
• Currency. Does the secondary source predate newer primary work on the same question? If so, the original may no longer represent the current evidence.
• Chain length. Is the secondary source itself citing another secondary source? Longer chains accumulate distortion.

If none of these apply, you can often keep the secondary citation and attribute the claim to it directly. The rest of this guide assumes at least one checkpoint fired.

Write down the sentence you intend to publish, then isolate the factual assertion inside it. Separate what you are claiming from how you are framing it. A useful pattern is: population, exposure or intervention, comparator, outcome, and direction.

For example, a draft sentence might read: "In adults with community-acquired pneumonia, early mobilization was associated with shorter hospital stay." The factual core is the population (adults with community-acquired pneumonia), the exposure (early mobilization), the outcome (length of stay), and the direction (shorter). Everything else is style.

This extraction matters because it becomes your acceptance test. When you reach the original study, you will check each element against what was actually reported. If the study examined a different population or a different outcome, your sentence needs to change, not the study.

Start from the citation in the secondary source. Capture the authors, title, journal, year, and any identifier such as a DOI. Then confirm you have the right work before reading it.

A DOI can be resolved through doi.org to the current destination registered for that identifier. Resolution helps locate a work, but bibliographic fields and the source text still need to be checked for identity and claim support. In practice, that means resolving the DOI, then comparing the title, authors, and year against the citation you started with. Mismatches are common when reviews cite preprints, conference abstracts, or corrected versions.

If the secondary source gives no identifier, search by author and title. Google Scholar documents searching by author and title, date sorting and date limits, Cited by and Related articles, citation export, library access links, email alerts, and ways to locate alternative versions of a work. Those features help you find the record and see whether a published version exists.

For a scholarly search interface that supports finding papers, you can run the title or author query in WisPaper search. If you prefer a public entry point that explains the search capability before you begin, start from WisPaper Scholar Search.

Checkpoint: Do the title, authors, year, and venue match the citation in the secondary source? If not, you may be looking at a different version. Note the discrepancy before proceeding.

Read the abstract for orientation, then go to the methods and results. Your goal is not to summarize the whole paper but to answer the acceptance test you wrote in Step 1.

Work through these questions in order:

1. Design. What kind of study is it? The design sets the ceiling on what the sentence can claim.
2. Population. Who was included, and who was excluded? Compare against your sentence.
3. Exposure or intervention. What exactly was measured or delivered, and how was it defined?
4. Comparator. What was it compared with, if anything?
5. Outcome. Which outcome was reported, how was it measured, and over what time frame?
6. Result. What did the authors report, in their own terms, including uncertainty?
7. Limitations. What did the authors themselves flag as constraints on interpretation?

Keep the authors' language. If they reported an association, do not upgrade it to an effect. If they reported a result in a subgroup, do not present it as the main finding.

These four judgments are often collapsed into a single verdict, which is where weak secondary citations do the most damage. Keep them distinct.

Evidence level describes the design and its standing for the question you are asking. A randomized trial and an observational cohort answer different questions with different strengths. Record the design plainly rather than assigning a grade you cannot defend.

Applicability asks whether the study population, setting, and care context resemble the one your sentence describes. A finding in one health system or age band may not transfer to another.

Indirectness asks whether the study measured the same exposure, comparator, and outcome as your claim. A study that measured a proxy outcome is indirect for a claim about the clinical outcome.

Study limitations are the constraints the authors and readers identify within the study itself: measurement issues, missing data, short follow-up, or confounding. These limit how firmly the result supports any sentence, including a correctly worded one.

A sentence can be accurate about the result and still overreach because of applicability or indirectness. That is why the rewrite in Step 5 depends on all four judgments, not just the reported number.

Suppose a narrative review states that early mobilization shortens hospital stay in adults with community-acquired pneumonia, and cites a primary study. Your draft sentence repeats that claim and cites the primary study.

Extract the claim. Population: adults with community-acquired pneumonia. Exposure: early mobilization. Outcome: length of hospital stay. Direction: shorter.

Locate the work. You resolve the DOI and compare title, authors, and year against the review's reference. They match, so you proceed.

Read the original. You find that the study was observational, enrolled adults in a single center, defined early mobilization by time to first out-of-bed activity, and reported an association between earlier mobilization and shorter stay. The authors note that sicker patients may have mobilized later, which could explain part of the association.

Apply the four judgments. Evidence level: observational, so the design supports association rather than causation. Applicability: single center, so transfer to other settings is uncertain. Indirectness: the exposure definition is specific, so check whether your sentence implies the same definition. Limitations: confounding by illness severity is a stated constraint.

Rewrite. The original draft said early mobilization "was associated with shorter hospital stay." That direction is supported, but the sentence should carry the design and setting: "In a single-center observational study of adults with community-acquired pneumonia, earlier mobilization was associated with shorter hospital stay, though the authors noted that illness severity may have influenced timing."

The rewrite is longer, but every element is traceable to the original. If the review had claimed a causal effect, the rewrite would need to drop that framing entirely.

Use this table as a working record for each sentence you trace. It keeps the audit trail visible when a co-author or reviewer asks how you verified a claim.

A completed row is your evidence that the sentence was verified rather than inherited.

Rewriting is where the trace pays off. Adjust the sentence along four axes until it matches the original.

• Verb strength. Match the verb to the design. Observational studies support association language; trials may support stronger phrasing about the intervention's effect, but only for the outcomes actually measured.
• Scope. Narrow the population, setting, or time frame to what the study reported.
• Outcome precision. Name the outcome the study measured, not a broader one it did not.
• Qualifiers. Add the constraint the authors flagged when it materially affects interpretation.

Then run a reverse check: read your new sentence and ask whether each element could be defended by pointing to a specific part of the original. If any element cannot, cut it or soften it.

If the original does not support the claim at all, you have two options. You can attribute the claim to the secondary source instead, if that source genuinely makes it, or you can remove the claim. Do not keep the primary citation attached to a claim the primary study does not support.

Tracing one sentence often produces several papers you will need again: the original study, a related study with a different population, and the review you started from. Keeping them together saves rework.

You can save or upload papers and organize them in My Library, which keeps the traced originals alongside your other sources. When you need to ask questions based on papers uploaded to My Library or added from search results, Library QA supports that question-and-answer workflow over your saved material.

For a longer task, such as tracing several claims across a guideline's reference list, Scholar Agent fits WisPaper's literature review, inspiration, or reproduction use cases. If your verification work is part of a larger manuscript or guideline project, Research Projects gives you a place to organize that work.

Knowing when to stop tracing is as important as knowing how to start. Use these rules to decide whether to proceed, pause, or escalate.

Stop and keep the secondary citation when your sentence is background, the claim is attributed to the secondary source, and no specific primary result is being asserted.

Stop and remove the claim when the original cannot be located after reasonable searching, or when the located work does not report the outcome your sentence names. A citation you cannot verify should not carry a factual assertion.

Pause and escalate to a second reader when the original's result is ambiguous, when the authors' own limitations would materially change your sentence, or when your rewrite would anchor a guideline recommendation. A second reader can check whether your rewrite matches the source.

Escalate to a full appraisal when the sentence depends on a subgroup result, a proxy outcome, or a population that differs from your target context. These are the cases where applicability and indirectness do the most work, and a single sentence may not be the right vehicle for the claim.

Re-check identity whenever a DOI resolves to a different version than the one cited. Treat the discrepancy as unresolved until you confirm which version the secondary source relied on.

Metadata errors propagate. A review may cite the wrong year, conflate two studies by the same author, or list a conference abstract as a full paper. Crossref exposes publisher-deposited scholarly metadata through its REST API, including titles, authors, venues, dates, identifiers, licenses, funding, ORCID and ROR identifiers, abstracts when deposited, and post-publication updates from trusted sources. That makes it a useful cross-check when a citation looks inconsistent.

When you build or revise a reference list, screening the BibTeX against real academic databases helps catch references that do not correspond to real works. TrueCite supports BibTeX citation screening for exactly that purpose, flagging hallucinated references.

Metadata checks confirm identity, not claim support. A reference can be perfectly formatted and still be attached to a sentence the study does not support. Always return to the source text for the claim itself.