Trial Registry vs. Journal Article: Which Record Answers Which Question?

A registry entry and a journal article about the same trial are not two versions of one record. They are produced at different times, by different hands, for different purposes. The registry entry is written before the results exist and describes what the investigators planned to do. The journal article is written after the data are in and describes what was actually done and found. When you are answering a focused evidence question — for a guideline, a systematic review, or a scholarly argument — the useful move is to read them as a pair: compare planned outcomes and methods against reported analyses and results, then reconcile whatever does not line up.

This article sets out a practical comparison for that task, using a single running example, and shows where each record type falls short on its own. It closes with a decision table and a combined workflow. It is written for clinicians and researchers appraising evidence, not for making individual treatment decisions.

Before choosing a record, name the question. Two questions that look similar pull you toward different sources:

• "What did this trial set out to measure, and how?" This is a question about design and intent. The registry entry is the natural first stop, because it captures the plan.
• "What did this trial report, in which population, with what result?" This is a question about findings. The journal article is the natural first stop, because it carries the analyses and results.

Most real evidence questions need both. A guideline panel asking whether an outcome was prespecified, or a reviewer checking whether a reported subgroup was planned, is asking a question that no single record answers cleanly. That is the reconciliation task.

To keep the comparison concrete, imagine a focused question: in adults with a given chronic condition, does a specified intervention reduce a named clinical event over 24 months, compared with usual care?

Hold four appraisal criteria constant across both record types:

1. Population — who was enrolled, and how were they defined?
2. Outcome — which outcome, measured how, at what time point?
3. Method — design, comparator, and analysis approach.
4. Result — direction, magnitude, and the analysis it came from.

Every comparison below is judged against these four criteria. The point is not that one record is better, but that each record answers some of these criteria well and others poorly.

The registry entry is strongest exactly where the journal article is weakest: it fixes the plan before the results are known. For the running example, the registry is where you look to establish:

• the primary outcome as originally specified, and any secondary outcomes;
• the planned sample size and the planned analysis;
• the eligibility criteria that defined the population;
• the comparator the trial intended to use.

This is what makes the registry useful for questions about prespecification. If a journal article reports an outcome as primary, the registry lets you check whether that outcome was primary in the plan.

The registry entry fails on the result criterion. It describes intent, not findings, and it does not carry the analyses and results you need to answer a question about effect. It also fails on method detail: the level of methodological description in a registry entry is not the level of a full report. And it can fail on currency — a plan can be amended, so a single snapshot of a registry entry may not reflect every change. For the running example, the registry alone cannot tell you whether the intervention reduced the clinical event; it can only tell you that the trial intended to measure it.

The journal article is strongest on result and on method detail. For the running example, it is where you find:

• the analyses actually performed, including any that were not in the plan;
• the results for the primary and secondary outcomes;
• the population as enrolled, which may differ from the population as planned;
• the limitations the authors themselves identify.

This is what makes the article the right source for a question about findings, and for judging study limitations.

The journal article fails on the prespecification question. A published report describes what was done; it does not, by itself, tell you what was planned before the data were seen. If the article reports a subgroup analysis or a switched primary outcome, the article alone cannot tell you whether that analysis was anticipated. It also fails on completeness of the plan: not every planned outcome is necessarily reported, and the article does not always make the gap visible. For the running example, the article can tell you the reported effect on the clinical event, but not on its own whether that was the outcome the trial set out to test.

Read the table as a division of labour, not a ranking. Each column is strong where the other is weak, which is precisely why the reconciliation step exists.

When the two records disagree, the disagreement is itself information. Work through it in a fixed order so you do not anchor on either record.

Compare the primary outcome in the registry with the primary outcome in the article. If they match, the prespecification question is largely settled for that outcome. If they differ, note the difference explicitly rather than silently adopting the article's framing.

Compare planned eligibility with the population actually enrolled and analysed. A reported result applies to the population in the analysis, so the population criterion must be settled before the result criterion means anything for your question.

Compare the planned analysis with the analysis reported. Note where the article reports an analysis that the plan did not specify, and where a planned analysis is not reported.

Only now read the result. State it against the outcome and population you have reconciled, not against whichever record you happened to open first.

These are different judgements and should be recorded separately:

• Evidence level concerns the design and the strength of the inference the study supports.
• Applicability concerns whether the reconciled population, intervention, and outcome match your question.
• Indirectness concerns how far the study's population, intervention, comparator, or outcome departs from the one you care about.
• Study limitations concern the shortcomings of the study itself, including those the authors report.

A trial can be at a strong evidence level and still be indirect for your question, or direct and still limited. Keeping the four judgements apart prevents a discrepancy in one from being mistaken for a verdict on another.

Reconciliation only works if you retrieve both records. That is a search-design problem, and the standard guidance is explicit about how to approach it.

The Cochrane Handbook (Chapter 4) recommends planning searches from your eligibility criteria, using multiple sources, combining free-text terms with subject headings, documenting the process, deduplicating records, screening in stages, linking multiple reports of one study, and rerunning searches before publication. Two of those recommendations matter most here: using multiple sources, because a registry and a journal article may not sit in the same index, and linking multiple reports of one study, because the registry entry and the article are two reports of the same trial and should be treated as one study, not two.

PubMed supports the mechanics: field tags, phrase searching, filters, Advanced Search, saved searches, Similar Articles, citation lookup, and links to full text when available, with search details showing how terms were translated and combined. MeSH — NLM's controlled and hierarchical vocabulary — supplies preferred headings, entry terms, scope notes, qualifiers, and tree positions that you can use to expand or focus a strategy. A well-built strategy uses both free-text terms and subject headings, so that a trial is caught whether it is described in registry language or in article language.

When you need to find the papers themselves, WisPaper's scholarly search is a place to run that search, and the Scholar Search entry point explains the paper-finding route for readers new to it.

Put the pieces in order and the task becomes repeatable.

1. State the question in population, intervention, comparator, and outcome terms, and decide whether it is a design question, a findings question, or both.
2. Plan the search from eligibility criteria, combining free-text and subject headings, and plan to draw on more than one source so registry and article records can both surface.
3. Retrieve and deduplicate, then link reports of the same study so the registry entry and the article are handled together.
4. Extract the plan from the registry: primary outcome, population, comparator, analysis.
5. Extract the report from the article: analyses, results, enrolled population, stated limitations.
6. Reconcile using the five steps above, recording each discrepancy rather than resolving it silently.
7. Judge separately: evidence level, applicability, indirectness, study limitations.
8. Rerun the search before you finalise, so a newer report or an amended entry is not missed.

For steps 3 and 4, keeping the records together is the practical bottleneck. My Library is where you can save or upload the papers and organise them, so the registry-linked article and its companion reports stay in one place while you reconcile. When you want to interrogate the saved set directly — for example, to check what a group of saved papers says about a planned versus reported outcome — Library QA lets you ask questions based on papers uploaded to My Library or added from search results.

For longer reconciliation work that spans many studies, Scholar Agent is the area for extended research tasks, including literature review. If the reconciliation is part of a larger piece of work with its own scope, Research Projects is where that project lives.

Use the left column to choose your entry point and the right column to close the loop. No row is answered by one record alone.