Open-Label vs Blinded: What You Can Conclude

What masking is for

A blinded trial is not trying to be rigorous for its own sake. It is trying to remove a specific set of alternative explanations for its own result.

If a participant knows they got the active drug, their report of pain, energy, sleep or recovery is measuring the drug plus their expectation of the drug. Blinding does not make the expectation go away. It puts an equal one in the other arm.

The five labels, decoded

ClinicalTrials.gov records carry a masking field with five values, plus a list of who was masked.

None — an open-label trial. Everyone knows who got what.

Single — one group is masked, usually the participants.

Double — two groups. On the retatrutide phase 2 record, NCT04881760, that is participants and investigators¹.

Triple — three groups.

Quadruple — participants, care providers, investigators and outcome assessors. STEP 1, NCT03548935, is quadruple-masked².

The label alone is not the whole story. The record for the BPC 157 hamstring trial, NCT07437547, adds a masking description: identical-appearing prefilled syringes prepared by an investigational pharmacy under a randomisation code, and MRI injury volume assessed by blinded central radiology review³. That is a much more specific claim than the word "quadruple", and it is the sort of detail worth reading for.

The uncomfortable evidence about blinding itself

Here is where an honest page has to slow down.

The largest study of whether blinding actually changes measured treatment effects — MetaBLIND, covering 142 meta-analyses and 1,153 trials from the Cochrane Database — found no evidence for an average difference in estimated treatment effect between trials with and without blinded patients, healthcare providers or outcome assessors⁴.

For trials not reported as double blind versus those that were, the ratio of odds ratios was 1.02, with a 95% credible interval of 0.90 to 1.13, across 74 meta-analyses⁴.

The authors' own reading is worth quoting in shape if not in full: these results could reflect that blinding is less important than often believed, or they could reflect limitations of the method, such as residual confounding and imprecision. They recommend replication, and they say blinding should remain a methodological safeguard⁴.

So the correct position is not "open-label trials inflate results by X". It is that the average inflation, if any, has not been demonstrated, while the specific mechanisms by which an open-label design can mislead remain entirely real and are visible case by case.

Where the mechanism bites hardest

Subjective primary endpoints. Pain scores, recovery ratings, sleep quality, "energy". These are reports, and a report is made by a person who knows something.

Decisions made by clinicians. Return-to-sport dates, hospital discharge, dose escalation, rescue medication. Someone chooses, and the chooser has a belief.

Everything that follows from expectation. Dropout, adherence, co-interventions, how hard a physiotherapist pushes.

Against that, blinding matters much less for a laboratory value read by a machine from a coded sample, or for all-cause mortality.

This is the practical test: name the primary endpoint, then ask whether a person who knew the assignment could have nudged it. If the answer is yes, the masking field is load-bearing.

Worked: the record that shows the problem

Consider NCT07752381, registered in 2026 for a study completed in November 2025. It is interventional, 40 participants, allocation not applicable, a single group, masking none. The primary outcome measures include change in high-sensitivity C-reactive protein and interleukin-6 — laboratory values — alongside change in self-reported swelling⁵.

Every person in that study received the product. There is no comparison arm. The self-reported measures are made by participants who know exactly what they took, over eight weeks, in a design with no way to separate the product from time, from rehabilitation, from regression to the mean, or from expectation.

None of which makes it a bad study. It is a legitimate design for its purpose. It simply cannot support a sentence of the form "this reduced inflammation", and if such a sentence is being built on it, the design is the reason to doubt it.

An extension is not automatically unmasked

A common shortcut is to assume that the long-term data from any drug programme are open-label. Sometimes they are. Often they are not, and it is worth checking rather than assuming.

The tesamorelin extension study, NCT00608023, ran 263 participants for a further period with quadruple masking retained, and re-randomised participants across tesamorelin-to-tesamorelin, tesamorelin-to-placebo and placebo-to-tesamorelin sequences, with fasting glucose and oral glucose tolerance at week 52 as its primary endpoints⁶.

That design is why its 52-week metabolic data can be read as a comparison rather than a time series. Our tesamorelin page covers what that programme did and did not establish.

What an open-label result is good for

Generating a hypothesis worth testing properly.

Measuring things a person cannot influence — pharmacokinetics, an assay, an imaging measurement read blind.

Describing safety events that are unambiguous and serious.

What it is not good for is settling whether a compound works, when the outcome is something a hopeful person reports about themselves.

The reading rule

Open the registry record. Find the masking field, then find the primary endpoint. If the endpoint is subjective and the masking is none, the trial has measured the drug and the expectation together and has no way to tell you which is which. That is not a small caveat — it is the whole conclusion. The same discipline applies when you read any registry record.