Briefing · 2026-10-08
Where the weak phase 3s are
9.8% of the phase-3 trials graded on this site cannot isolate the effect of the drug by design. This briefing shows what those trials have in common and where they cluster.
Every trial on this site carries a design grade: strong, moderate or weak. The grade describes how the trial was built to test the drug. A trial grades weak when it lacks what lets a reader attribute an outcome to the drug: a control arm, randomisation, blinding, or a registry record that documents them.
The figures on this page are computed from the live grades each time the page loads. Today they cover 2,212 registered phase-3 trials on assets this site tracks.
How many grade weak
216 of the 2,212 graded phase-3 trials grade weak: 9.8%. The weak trials are smaller. Their median enrolment is 149 participants; the median for the rest is 479.
Weak share by therapeutic area
The share is highest in Gastroenterology & Hepatology: 31 of 163 trials, 19.0%. Neurology has the most weak trials: 46 of 274, 16.8%. Oncology is at 4.5%. Each line shows the share with its 95% interval, and the dashed line marks the share across all areas. An area with few graded trials has a wide interval.
What the weak trials have in common
These features come from the written reasoning for each weak grade. One trial can show several.
What kind of study they are
Many weak phase-3 trials were not built to test whether the drug works. 102 are safety, tolerability or extension studies, which measure long-term exposure in patients who take the drug. 33 grade weak because the registry record does not document a control arm, randomisation or blinding.
A weak grade states what a trial can show. It does not say whether the drug works.
Why neurology has the most weak trials
Most of the weak neurology trials are long-term open-label safety studies, many of them of drugs already on the market: anti-seizure medicines, multiple sclerosis therapies, and Parkinson's disease and migraine treatments. Several test a new formulation, such as a buccal film, a nasal spray, a patch or an intestinal gel, in a single open-label group. A few were randomised and blinded but enrolled fewer than 15 patients.
The full list is below, with the reasoning for each grade.
46 weak phase-3 trials in Neurology
This was a single-group, open-label study with 447 enrolled participants and no placebo or active comparator, so the reported pain-free headache counts cannot establish how much benefit was caused by almotriptan.
The study was non-randomized, open-label, and had no comparator arm; all 800 participants received apomorphine.
This phase 3 study randomised 744 adults with episodic migraine to atogepant 60 mg daily or an investigator-selected oral standard-of-care preventive for 52 weeks, open-label.
The trial randomized 545 participants to quadruple-blinded atogepant or active-controlled topiramate, with matching placebos, and used treatment discontinuation due to adverse events as its primary clinical endpoint.
Although the trial was designed as a randomized, blinded, placebo-controlled study, it enrolled only 7 participants and was stopped early by the sponsor before anyone completed it.
This was a single-group, open-label extension of the completed double-blind GWEP1521 trial (NCT02544763), with 199 participants continuing into cannabidiol treatment and no concurrent placebo or active-control arm.
The study was open-label and non-randomized, with no placebo or active comparator, and its primary endpoint was adverse events rather than a controlled seizure or other treatment-effect comparison.
The study was neither randomized nor blinded and had no comparator arm; all participants received cenobamate with their existing antiseizure medication.
The study treated 130 participants with diazepam buccal film in a single open-label group, without randomization, blinding, or a control arm.
The study was open-label, nonrandomized, and had no placebo or active comparator, so it provides uncontrolled safety and treatment-exposure information rather than comparative evidence of effectiveness.
The study was open-label, nonrandomized, and had only one STS101 treatment arm, so its results cannot distinguish treatment effects from the natural course of migraine or other factors.
This dose-blind extension study (ENDORSE) continued dimethyl fumarate 240 mg twice or three times daily for up to about 10 years in 1,736 patients with relapsing-remitting multiple sclerosis from the DEFINE and CONFIRM trials.
The study was a single-group, open-label trial with 136 participants enrolled and no placebo or active comparator.
The study was a single-group, open-label trial with no placebo or active comparator, although it included 464 participants who rolled over from ALK870-A302.
The study treated 50 children in a non-randomized, unblinded, single-arm design with no placebo or active comparator.
The study was a single-arm, open-label safety extension enrolling 124 participants who had completed the prior MT-1186-A01 trial, with no placebo or active comparator and no formal efficacy comparison.
This trial had no randomization, no blinding, and no control group; all 244 participants received open-label ABBV-951, with primary outcomes limited to safety and tolerability endpoints.
The study assigned participants to fosphenytoin dose groups without randomization, blinding, or a comparator arm.
The study was a nonrandomized, single-group, open-label trial of gabapentin, with 90 participants enrolled and 89 treated.
Although participants were randomized, the study was open-label and compared two doses of galcanezumab without a placebo or separately blinded control arm.
The study enrolled 95 participants in a single open-label group without randomization, blinding, or a comparator arm.
This was a 52-week open-label, single-arm continuation of study 6002-014, with no randomization, blinding, or comparator group.
This was a small, single-arm, open-label study of 20 participants, with no randomization, blinding, or comparator.
This was an open-label, single-arm study with no randomization, no blinding, and no placebo or active comparator arm.
The record itself is contradictory about randomization (design field says 'non-randomized' while the participant-flow text says patients 'were randomized'), but masking is explicitly listed as none and there is no placebo or comparator arm - both study arms simply used different active doses of the same drug.
This was a 354-participant, open-label, single-group study with no randomization, blinding, or comparator arm.
This was a single-arm, open-label study with no comparator group, and its primary endpoint measured only safety and tolerability, so it makes no controlled test of whether the drug works.
The study was a single-arm, open-label trial of 35 participants, including only 5 with severe renal impairment, with no control group or blinding.
The main study gave ocrelizumab to every participant without randomization, blinding or a comparison group, so its disability outcomes cannot be compared against another treatment.
The study enrolled 132 participants and measured the clinical outcome of seizure-frequency change, but the registry does not document randomization, blinding, or a comparator arm.
The registry lists 94 participants receiving oxcarbazepine but does not identify a randomized, blinded, or comparator arm.
Although randomized and blinded, this trial was terminated after enrolling only 11 children across three arms, far too few to draw any reliable conclusion about the drug's effect on relapse, and no statistical comparison was reported.
The study was a single-group, open-label trial with 180 participants and no placebo or active comparator, so it could not provide a controlled statistical comparison of perampanel's effects.
The study was a single-arm, open-label trial without randomization, blinding, or a comparator arm, involving 91 enrolled participants.
This was an open-label study with no randomization and no concurrent placebo or active-control arm, although it included 605 participants and some had continued from earlier studies.
The study was single-arm, open-label, and had no placebo or active comparator, so its primary safety endpoint was descriptive rather than a randomized comparison.
The study was a single-group, open-label trial without randomization or a placebo or active comparator.
The study was nonrandomized and unblinded, and its comparison group received no intervention rather than placebo or active treatment.
Although this trial was randomized and triple-blinded with a placebo comparator, it enrolled only 8 participants (4 per arm), which is severely underpowered for any meaningful primary-endpoint comparison.
Although the registry reports a randomized, double-blind design with 506 participants, it lists no comparator arm, so the treatment cannot be evaluated against another treatment or placebo from the available record.
The study was an open-label, single-arm trial with 185 participants and no placebo or active comparator, so its findings are descriptive rather than a controlled comparison.
This was a single-arm, open-label study of 167 participants with no randomization, blinding, or comparator arm.
The study was nonrandomized, open-label, and had no placebo or active comparator, although 656 adolescents were enrolled.
This study was not randomized and was open-label, with no blinding; the comparison group was untreated healthy volunteers rather than a placebo-treated MS population.
This was a nonrandomized, unblinded, single-group study without a placebo or active comparator, so changes in symptoms or adverse events cannot be reliably attributed to valbenazine.
Although 165 participants were reportedly randomized and treatment was double-blinded, the study is listed as a single-group design with no placebo or active comparator.
How to read a phase-3 result
When a phase-3 result is quoted, check which design produced it. A response rate from a single-arm study and one from a randomised, blinded trial are different kinds of evidence. Each asset page on this site shows its trial grades with the reasoning.
Method
Corpus: phase-3 trials with a published design grade, on assets visible on this site, counted once per registered trial (NCT number). A trial graded on two assets counts once. Where its two grades differ (15 trials), this page uses the stronger grade.
Areas: a trial counts in every therapeutic area whose tracked indications match its conditions text, and the chart shows areas with at least 25 graded trials. 110 trials match no tracked indication and are left out of the chart only.
Features and study kinds are read from the written reasoning for each weak grade, so one trial can count under several, and some match none. "Phase 3" is the registry's own label: sponsors register some extension and safety studies as phase 3.