Reading a clinical trial

Does the paper report the trial’s registered primary outcome?

Compare the dated trial registration with the paper and its supplement. Check the outcome definition, time point and analysis population before citing the result.

The short answer

A paper can report the right trial and still answer a different question from the one you want to cite. Put the dated registry entry beside the article. Compare what was measured, how it was measured, when it was measured and who was included.

Start with the trial ID and the date of the plan

Look for the registration number in the abstract or Methods. For a trial on ClinicalTrials.gov, search its NCT number. If the paper names another registry, open that record instead. Check the intervention, comparator and population, then establish whether you are reading the main report, a subgroup analysis or a follow-up. More than one paper can cite the same trial ID.

In ClinicalTrials.gov, use Record History to find the version available before enrolment began, if there was one. Save its date, the trial start date and any later amendment that changes the outcome. If the first public entry came after enrolment started, record that timing. To establish what was planned earlier, look for a dated protocol or statistical analysis plan.

Build a comparison you can check line by line

Use a separate row for each primary or co-primary outcome. In every cell, keep a registry version and field or a paper page and table number. Check the supplement before marking an item as not located. The ClinicalTrials.gov definitions distinguish the outcome measure from its time frame.

CheckRecord on both sidesQuestion to resolve
RolePrimary, co-primary, secondary or exploratoryDid its stated role change?
Definition and metricInstrument, units, threshold, mean change or proportionWould these summaries answer the same question?
Time pointExact follow-up time or assessment windowIs a shorter follow-up being highlighted?
Analysis populationWho was included and how many were analysed in each arm, where reportedDo exclusions or missing data change the denominator?

Same symptom, different endpoint

Fictional example: the registry specifies mean change from baseline on a symptom scale at week 12. The abstract highlights the percentage of participants who improved by at least two points at week 4. Both refer to the same symptom, but the statistic and time point differ. Search the results tables and supplement for the week-12 mean change.

If Table 2 reports that original analysis and identifies the week-4 result as secondary, cite Table 2 for the registered outcome. If you cannot locate it, record the files and sections checked and the question still open. A documented amendment, a secondary publication or incomplete reporting can produce different findings in this check.

In a study published on 19 July 2019, Chen and colleagues found at least one primary outcome change in 130 of 389 trials with a clearly described, prospectively registered primary outcome. That proportion describes their sample. It cannot tell you why the wording differs in the paper in front of you.

Try the check with a paper you are reading

“Use only the dated registry extract, protocol and article files I supplied. For each primary outcome, quote a short identifying phrase from each source and give the version date, page, table or field. Compare role, metric, time point and analysis population. Mark a match, a documented amendment, a difference that needs explanation or an item you could not locate. For missing evidence, name the next document to check. Do not infer the authors’ motives.”

Keep the dated documents together in MedSafeAI’s AI Library. Try the prompt on one outcome, open the cited table yourself and save a one-sentence conclusion with its source location.

Sources & further reading

Follow the original sources to check the details. Product pages describe their own products; they are not independent clinical evaluations.

  1. ClinicalTrials.gov: protocol registration data definitions clinicaltrials.gov
  2. ClinicalTrials.gov: how to read study results and record history clinicaltrials.gov
  3. Chen et al.: registered and published primary outcomes (2019) jamanetwork.com

For healthcare professionals. AI responses and agreement between models do not replace source verification or clinical judgment.

PUT IT INTO PRACTICE

Bring your next question to MedSafeAI.

Explore the models, compare perspectives and keep the evidence in view. Use MedSafeAI on the web, iPhone, iPad and Android.

See plans

Keep exploring

Journal