Reading the Study a Clinic Shows You

Published 17 min read

Document links
6
Quoted passages
9
Fact-checked
18 September 2026
Reviewed
every twelve months

Sources on this page: clinicaltrials.gov · www.ftc.gov · www.ecfr.gov · www.icmje.org · www.fda.gov

Reading the Study a Clinic Shows You

The existence of a study does not tell you how strong the evidence is. A study is evidence; the question is what kind, and how much weight it can carry. It occupies a rung on a ladder, and the whole job is working out which rung.

The hard part is not understanding the science. It is knowing what to look for, and where each thing lives. Some of it is in the abstract, some only in the methods, some in a registry record, and some in the results section or nowhere at all.

Find the paper from the citation

Start with the reference you were given. A name and a year are normally what a search needs. A registry number, written as NCT followed by digits, locates a study record directly.

If no citation is offered, that is your first finding. A claim presented as evidence based without a retrievable source has not yet been evidence based for you. Ask for the reference in writing.

1. Was it in people?

This is the fastest screen for whether a paper can directly support a claim about people, and it costs nothing to check. It does not disqualify the study; it tells you what the study is for. The FTC guidance states the point plainly:

“Animal and in vitro studies…aren’t sufficient to substantiate health-related claims”

Source: FTC, Health Products Compliance Guidance. Accessed 18 September 2026.

In vitro means in glassware, in a dish. Work in cells or in animals can be genuinely important and still not support a claim about what will happen in a person. If the paper you have been handed is a laboratory or animal study, you have learned something real about the state of the evidence, and it is not what you were asked to conclude.

2. What was the design?

Three broad shapes, and they answer different questions.

Observational. Investigators recorded what happened without assigning treatment. Useful for describing, limited for attributing cause.

Single arm. Everyone received the treatment and there was no concurrent comparison group. Without one, the design provides no internal basis for distinguishing the treatment’s effect from the passage of time, from expectation, or from other care received alongside it. Those explanations are not ruled out by the study; they are left open by it.

Controlled and randomized. Participants were assigned to treatment or comparison, ideally by chance. This is the design built to answer the causal question. The FTC guidance describes randomized controlled human trials as the most reliable form of substantiation for health benefit claims, allowing high quality epidemiologic evidence to substitute only:

“in those limited cases where: 1) it is considered an acceptable substitute for RCTs by experts in the field; and 2) RCTs aren’t otherwise feasible.”

Same source.

Also ask whether anyone was blinded. When participants or assessors know who received what, that knowledge can influence what is reported and recorded.

3. How many people, and how many finished?

Two numbers, not one.

Enrollment tells you how many started. Completion tells you how many were still being measured at the end. A study that enrolled sixty and analysed thirty is reporting on half the people it started with.

Those who left may differ from those who stayed, and that difference can pull the result in either direction. Whether you can see it depends on the paper: some report how many left, why, and how the baseline characteristics of leavers compared with completers, and some run an analysis that tests whether the conclusion survives their absence. So this field has a second part. Check whether the paper explains who left and why. If it does, you have more than a number. If it does not, record that.

A small study is not worthless. It is bounded. It can generate a signal worth testing and cannot settle a question on its own.

4. Who was actually enrolled?

Every study has entry criteria, and they are written down. Copy them out rather than summarising them; the wording is the point.

Age range. Diagnosis and how it was confirmed. Disease stage or severity. What treatments participants had already tried. What conditions excluded someone from taking part.

Where this page stops, and it stops here deliberately. Writing down who was enrolled is a records task and you can do it. Deciding whether a study’s population is close enough to your own situation for its result to mean something for you is a clinical judgment, and this page does not equip you to make it; not because the idea is difficult, but because doing it properly needs your diagnosis, your history and your examination, none of which a web page has.

So the output of this field is not a conclusion. It is a sentence you can hand to a clinician: this study enrolled people with these characteristics; how does that compare with my situation? That question is answerable by someone who knows your case. Answered at your desk, it is a guess wearing the clothes of a finding.

5. What did they measure, and was it prespecified?

A study announces in advance what it is trying to show. That is the primary outcome measure, and a study can name more than one. Where the report includes a formal sample-size calculation, check which outcome and which assumptions that calculation was built on; that tells you what the study was actually sized to answer, which is not always the same as everything it reports.

Two distinctions matter here.

Primary versus everything else. Studies can report many measures. When many are reported, differences can appear in some of them without the treatment having produced them. A positive secondary result in a study whose primary outcome was not met is a different thing from a positive primary result.

Surrogate versus clinical. A surrogate outcome is a marker, such as an imaging finding or a laboratory value. A clinical outcome is something a person experiences: pain, function, an ability regained. Improvement in a marker does not guarantee improvement in life.

A favourable result that was not named in advance should generally be read as exploratory, unless the record documents an amendment and the timing of that amendment gives it context. Protocols are sometimes changed for legitimate reasons and before anyone has seen the results, which is why the version history further down this page matters rather than the bare fact of a change.

Exploratory is not an insult. The FTC puts the consequence in a sentence:

“post hoc analysis that departs from the originally stated study protocol (e.g., an analysis that looks at various smaller subgroups of the study population) may identify areas for future exploration, but doesn’t generally provide reliable evidence to substantiate a claim.”

Source: FTC, Health Products Compliance Guidance. Accessed 18 September 2026.

So the finding is a hypothesis for the next study rather than a conclusion from this one. Different evidentiary weight, not no weight.

Where “in advance” is actually recorded, and why the obvious method gives the wrong answer. The paper is written after the results are known, so it is not the document that records what was declared beforehand.

The registry is, but not the registry page as it looks today. A registry record can be edited by whoever submitted it, including after the study has run. So if you compare the paper against the current record and they agree, you have learned nothing about prespecification: they may agree because the record was brought into line with the paper.

The version history is the part you actually need, and it is public. On ClinicalTrials.gov every record carries a Record History view listing each version with its date, and the platform tracks which versions changed the outcome measures. The study we used as a worked example is NCT02837510, and we are naming it so you can check every figure on this page against the record rather than against us. Its history had fourteen versions between 2016 and 2025, and the outcome measures changed once, in the version the site displays as ver. 14, submitted 27 February 2025. The version before it, ver. 13 of 11 December 2023, changed the study status and nothing else.

One thing to know before you open it, because it caught us. The registry’s public interface numbers versions from one, so the newest here is ver. 14. The data interface behind it numbers the same versions from zero, so the same record is version 13 there. An earlier version of this page quoted the second number, which is correct in the data and wrong on the screen you are looking at. Use the number the Record History page shows you, and if you ever compare notes with a figure from somewhere else, check which of the two counts it came from. Open the record on ClinicalTrials.gov and use the Record History view; the numbers below came from there on 18 September 2026 and the record may have moved on since.

And there is a trap inside that trap. The earliest version is not automatically what was declared before the study ran, because a study can be registered after it has already started. That is not a loophole, and it is not one rule either. Two different requirements are in play and they do not say the same thing.

The US federal deadline for an applicable clinical trial runs from enrolment:

“…must submit the clinical trial registration information … not later than December 26, 2007, or 21 calendar days after the first human subject is enrolled, whichever date is later.”

Source: 42 CFR 11.24(a). Accessed 18 September 2026.

The publication-side policy runs from before enrolment:

“the ICMJE recommends that all medical journal editors require registration of clinical trials in a public trials registry at or before the time of first patient consent for enrollment as a condition of consideration for publication. ICMJE uses the date trial registration materials were first submitted to a registry as the date of registration.”

Source: ICMJE, Clinical Trial Registration. Accessed 18 September 2026.

So a record entered after a trial is under way can still meet the federal deadline while falling outside the ICMJE policy. Whether a given timing satisfies a requirement depends on which requirement is being applied, and a registry page does not tell you which one anyone was working to. Note also that the ICMJE uses first submitted as the registration date, which is the same field step 2 above asks you to compare against the start date. But it means the first version can already be describing something in progress.

Our own worked example shows it. For NCT02837510, the registry gives the study start as May 2016, flagged as actual rather than estimated, and the first submission to the registry as 3 June 2016. The earliest version therefore post-dates the start. Anyone reading it as what was planned in advance would be wrong, and nothing on the page announces that.

So the method is five steps, and the second one is what makes the rest mean anything:

1. Note the paper’s date. Publication date, or the analysis period if the paper gives one.

2. Compare the study start date with the first submitted date. Both are on the record. If first submission comes before the start, what you can say is precise and worth saying in those words: the study was registered prospectively relative to its start date, and this is the earliest publicly registered version. That is a statement about the record, and it is as far as the record takes you. If first submission comes after, write that down too: what you have is the earliest recorded version, which is not the same as a record of what was planned.

3. Read the primary outcome from that earliest version, not from the current page.

4. Look at whether the outcomes changed afterwards, and on which dates. The Record History lists every version with its date.

5. Set the dates side by side and say only what they show.

Three fields are worth comparing across versions: the primary outcome, the enrolment number against what the paper says was analysed, and the outcome timing or time point.

A change in any of them is not proof of anything improper. Studies are amended for legitimate reasons, and the registry exists so that amendments are visible rather than silent.

Where this field stops. Whether the timing of a change carries methodological weight depends on things the registry does not show you: when the data were locked, when anyone was unblinded, whether an interim analysis had already run. You can establish the sequence; reading meaning into it is a question for someone who does this professionally. What you end up with is four facts and a question: registered on this date, started on that one, outcome changed here, paper reports this. Four facts and a question is a usable thing to carry into a conversation; an impression is not.

Source: ClinicalTrials.gov record history, read through the platform’s public interface and its API, 18 September 2026. The reader-facing route is the Record History tab on any study page.

6. How long did they follow?

Duration is the quietest weakness. A treatment for a long term condition studied over six weeks tells you about six weeks.

Ask how long participants were followed after treatment, and whether the reported benefit was still present at the last measurement.

7. How big was the effect, and how uncertain?

There is a difference between a result that is statistically detectable and a result that matters to a person. A small average change across a group can reach significance in a large study while being imperceptible to an individual.

Look for two things: the size of the difference, and the range of uncertainty reported around it. That range may be given as a confidence interval, and there are two things to read off it rather than one.

Width tells you about precision. A wider interval means the estimate is less precise, and that is all width on its own tells you.

Position is the second reading, and it is the one people skip. Where does the interval sit relative to no effect at all, and relative to a difference that would matter to you in practice? A wide interval can still sit entirely on one side of no effect. A narrow one can sit precisely on a difference too small to notice. Width and position answer different questions and neither substitutes for the other, which is why “wide interval, so it could be anything” is not a reading of the result.

If no measure of uncertainty is reported at all, write that down as the answer to this field, because it is one.

8. Who paid, and what was the sponsor’s role?

Look for a funding statement and a conflict-of-interest statement. Where a paper carries them, they sit under headings such as Funding, Disclosures or Competing interests; search the document for those words rather than scrolling. If you cannot find either, that is the answer to this field: record it as not stated rather than as none.

Funding by an interested party does not make a study wrong. It is a fact to hold alongside the others. The more specific question is what the sponsor’s role was: did they design the study, hold the data, or write the paper?

9. What phase, and what that phase is for

If the paper carries a phase designation, it tells you what question the study was built to answer. Not every study has one: the phase system belongs to the development pathway for drugs and biological products, and a study can be perfectly legitimate without sitting in it. An absent phase is not a defect, it is simply an empty row, and what fills the gap is the design field above.

Where a phase is stated, the FDA describes Phase 1 as being for “Safety and dosage,” Phase 2 for “Efficacy and side effects,” and Phase 3 for “Efficacy and monitoring of adverse reactions.”

Source: FDA, Step 3: Clinical Research. Accessed 18 September 2026.

One correction to a simplification you will encounter. It is said that Phase 1 has nothing to do with effectiveness. That is not what the FDA says. The same page describes researchers gathering “early information about how effective it is to determine how best to administer the drug to limit risks and maximize possible benefits.”

So the accurate statement is narrower: Phase 1 is primarily about safety and dosing, may gather early information about effect, and is not designed to establish clinical effectiveness.

Put the answers side by side

Write the nine answers in a column. Set out that way, the gaps are visible without anyone having to judge the science: the empty rows are the questions, and they are the ones worth asking.

A column reading randomized design · primary clinical outcome met · a year of follow up describes a different kind of study from one reading single arm · twelve people · surrogate marker · six weeks. Saying so is not a criticism of anyone; it is what the two columns say.

Field 4 stays out of this comparison, deliberately. Whether the enrolled population is close enough to your own situation is not a row you fill in and weigh; it is the row you carry to a clinician. A column can tell you how strong a study is as a study. It cannot tell you what that study means for you, and the nine fields are not built to.

If several fields cannot be filled in from what you were given, that is also a result. It means the claim has not yet been shown to rest on what it was said to rest on.

Where to go from here

If the study checks out but the marketing language around it is doing more work than the study does, that gap is examined in What “Evidence-Based” and “Evidence-Guided” Mean in Marketing Copy. If what you were offered as evidence turns out to be a patient account rather than a study, see Testimonials and Outcome Stories: Where They Sit in the Evidence Ladder.

Before you leave this page: print the nine fields as a blank list and take it with you. Filling it in takes minutes when the paper is in front of you and is very hard to do from memory afterwards.

What this does not mean

This page does not evaluate any study, and it does not tell you whether a treatment works. It describes what to look for and where to find it. Judging whether a body of evidence supports a treatment for your circumstances is a clinical matter and requires a qualified professional who knows your history.

Official sources cited on this page. Every quotation above is taken from one of these. The registry entry itself is what you check a study against, not our description of one.

What the other fields rest on, and a correction to what this page used to say here. An earlier version of this section stated that the appraisal fields were not drawn from a cited source. That was wrong, and it was wrong because we had not opened the right document rather than because no document existed. The FTC’s Health Products Compliance Guidance, which this page already cites, names all but two of them directly, and the two it does not are identified at the end of this section.

On the parameters themselves:

“In evaluating the reliability of such testing, the FTC will consider several parameters, such as sample size, duration, and outcome measures, that will vary depending on the exact nature of the hypothesis being tested and accepted norms in the relevant field. Assessing whether a study is well-designed and well-conducted, and whether the data has been properly analyzed and interpreted, are tasks that should be undertaken by someone with appropriate expertise.”

On reporting every outcome rather than the favourable ones:

“Studies that use multiple outcome measures should report all outcomes, rather than selectively reporting positive outcomes. Such studies also should include a statistical adjustment to account for the increased likelihood that, when multiple outcomes are measured, a positive result on any one of the measures may be due to chance.”

On whether the size of an effect matters to a person rather than to a statistic:

“Any statistically significant results must translate to a benefit that is clinically meaningful for consumers. Some results that are statistically significant may be too small to provide real consequences for consumer health.”

On people leaving the study:

“Subject dropout rates, non-compliance, or concurrent changes in diet or other health-related behaviors should be carefully assessed to ensure they don’t undermine any findings.”

Source for all four: FTC, Health Products Compliance Guidance, accessed 18 September 2026. Read from the guidance document itself rather than its landing page, which is what the earlier version of this section had opened.

Read the last sentence of the first quotation again, because it is the boundary this whole page works inside. Assessing a study is a task for someone with appropriate expertise. What this page gives you is the set of facts to carry to that person, not a substitute for them.

Two things on the list are still ours rather than the guidance’s: the primary-versus-secondary hierarchy and the surrogate-versus-clinical distinction in field 5. Those are stated as things to look for, not as rules from an authority.

Written and fact-checked against the sources above on 18 September 2026. No independent expert review has taken place yet, and this page says so rather than implying one. This page is reviewed every twelve months.

Leave a comment

Your email address will not be published. Required fields are marked *