This page shows what was measured, who it was measured in, and what that does not settle.
What Rosiglitazone does in the body
Used to lower blood sugar in type 2 diabetes.
Rosiglitazone switches on a master control gene inside fat cells. Those cells then become better at storing fat, which pulls fat out of muscle and liver where it interferes with insulin signalling. Blood sugar falls and stays down. The same reprogramming makes the body retain salt and water, which is why heart failure gets worse, and shifts bone marrow stem cells towards making fat instead of bone, which is why fractures increase.
What happened in people
A review of 42 studies found a concerning rise in heart attacks, but the size of that rise was uncertain.
✓ Reviewed first-read answer
Where this came from
A person wrote this and a reviewer approved it against this exact record. It carries no effect size.
A reviewer approved this against this record. The reviewed-claim record carrying the exact population and effect size does not exist yet.
No source is stored against this line.
The limit that matters most
A later overall heart-safety study could not settle the heart-attack question specifically.
Where it acts
Adipocyte nucleus, principally subcutaneous adipose tissue, with secondary effects in skeletal muscle and liver
Kind of result
The kind of result is not recorded
Supervision
Professional supervision is normally required. A regulator classifies this substance in a way that restricts how it is supplied.
What the registries record it as
Its recorded molecular formula is C18H19N3O3S, weighing 357.4.
PubChem record · 77999 · read 2026-08-29
Where each sentence above came from
No recorded sentence describes the change this makes in a way that stands on its own. The page opens with what it is taken for instead, and the recorded explanation follows below.
Shown as the opening line on this page.
The limit a reviewer approved as the one that matters most here.
The four opening statements run to 103 words.
Words this page uses
Four words worth knowing first
Chosen from what this page shows, with each one explained before the word it depends on.
Stand-in result
A stand-in result is a number measured because the real result takes too long.
A picture of it, and where the picture fails
It is like judging a journey by the speedometer rather than by arriving.
Where that stops being true. Speed does predict arrival. Many stand-in results do not predict the real one.
What people get wrong. A stand-in result is often reported as the result itself.
A surrogate endpoint substituted for a clinical endpoint, valid only where the substitution has been shown to hold.
Randomisation
Randomisation means chance decides who gets which treatment.
A picture of it, and where the picture fails
It is like a coin toss deciding the groups.
Where that stops being true. A coin toss is fair once. Fairness here comes from doing it for everyone.
What people get wrong. It is thought to make groups identical. It makes them similar on average, including on things nobody measured.
Allocation of participants to arms by a chance process, so that unmeasured factors are balanced in expectation.
Confidence interval
A confidence interval is the range the true answer is likely to sit in.
A picture of it, and where the picture fails
It is like a weather forecast giving a range rather than one number.
Where that stops being true. A forecast range covers tomorrow. This one covers what the study could resolve.
What people get wrong. The middle number is read as the answer. A range crossing zero means no change is still possible.
An interval estimate that would contain the true parameter in a stated proportion of repeated studies.
Pathway
A pathway is a chain of steps inside a cell, each one setting off the next.
A picture of it, and where the picture fails
A pathway is like a row of dominoes.
Where that stops being true. Dominoes fall one way. Pathways loop back and can switch themselves off.
What people get wrong. Changing one step is read as changing the outcome. Other steps often absorb it.
An ordered series of molecular interactions producing a defined cellular change.
What happened in people◇Read from sources, not yet reviewed
What was measured, goal by goal
One row for each goal a registered study measured something for. One column for each kind of thing that could be measured.
Registered studies list 40 outcome measures that RNAWiki could read. A registered study says what someone planned to measure. It does not say what they found. 0 of the matched studies tested this substance, and 0 posted a result.
There is no single score. A strong test result and a weak life result are different facts.
Goals down the side, kinds of measurement across the top. Each cell says what kind of thing was registered, not what was found.
Goal
Life outcome
What a body can do
How a person feels
A test result
A step in the body
Harms
How long
Who was studied
Blood sugar
∅Nothing in the sources checkedNo registered study lists a life outcome for this goal.
∅Nothing in the sources checkedNo registered study lists a performance measure for this goal.
∅Nothing in the sources checkedNo registered study lists a symptom measure for this goal.
△Only a number moved12 registered test measure.
∅Nothing in the sources checkedNo registered study lists a body-step measure for this goal.
—Not recordedHarms were not a registered measure for this goal.
—Not recordedNo finished study window is recorded.
…Waiting for a reviewerWho was studied is listed further down the page.
Cholesterol
∅Nothing in the sources checkedNo registered study lists a life outcome for this goal.
∅Nothing in the sources checkedNo registered study lists a performance measure for this goal.
∅Nothing in the sources checkedNo registered study lists a symptom measure for this goal.
△Only a number moved5 registered test measure.
∅Nothing in the sources checkedNo registered study lists a body-step measure for this goal.
—Not recordedHarms were not a registered measure for this goal.
—Not recordedNo finished study window is recorded.
…Waiting for a reviewerWho was studied is listed further down the page.
Mood
∅Nothing in the sources checkedNo registered study lists a life outcome for this goal.
∅Nothing in the sources checkedNo registered study lists a performance measure for this goal.
…Waiting for a reviewer1 registered symptom measure.
∅Nothing in the sources checkedNo registered study lists a test result for this goal.
∅Nothing in the sources checkedNo registered study lists a body-step measure for this goal.
—Not recordedHarms were not a registered measure for this goal.
—Not recordedNo finished study window is recorded.
…Waiting for a reviewerWho was studied is listed further down the page.
Which registered measures put each goal on this table
Blood sugar
hba1c from baseline to week 24; diabetes; effect on hba1c levels; hba1c following two years of treatment; insulin sensitivity; plasma glucose concentration; hba1c change from baseline at week 28; hba1c
Cholesterol
pre and post intervention triglyceride levels; pre and post intervention ldl cholesterol levels; pre and post intervention hdl cholesterol levels; effect of treatment on peak ldl particle size; fasting triglyceride level
Mood
hamilton depression rating scale
Sorted by fixed word lists, version v1. A name the rules do not recognise stays unsorted rather than moving to the nearest column.
What each mark on this table means
∅ Nothing in the sources checked
No registered study lists a life outcome for this goal.
△ Only a number moved
12 registered test measure.
— Not recorded
Harms were not a registered measure for this goal.
… Waiting for a reviewer
Who was studied is listed further down the page.
What happened in people◇Read from sources, not yet reviewed
What happened in people
Each card is one study: the exact question it asked, who was in it, and whether it showed what it set out to show.
Pooled odds ratio for myocardial infarction and for death from cardiovascular causes
✓ The study showed what it set out to show
Who was studied
Nissen and Wolski meta-analysis of 42 randomised trials
How many people
42
Study design
Meta-analysis of trials longer than 24 weeks with a randomised non-rosiglitazone control
Compared against
Not recorded for this study
Kind of result
Living longer, or avoiding a major event
What was found
Myocardial infarction odds ratio 1.43 (95% CI 1.03 to 1.98), P = 0.03; cardiovascular death 1.64 (95% CI 0.98 to 2.74), P = 0.06
Repeated elsewhere
Partially Replicated
What this does not prove. Meeting one endpoint in one population does not carry to other goals or other people.
The limits of this study, and where it came from
Main limit. The authors state their principal limitation directly: no access to original source data, so no time-to-event analysis was possible. The analysis pools trials designed for glycaemic endpoints.
How far it carries. This study is recorded from the curated record, not from the registry match.
Form and route. Oral tablet, 4 to 8 mg daily
Interval reported. 95% CI 1
Written into the record, not signed off as a reviewed claim.
Primary endpoint hazard ratio 0.99 (95% CI 0.85 to 1.16); myocardial infarction 1.14 (0.80 to 1.63); heart failure hospitalisation or death 2.10 (1.35 to 3.27)
Repeated elsewhere
Replicated
What this does not prove. Meeting one endpoint in one population does not carry to other goals or other people.
The limits of this study, and where it came from
Main limit. Open-label design and sponsor adjudication were the objections that led to the FDA-required readjudication. The investigators describe the myocardial infarction data as inconclusive.
How far it carries. This study is recorded from the curated record, not from the registry match.
Form and route. Oral tablet, 4 to 8 mg daily
Interval reported. 95% CI 0
Written into the record, not signed off as a reviewed claim.
ClinicalTrials.gov NCT00379769 — RECORD (NCT00379769) · a recorded source, not a stored snapshot
Cardiovascular or unknown-cause death, myocardial infarction or stroke, readjudicated under standard and new FDA endpoint definitions
✓ The study showed what it set out to show
Who was studied
Duke Clinical Research Institute readjudication of RECORD
How many people
4447
Study design
Independent readjudication of cardiovascular endpoints in the original trial data
Compared against
Not recorded for this study
Kind of result
Living longer, or avoiding a major event
What was found
Composite hazard ratio 0.95 (95% CI 0.78 to 1.17) against 0.93 (0.74 to 1.15) originally; myocardial infarction 1.13 (0.80 to 1.59); mortality 0.86 (0.68 to 1.08)
Repeated elsewhere
Replicated
What this does not prove. Meeting one endpoint in one population does not carry to other goals or other people.
The limits of this study, and where it came from
Main limit. Only 328 additional person-years of follow-up were ascertained across 25,833 person-years, so the exercise tested the adjudication rather than extending the data.
How far it carries. This study is recorded from the curated record, not from the registry match.
Form and route. Oral tablet, 4 to 8 mg daily
Interval reported. 95% CI 0
Written into the record, not signed off as a reviewed claim.
ClinicalTrials.gov NCT00379769 — RECORD (NCT00379769) · a recorded source, not a stored snapshot
What we know
RNAWiki holds 3 written-up human studies for this substance, each with the question it asked.
How we know it
Each card names the study, the number of people and what the study set out to measure.
What this does not prove
These summaries were written by a person and have not been signed off as reviewed claims.
Why it matters
A study that failed is as informative as one that worked, and both are here.
What we still do not know
No reviewed claim exists, so no exact population, effect size or interval is published yet.
What happened in people◇Read from sources, not yet reviewed
How close this is to real life
The top step is something a person would feel or care about. The bottom is a guess from software. Filled steps are where evidence exists for this substance.
■Living longer, or avoiding a major eventEvidence recorded. Death, a heart attack, a stroke, a hospital stay.2 registered measures of this kind. No reviewed result.
□What a body can do day to dayNo evidence recorded. Walking, dressing, breathing, recovering.No registered study measures this.
□Measured performanceNo evidence recorded. How much was lifted, how far was run, how fast.No registered study measures this.
■Symptoms and quality of lifeEvidence recorded. Pain, tiredness, mood, sleep, as the person rated it.2 registered measures of this kind.
■A number that stands in for healthEvidence recorded. Cholesterol, blood sugar, bone density. A stand-in, not the thing itself.20 registered measures of this kind.
■A step measured inside a personEvidence recorded. Something measured in human tissue or human cells.A human record sits on the stored organism ladder.
■AnimalsEvidence recorded. Mice, rats, dogs. Useful for ideas, not proof about people.Stored records in Mouse, Rat, Non-human primate. A result in animals says what to test next. It does not say what happens in people.
■Cells in a dishEvidence recorded. Cells or chemistry on a bench, far from a whole body.Stored mechanism abstracts describe steps measured in cells or tissue.
□A guess from softwareNo evidence recorded. Nobody measured it. RNAWiki never publishes one as a finding.RNAWiki stores no prediction for this record. Software can suggest a link. RNAWiki does not publish one as a finding.
Higher on these steps means closer to something a person would feel. It does not mean better done.
What it changes in the body◇Read from sources, not yet reviewed
The path through the body
From what a person takes to what changes. A solid line is a step measured in people. A dashed line is a step nobody has measured that way.
Start
Rosiglitazone
What a person takes: Oral tablet, 4 to 8 mg daily.
The measurement behind this step
Oral rosiglitazone maleate, once or twice daily, alone or in fixed combination with metformin or glimepiride. Nearly complete absorption, high plasma protein binding, hepatic metabolism principally by CYP2C8 with no active metabolites of consequence.
Getting in
An oral tablet, once or twice daily
Taken by mouth, alone or alongside metformin or a sulfonylurea.
──Measured in people. Written by a person from the study named beside the step. Not signed off as a reviewed claim.
The measurement behind this step
Oral rosiglitazone maleate 4 to 8 mg daily, as monotherapy or in combination. Nearly complete absorption, high plasma protein binding, metabolised principally by CYP2C8.
It crosses into fat cells and travels to the nucleus, where the genetic instructions are kept.
──Measured in people. Written by a person from the study named beside the step. Not signed off as a reviewed claim.
The measurement behind this step
Lipophilic enough to cross the plasma and nuclear membranes, reaching the PPAR-gamma ligand-binding domain in the adipocyte nucleus, with lower expression in muscle and liver.
It docks into a master control protein that decides which genes a fat cell switches on.
╌╌Described, with no measurement named. No measurement is named for this step, so it is drawn as an unconfirmed link.
The measurement behind this step
High-affinity agonism at PPAR-gamma, which heterodimerises with the retinoid X receptor, releases corepressors, recruits coactivators, and binds peroxisome proliferator response elements in target gene promoters.
Fat cells are reprogrammed — and so are kidney and bone
Fat cells become better at storing fat, pulling it away from muscle and liver. The same switch tells kidney tubules to retain salt and marrow stem cells to become fat rather than bone.
╌╌Described, with no measurement named. No measurement is named for this step, so it is drawn as an unconfirmed link.
The measurement behind this step
Transcriptional reprogramming increases adipocyte differentiation, adiponectin secretion and GLUT4 expression, lowering ectopic lipid in muscle and liver. The same receptor upregulates epithelial sodium channel expression in the collecting duct and shifts mesenchymal lineage commitment from osteoblast to adipocyte.
Durable glucose control; heart failure doubled, and a contested infarction signal
Blood sugar falls and stays down for years. Heart failure hospitalisation roughly doubles and fractures rise in women. The heart-attack question was answered differently three times.
╌╌Described, with no measurement named. No measurement is named for this step, so it is drawn as an unconfirmed link.
The measurement behind this step
Measured: HbA1c lower at 5 years than the comparator. Measured: heart failure hospitalisation or death hazard ratio 2.10 (1.35 to 3.27). Contested: myocardial infarction odds ratio 1.43 (1.03 to 1.98) in meta-analysis, hazard ratio 1.14 (0.80 to 1.63) in RECORD, 1.13 (0.80 to 1.59) readjudicated.
No suggested links are held for this record, so nothing is hidden from this path.
What we know
The record describes 5 steps from taking it to an effect in a person.
How we know it
Each step names the study behind it in the technical detail beside it.
What this does not prove
A change inside the body is a reason to look. It is not a result in a person.
Why it matters
A reader can see exactly where the chain stops being measured in people.
What we still do not know
No reviewed claim binds any of these steps to a named population.
What it would be like to take◇Read from sources, not yet reviewed
Felt, measured, or meaningful
Three different things that get called the same word. Registered studies measured all three, and mixing them is how a blood test becomes a health claim.
Felt
Things a person could notice without a test.
hamilton depression rating scale
clinical global impression severity scale
Measured
Things only a test, a scale or a device shows.
hba1c from baseline to week 24
effect on hba1c levels
hba1c following two years of treatment
flow mediated dilation
insulin sensitivity
pre and post intervention triglyceride levels
pre and post intervention ldl cholesterol levels
pre and post intervention hdl cholesterol levels
plasma glucose concentration
hemoglobin a1c at week 24
and 10 more.
Meaningful
Things that change how a life goes, not only a number.
A registered study says what someone planned to measure. It does not say what they found. A number moving is not the same as a life going better. The two are shown apart here.
Felt, but not shown to help. A person notices a change. No study showed it leads anywhere good.
Measured, but not felt. A number moves. The person notices nothing. Both can be true.
Measured, but not shown to matter. A number moves in the good direction. Nobody showed that lives went better.
Matters, but takes years. The result that counts may take longer than anyone would keep watching.
Names that fit none of the three (16)
primary major macrovascular events
treatment failure
diabetes
safety
atheroma volume to month 18
a1c at month 6
asiiauc during a hyperglycemic clamp test
carotid intima media thickness
safety and tolerance of medications
forearm blood flow
liver/spleen at 6 months
time from randomization to the primary action
clinical response
proteinuria
novel cardiovascular risk factors modification
adverse events
These are harms, study-process measures, or names the rules did not recognise. They are shown rather than dropped.
What is missing or unclear◇Read from sources, not yet reviewed
Were people like you studied?
RNAWiki cannot tell whether a study fits you. It can show who was in it, and where the result stops carrying.
Who was studied
People similar to this profile were included.
Very few people. The restrictions were lifted, but prescribing never recovered, and the alternatives that arrived afterwards are better. Rosiglitazone is recorded in Drugs@FDA with Avandia products discontinued.
Who is missing from the studies
Which groups were under-represented in the studies has not been recorded for this substance. Pregnancy, breastfeeding, children, older adults, and people with liver or kidney conditions are commonly excluded from trials; nothing here says whether they were.
Where the result stopped carrying
A restricted access programme was imposed in 2010, four years before the readjudication that led to its removal
The trial designed to settle cardiovascular safety was open-label and sponsor-adjudicated, which is why its result was not accepted at the time
Prescribing never recovered after the restrictions were lifted; Avandia products are recorded as discontinued in Drugs@FDA
This is a scope explorer, not a diagnosis engine.
It shows who was studied so you can see whether people similar to you were included.
RNAWiki cannot tell whether a study fits you. It can show who was in it.
What it would be like to take◇Read from sources, not yet reviewed
Why it might seem to do nothing
A real effect and a noticed effect are different things. These are the recorded reasons the two come apart.
It was studied for a different goal
The studies measured something else entirely.
On this record: Some registered studies measured things that match no goal on this page.
It was studied in different people
The people in the studies were not much like the reader.
On this record: Which groups were under-represented in the studies has not been recorded for this substance. Pregnancy, breastfeeding, children, older adults, and people with liver or kidney conditions are commonly excluded from trials; nothing here says whether they were.
The evidence may simply be wrong
Small early studies often do not hold up.
On this record: RNAWiki has not yet published a reviewed conclusion for this use.
Other reasons RNAWiki checked and found nothing for (10)
A different form was studied
The studied form is not the form on the shelf. Nothing in this record points to this reason.
There was nothing to correct
Where a level is already normal, topping it up may change nothing. Nothing in this record points to this reason.
Something else had to happen too
In the studies it was paired with training, a diet or another treatment. Nothing in this record points to this reason.
The change is too small to feel
A real change can still sit below what a person notices. Nothing in this record points to this reason.
Day-to-day swing hides it
Sleep, food, stress and time of day move most numbers more than this would. Nothing in this record points to this reason.
Not enough time yet
The studies ran longer than the person has waited. Nothing in this record points to this reason.
It was not taken as studied
Missed days change the result, and studies count them. Nothing in this record points to this reason.
Something else changed at the same time
Two changes at once cannot be told apart afterwards. Nothing in this record points to this reason.
The product may not be what it says
Contents of a sold product are not always what the label states. Nothing in this record points to this reason.
No recorded reason
RNAWiki has nothing stored that would explain it. Nothing in this record points to this reason.
None of these is a reason to take more. Taking more is not a step this page ever suggests.
What it would be like to take◇Read from sources, not yet reviewed
What taking it involves
The recorded facts about getting hold of it and using it. Nothing here is a suggestion about what you should do.
How it is supplied
Withdrawn
❝ Quoted from a source
Where this came from
Copied from a source RNAWiki stored, with the source named.
Read from the recorded approval status.
No source is stored against this line.
Form and route
Oral tablet, 4 to 8 mg daily
✎ Source-linked record
Where this came from
A person wrote this into the record with the study named beside it. No reviewer has signed it off.
The form and route recorded on this substance.
No source is stored against this line.
Who oversees it
Professional supervision is normally required. A regulator classifies this substance in a way that restricts how it is supplied.
❝ Quoted from a source
Where this came from
Copied from a source RNAWiki stored, with the source named.
Suppression classes recorded: S8.
No source is stored against this line.
What is in the pack
Oral rosiglitazone maleate, once or twice daily, alone or in fixed combination with metformin or glimepiride. Nearly complete absorption, high plasma protein binding, hepatic metabolism principally by CYP2C8 with no active metabolites of consequence.
✎ Source-linked record
Where this came from
A person wrote this into the record with the study named beside it. No reviewer has signed it off.
Recorded notes on how this is sold and what that means for what is in the pack.
No source is stored against this line.
Where it is registered
Regulatory records are listed in the technical disclosure at the foot of this page.
# Counted from records
Where this came from
A count of rows RNAWiki holds. It describes our records, not your body.
Register entries are stored per jurisdiction and shown with their dates.
No source is stored against this line.
Why people stop
Not recorded.
∅ Nothing recorded
Where this came from
RNAWiki holds nothing here and says so rather than guessing.
No record of why people stopped taking it is stored for this substance.
No source is stored against this line.
What it would be like to take◇Read from sources, not yet reviewed
What can go wrong
Sorted by where each line came from. A regulator's label and a report somebody sent in are not the same kind of fact.
·Worth asking a clinician aboutWritten into the record from the studies named on this page
Heart failure is the confirmed and undisputed harm: hazard ratio 2.10 (95% CI 1.35 to 3.27) for hospitalisation or death from heart failure, driven by PPAR-gamma-mediated sodium and fluid retention, with a boxed warning against use in symptomatic heart failure. Upper and distal lower limb fractures are increased, mainly in women. Weight gain, oedema and macular oedema occur. The myocardial infarction signal remains formally unresolved: 1.43 (1.03 to 1.98) by meta-analysis, 1.14 (0.80 to 1.63) in RECORD, 1.13 (0.80 to 1.59) on independent readjudication. Access restrictions imposed in 2010 were removed in 2013 and the programme eliminated in 2015.
Nobody counted how many people took this and were fine, so this cannot be turned into a rate.
Over a long time. No completed tested study window is recorded, so long-term effects are not established by the registry record.
Groups the studies covered thinly. Which groups were under-represented in the studies has not been recorded for this substance. Pregnancy, breastfeeding, children, older adults, and people with liver or kidney conditions are commonly excluded from trials; nothing here says whether they were.
What is missing or unclear◇Read from sources, not yet reviewed
Does the exact form matter?
Evidence belongs to the exact thing that was tested. A different salt, a different mixture or a different preparation is a different question.
The form that was studied
Oral tablet, 4 to 8 mg daily
✎ Source-linked record
Where this came from
A person wrote this into the record with the study named beside it. No reviewer has signed it off.
The form used in the studies cited on this page.
No source is stored against this line.
What is sold
Nearly complete absorption, high plasma protein binding, hepatic metabolism principally by CYP2C8 with no active metabolites of consequence.
✎ Source-linked record
Where this came from
A person wrote this into the record with the study named beside it. No reviewer has signed it off.
Recorded notes on which forms are sold and how they compare.
No source is stored against this line.
What is recorded as being sold
2 products list this as an active ingredient in the United States drug directory. 2 of them contain it and nothing else.
FDA National Drug Code directory · 55111-063 · read 2026-08-29
They are sold as powder.
FDA National Drug Code directory · 55111-063 · read 2026-08-29
Evidence carries across two names for one substance. It does not carry across a salt, a mirror form or a mixture.
What it would be like to take◇Read from sources, not yet reviewed
What you could measure
This one is decided with a clinician, so this section holds questions to ask rather than a plan to run.
Self-experiment planning is not offered for a prescription or clinician-supervised medicine, withdrawn medicine. The questions below are for a clinician or pharmacist.
Questions worth asking
Which of the trials of Rosiglitazone studied people like me?
What was measured, and for how long?
Was the result a laboratory value or a health outcome?
What would we watch for, and when would we stop?
Tracking can show whether something changed for you. It cannot show what caused it.
RNAWiki records evidence. It does not say whether this substance is right for you.
What is missing or unclear◇Read from sources, not yet reviewed
Claims that go past the evidence
Things said about this substance that sound reasonable, and the exact step that is missing between the evidence and the claim.
✗Goes past the evidence
That RECORD refuted the meta-analysis on myocardial infarction — its interval, 0.80 to 1.63, contains both no effect and the meta-analytic estimate
Why it sounds right. It follows from something real on this page: a body step, an animal result or a related finding.
The missing step. Nobody measured this in people for this use, or the study that did measure it did not show it.
What would change this
A study in people, for this exact use, measuring this exact thing, that a reviewer signs off.
RNAWiki has not yet published a reviewed conclusion for this use.
✗Goes past the evidence
That the 2013 readjudication constituted new evidence; it produced almost identical hazard ratios and added 328 person-years
Why it sounds right. It follows from something real on this page: a body step, an animal result or a related finding.
The missing step. Nobody measured this in people for this use, or the study that did measure it did not show it.
What would change this
A study in people, for this exact use, measuring this exact thing, that a reviewer signs off.
RNAWiki has not yet published a reviewed conclusion for this use.
✗Goes past the evidence
That removal of the restrictions implies the drug was exonerated; the heart failure and fracture harms were confirmed throughout
Why it sounds right. It follows from something real on this page: a body step, an animal result or a related finding.
The missing step. Nobody measured this in people for this use, or the study that did measure it did not show it.
What would change this
A study in people, for this exact use, measuring this exact thing, that a reviewer signs off.
RNAWiki has not yet published a reviewed conclusion for this use.
What is missing or unclear◇Read from sources, not yet reviewed
What nobody knows yet
Open questions, each with why it is open and what would close it.
How fast does the body clear it?
The sources RNAWiki checked hold nothing for this field.
Why it matters. Without this, nothing on this page can say how long anything lasts.
What would answer it
A stored source that records it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
Missing populations
Which groups were under-represented in the studies has not been recorded.
Why it matters. Recorded by the evidence model as a gap on this record.
What would answer it
A study that measures it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
Missing long-term data
No completed tested study window is recorded.
Why it matters. Recorded by the evidence model as a gap on this record.
What would answer it
A study that measures it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
No reviewed conclusion
RNAWiki has not yet published a reviewed conclusion for this use.
Why it matters. Recorded by the evidence model as a gap on this record.
What would answer it
A study that measures it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
Missing interaction studies
No interaction was found in the registers checked. Not finding one is not the same as showing there is none.
Why it matters. Recorded by the evidence model as a gap on this record.
What would answer it
A study that measures it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
Formulation uncertainty
Several salts, forms or products of Rosiglitazone are recorded. Results from one form may not transfer to another.
Why it matters. Recorded by the evidence model as a gap on this record.
What would answer it
A study that measures it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
Mechanism not reviewed
No reviewed mechanism story exists for this substance.
Why it matters. Recorded by the evidence model as a gap on this record.
What would answer it
A study that measures it, and a reviewer to sign it off.
Last checked 2026-09-04. Searched: The sources listed in the receipts at the foot of this page.
What happened in people◇Read from sources, not yet reviewed
Check any of this yourself
Every line above traced back to the study it came from. This is the technical layer, and the vocabulary changes here.
Every line above can be traced to the study named beside it. Follow the link and read it.
Meta-analysis of 42 trials: odds ratio 1.43 for myocardial infarction
In plain words
A 2007 analysis pooling 42 trials found people on rosiglitazone had about 43 per cent higher odds of a heart attack, and a borderline increase in cardiovascular death.
What was measured
Pooled odds ratio for myocardial infarction and for cardiovascular death across 42 randomised trials
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
Nissen and Wolski searched the published literature, the FDA website and the manufacturer's trial registry, including studies of more than 24 weeks with a randomised control group not receiving rosiglitazone and available outcome data. Of 116 potentially relevant studies, 42 met criteria. Mean subject age was about 56 and mean baseline glycated haemoglobin about 8.2 per cent. Combined by a fixed-effects model, the odds ratio for myocardial infarction was 1.43 (95% CI 1.03 to 1.98, P = 0.03) and for death from cardiovascular causes 1.64 (95% CI 0.98 to 2.74, P = 0.06). The authors state their own principal limitation: no access to original source data, so no time-to-event analysis was possible.
Written into the record, not signed off as a reviewed claim
RECORD: non-inferior on the primary endpoint, and heart failure doubled
In plain words
A dedicated 4,447-patient trial found no overall cardiovascular excess, but hospitalisation or death from heart failure occurred in 61 patients on rosiglitazone against 29 on the comparison treatment.
What was measured
Hazard ratios for cardiovascular hospitalisation or death, and for heart failure, over 5.5 years
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
RECORD (NCT00379769) randomised 4,447 patients with type 2 diabetes on metformin or sulfonylurea monotherapy, mean HbA1c 7.9 per cent, to added rosiglitazone (2,220) or to metformin plus sulfonylurea (2,227), in a multicentre open-label design with a non-inferiority margin of 1.20. Over a mean 5.5 years, 321 rosiglitazone and 323 control patients reached the primary endpoint of cardiovascular hospitalisation or death, hazard ratio 0.99 (95% CI 0.85 to 1.16), meeting non-inferiority. Component hazard ratios were 0.84 (0.59 to 1.18) for cardiovascular death, 1.14 (0.80 to 1.63) for myocardial infarction and 0.72 (0.49 to 1.06) for stroke. Heart failure causing hospitalisation or death occurred in 61 against 29 patients, hazard ratio 2.10 (1.35 to 3.27), risk difference 2.6 per 1000 person-years (1.1 to 4.1). Upper and distal lower limb fractures were increased mainly in women.
Written into the record, not signed off as a reviewed claim
The readjudication produced 0.95 where the original produced 0.93
In plain words
An independent group re-reviewed every death, heart attack and stroke in RECORD using the original records. The answer came out essentially unchanged.
What was measured
Hazard ratio for the composite of cardiovascular death, myocardial infarction or stroke, readjudicated versus original
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
The FDA required a reevaluation of cardiovascular outcomes in RECORD, carried out at the Duke Clinical Research Institute. Original data were used to systematically identify all potential deaths, myocardial infarctions and strokes; site investigators were approached for additional source documents and information on participants lost to follow-up; suspected events were readjudicated under standard procedures. Follow-up for mortality totalled 25,833 person-years including 328 additional person-years found during the effort. The readjudication identified 184 cardiovascular or unknown-cause deaths (88 rosiglitazone, 96 comparator), 128 participants with myocardial infarction (68 versus 60) and 113 with stroke (50 versus 63). The hazard ratio for cardiovascular or unknown-cause death, myocardial infarction or stroke was 0.95 (95% CI 0.78 to 1.17), against 0.93 (95% CI 0.74 to 1.15) originally. Myocardial infarction was 1.13 (0.80 to 1.59) and mortality 0.86 (0.68 to 1.08), the same as before. Analyses under new FDA endpoint definitions were similar.
Written into the record, not signed off as a reviewed claim
Restrictions removed in 2013 on an analysis that confirmed the previous one
In plain words
The 2010 access restrictions were lifted after the readjudication. But the readjudication agreed with the original result, which had already existed in 2009 when the restrictions were imposed.
What was measured
Not recorded
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
The sequence is: 2007 meta-analysis finds an odds ratio of 1.43 for myocardial infarction; 2009 RECORD reports a primary-endpoint hazard ratio of 0.99 and a readjudicated composite of 0.93; 2010 the FDA imposes a restricted access programme; 2013 the Duke readjudication returns 0.95 for the same composite and the restrictions are removed; 2015 the programme is eliminated entirely. The 2013 number is not new information about the drug — the readjudicators say so, noting only a modest number of additional person-years ascertained. What changed between 2010 and 2013 is confidence in the RECORD result rather than the RECORD result itself, because the trial's open-label design and sponsor adjudication were the objections, and an independent readjudication answered those objections without altering the estimate. That is a legitimate reason to change a regulatory position and it is not the same thing as new evidence.
Written into the record, not signed off as a reviewed claim
Heart failure and fractures were never in dispute at any stage
In plain words
While the heart attack question reversed twice, the doubling of heart failure hospitalisation and the excess of fractures in women were consistent throughout and never contested.
What was measured
Hazard ratio 2.10 for heart failure hospitalisation or death; excess limb fractures in women
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
RECORD confirmed a hazard ratio of 2.10 (95% CI 1.35 to 3.27) for heart failure causing hospitalisation or death, a risk difference of 2.6 per 1000 person-years, and an increase in upper and distal lower limb fractures mainly in women. Both are mechanistically explained: PPAR-gamma agonism upregulates epithelial sodium channel expression in the renal collecting duct, causing sodium and fluid retention, and shifts mesenchymal stem cell commitment from osteoblast towards adipocyte, reducing bone formation. Neither finding was reversed by the readjudication, because neither was ever the subject of it. The public argument concentrated entirely on the disputed endpoint and left the undisputed ones almost unmentioned, which is a common and instructive distortion.
Written into the record, not signed off as a reviewed claim
RECORD was not powered for myocardial infarction and its interpretation says so
In plain words
The trial that cleared the drug of an overall cardiovascular excess explicitly said its data were inconclusive about heart attacks specifically.
What was measured
That RECORD refuted the myocardial infarction signal in the 2007 meta-analysis
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
The RECORD investigators wrote that the data are inconclusive about any possible effect on myocardial infarction, while concluding that rosiglitazone does not increase overall cardiovascular morbidity or mortality against standard glucose-lowering drugs. The myocardial infarction hazard ratio was 1.14 with a confidence interval of 0.80 to 1.63 originally and 1.13 (0.80 to 1.59) on readjudication — intervals that comfortably contain both no effect and the 1.43 point estimate from the meta-analysis. So the meta-analysis and the trial are not in contradiction on this endpoint; the trial simply lacks the events to resolve it. Reading RECORD as a refutation of Nissen and Wolski asks a non-inferiority trial on a composite endpoint to settle a component it was never designed to settle.
Written into the record, not signed off as a reviewed claim
The lasting consequence is a rule, not a verdict on the drug
In plain words
The episode changed how every diabetes drug is approved. Cardiovascular outcome trials became a requirement, and that is why the later classes have outcome data at all.
What was measured
Not recorded
Effect estimate
No reviewed effect estimate exists for this record.
Limits
See the technical detail.
From the source
The rosiglitazone controversy produced a durable change in what a glucose-lowering drug has to demonstrate: cardiovascular safety as an explicit outcome, in dedicated trials, rather than as an assumption inherited from glycaemic control. Every SGLT2 inhibitor and GLP-1 receptor agonist cardiovascular outcome trial exists downstream of that requirement, and several of them found benefit rather than mere non-inferiority. The drug that triggered this now has an unrestricted label almost nobody uses. The regulatory settlement outlived the argument that caused it, which is the most durable outcome any entry in this file has produced.
This order is fixed in code and does not count clicks or time on the page.
What is not here
4 questions this page could not answer
These sections were prepared and left out because there was nothing to put in them. They are listed rather than hidden, so the gaps in this record are visible.
How long anything takes — found nothing in the sources checked.
What it may clash with — found nothing in the sources checked.
Other ways to the same goal — found nothing in the sources checked.
What changed on this page — found nothing in the sources checked.
The record as stored
The full record, for auditing
Everything above is built from what is below. This layer keeps the technical vocabulary, record identifiers and every stored row, so a reader who wants to check the page can.
The older medicine-wide conclusion held in this record
Written for a clinical reader, and about the medicine as a whole rather than about one indication, population and set of trials. RNAWiki does not treat it as a programme conclusion and no reviewer has signed it off in that form. It is here word for word because it is what the record holds.
A PPAR-gamma agonist restricted in 2010 after a meta-analysis of 42 trials found an odds ratio of 1.43 for myocardial infarction, and released in 2013 after an independent readjudication of the RECORD trial returned a hazard ratio of 0.95 against the original 0.93 — while the harms that were never in dispute, a doubling of heart failure hospitalisation and an excess of fractures in women, remained exactly as first reported.
Recorded evidence blocks (10)
Q2
What did Rosiglitazone's largest trial (1499650 people) and its longest (16 years) measure?
1499650 people in Rosiglitazone's largest registered study, 16 years in its longest registered window, measuring Peritoneal fluid cytokine concentrations. ClinicalTrials.gov · 2026-09-01
45 phase3, 37 phase4, 33 phase2, 23 na, 22 phase1, 8 na or unstated; NCT00004180; 2015-05; no ageing endpoint recorded. Last human test completed 2021, NCT02694874.
Interpretation These counts include studies where Rosiglitazone was a comparison treatment, a background treatment or an exposure in an observational study, and the longest window may be a planned end date. Trial roles have not yet been classified for this record.
Show the evidence
phase3
45
phase4
37
phase2
33
na
23
phase1
22
na or unstated
8
1 more recorded row
Last recorded human testNCT02694874
2021-12
recorded 2026-09-01 · last checked 2026-09-04
Q3
From mouse to human: where has Rosiglitazone shown lifespan?
Which of a1c at month 6, acute insulin response to glucose and adverse events did Rosiglitazone's trials measure?
a1c at month 6, acute insulin response to glucose and adverse events lead 40 outcome terms across Rosiglitazone's trials. ClinicalTrials.gov · 2026-09-01
diabetes, death, safety, effect on hba1c levels, atheroma volume to month 18 and hba1c following two years of treatment follow.
Show the evidence
primary major macrovascular events
1
hba1c from baseline to week 24
1
treatment failure
1
diabetes
1
death
1
safety
1
14 more recorded rows
effect on hba1c levels
1
atheroma volume to month 18
1
hba1c following two years of treatment
1
flow mediated dilation
1
insulin sensitivity
1
a1c at month 6
1
asiiauc during a hyperglycemic clamp test
1
carotid intima media thickness
1
pre and post intervention triglyceride levels
1
pre and post intervention ldl cholesterol levels
1
pre and post intervention hdl cholesterol levels
1
safety and tolerance of medications
1
forearm blood flow
1
plasma glucose concentration
1
recorded 2026-09-01 · last checked 2026-09-04
Q7
What will the one ongoing trial of Rosiglitazone report, and when?
"Anti-PD-1 mAb Plus Metabolic Modulator in Solid Tumor Malignancies"; n 72; "Best overall response"; 2032-04-30
recorded 2026-09-01 · last checked 2026-09-04
Q8
Which 75 trials of Rosiglitazone posted no result?
Posted no result
75 of 75 completed trials
Registrations
NCT02526615, NCT00358124, NCT00333723, NCT00501020, NCT00018382 and NCT00044460, and 69 more
Completion dates
oldest 2001-12; newest 2021-12
Show the evidence
Trial
NCT02526615
2001-12
NCT00358124
2002-06
NCT00333723
2003-01-06
NCT00501020
2003-02-13
NCT00018382
2003-03
NCT00044460
2003-04-10
14 further recorded trials
NCT00182052
2003-08
NCT00197132
2003-08
NCT00306696
2004-01
NCT00331487
2004-03
NCT00329225
2004-04
NCT00500955
2004-06
NCT00285142
2004-11
NCT00306644
2004-11
NCT00304993
2005-02
NCT00231387
2005-05
NCT00372086
2005-09
NCT00039663
2005-10
NCT00422955
2005-10
NCT00489229
2005-12
Q9
At the median, Rosiglitazone's trials enrolled 68 people — anything larger?
Median enrolment
68
Largest enrolment
1499650
Registered trials counted
156
Q10
Was Rosiglitazone studied with fasting and exercise?
fasting and exercise are named in Rosiglitazone's label sentences: "Pioglitazone significantly reduced triglycerides and fasting insulin when compared with placebo while rosiglitazone showed a modest reduction of LH when compared with metformin." openfda-label+europepmc · 2024-04-29
2 recorded statements; fasting, exercise
Show the evidence
fasting
Pioglitazone significantly reduced triglycerides and fasting insulin when compared with placebo while rosiglitazone showed a modest reduction of LH when compared with metformin.
exercise
We conducted a three-arm, parallel-group, randomized, controlled trial to compare the effects of rosiglitazone and physical exercise on endothelial function in patients with coronary artery disease and impaired fasting glucose or impaired glucose tolerance over a 6-month period.
recorded 2024-04-29 · last checked 2026-09-04
Q11
What is recorded about Rosiglitazone and sirtuin?
"This study aimed to investigate the function of Lipin-1 in 3T3-L1 preadipocytes and its interaction with SIRT1, SRSF10, and PPARγ in promoting browning-like transcriptional responses. <b>Methods:</b> Mouse 3T3-L1 preadipocytes were treated during differentiation with either rosiglitazone (RGZ), the SIRT1 activator SRT1720, or the SIRT1…" — where Rosiglitazone and sirtuin appear together. Europe PMC · pathway abstract search · 2025-08-25
"This study aimed to investigate the function of Lipin-1 in 3T3-L1 preadipocytes and its interaction with SIRT1, SRSF10, and PPARγ in promoting browning-like transcriptional responses. <b>Methods:</b> Mouse 3T3-L1 preadipocytes were treated during differentiation with either rosiglitazone (RGZ), the SIRT1 activator SRT1720, or the SIRT1 inhibitor EX527."
autophagy
PMID 38895627
"Nuclear protein extraction and immunoprecipitation were used to detect the proteins interaction. <b>Results:</b> In this study, we reported that rosiglitazone promoted adipocyte browning and inhibited autophagy."
PMID 38895627
"Autophagy inhibition by rosiglitazone does not prevent mitochondrial clearance, which was considered to promote adipose whitening."
PMID 38895627
"Inhibition of NRF2 by ML385 reversed autophagy inhibition and the pro-browning effect of rosiglitazone. <b>Conclusion:</b> Our study linked autophagy inhibition with rosiglitazone-promoted browning of adipocytes and provided a mechanistic insight into the pharmacological effects of rosiglitazone."
mTOR
PMID 37918553
"Moreover, our experiments with the agonist (rosiglitazone) and antagonist (GW9662) of PPARγ confirm that the studied EDP acts through the PPARγ pathway affecting mTOR and finally autophagy."
PMID 36979839
"This study tested the hypothesis that rosiglitazone (RGZ) has an antidepressant impact on dexamethasone (DEXA)-induced depression by analyzing the function of the pAKT/p38MAPK/mTOR pathway and NGF through regulation of AMPK."
AMPKPMID 36979839
"This study tested the hypothesis that rosiglitazone (RGZ) has an antidepressant impact on dexamethasone (DEXA)-induced depression by analyzing the function of the pAKT/p38MAPK/mTOR pathway and NGF through regulation of AMPK."
IGF-1PMID 34485865
"Rosiglitazone, a synthetic peroxisome proliferator-activated receptor γ (PPARγ) ligand, has been reported to reduce growth hormone (GH) and insulin-like growth factor-1 (IGF-1) in 10 patients with acromegaly."
mTORPMID 36972168
"Our experiments with the PPARγ agonist (rosiglitazone) and antagonist (GW9662) suggests that TBC acted in the A549 cell line probably through activation of the mTOR-PPARγ pathway and could interfere with the p62 autophagy pathway."
AMPKPMID 35485096
"The inhibitory effect of rosiglitazone on expression of p-Akt/t-Akt, PPARa, p-FoxO1/t-FoxO1, and p-AMPK/t-AMPK was significantly (p<0.01) alleviated in the adipocytes by salicin ether."
recorded 2025-08-25 · last checked 2026-09-04
Where it is registeredIdentifiers, relations and other names
ClinicalTrials.gov — US Government work · Europe PMC — metadata only, under the recorded legal gate · derived from this page's own recorded fields; no external licence · mixed register set; per-register licences in docs/specs/corpus-20k-sources.md · openFDA / DailyMed — US Government work
✗ required summary fields resolved: 4 required field(s) not terminal: Why people use it, Best-supported result, Biggest unanswered question, Human evidence
✓ public claims reviewed: 0 reviewed claim(s); drafts are never rendered
✓ source coverage passed: 9 source rows
✓ no critical contamination: no quarantine open
✓ canonical metadata passed: slug and display name present
✓ no raw internal fields: enforced by the copy-contract test over the rendered page
This is a record of evidence. It is not medical advice, and it does not say this substance suits you. Nothing here says any substance on RNAWiki is appropriate for a child.