GLP-1 Evidence

Not medical advice. A record of published research and our assessments of it. The limits

Impact of semaglutide on high-sensitivity C-reactive protein: exploratory patient-level analyses of SUSTAIN and PIONEER randomized clinical trials

Mosenzon O, Capehorn MS, De Remigis A, Rasmussen S et al. · Cardiovascular diabetology · Sep 2, 2022
DOI 10.1186/s12933-022-01585-7 · PubMed 36056351 · NCT02906930 · NCT02863328 · NCT02827708 · NCT01885208

Design
Randomized trial · 2482 participants · Mixed outcome
Match to healthy normal-weight adults aged 55–75
Different population[Auto] Participants had type 2 diabetes (age/BMI not reported in abstract).
Could weight loss explain it?
Specifically tested[Auto] Abstract addresses weight-loss independence: "Mediation analyses assessed the effect of change in glycated hemoglobin (HbA1c) and/or change in body weight (BW) on hsCRP reductions."
Study tier
Study tier 1[Auto] Large randomized trial with clinical outcomes (auto-provisional; risk of bias and consistency not yet assessed).
Assessment
Version 1 · automatic, not yet reviewed by a person · Sep 13, 2026

Study facts come from the paper. Population match, weight-loss explanation and study tier are our judgments, made against the reference group of healthy normal-weight adults aged 55–75.

[Auto, unreviewed; quoted from abstract conclusions] Semaglutide reduced hsCRP ratios-to-baseline versus comparators in subjects with type 2 diabetes (not significant with CKD). This effect was partially mediated via reductions in HbA1c and BW and potentially by a direct effect of semaglutide. Semaglutide appears to have an anti-inflammatory effect, which is being further investigated in ongoing trials.

01Findings

What the study reported

Drugs
Semaglutide, Exenatide
Dose
3.0 mg
Route
oral
Comparator
placebo
Primary outcome
Not extracted
Effect
Not extracted
95% confidence interval
Not extracted
Follow-up
Not stated
Adverse events
Not extracted
Limitations
Auto-classified from abstract only; effect estimates, adverse events and limitations not extracted. Requires manual review.

Who was studied

Diabetes status
type 2 diabetes present (all or most)
Ckd status
chronic kidney disease present in population (see abstract)
Baseline condition
chronic kidney disease
Sample size
2482

Study quality details

Study design
Randomized controlled trial
Sample size
2482
Randomization
yes
Blinding
not stated
Comparator
placebo
Follow up duration
not stated
Outcome type
mixed
Replication
not assessed (auto)
Consistency with other evidence
not assessed (auto)
Population applicability
INDIRECT
Statistical precision
CI reported: 95% CI]: 0
Funding conflicts
unclear
Peer review status
yes
01bRisk of bias

How much the result can be relied on

Risk of bias is judged per result, not per paper: the same study can report one marker at low risk and another at high. Two reviewers assess each result independently against a written guide, and every domain judgment below carries its reasoning and the sentences it rests on.

These words are not quality scores. On the RoB 2 scale, High risk of bias is the worst rating a result can receive and Low risk of bias is the best — the opposite of how the same words read on some other scales. Levels are always written out in full here for that reason.

Assessed, not settled (8)

Recorded rather than omitted. Leaving these out would make the appraisal look cleaner than it is, and a disagreement between two careful readers is itself worth knowing.

hs-CRP, ETR to baseline, oral semaglutide 7 mg and 14 mg vs comparator (both maintenance doses reported separately) · week 26 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		As O1_SUSTAIN3. Item 6 presumption applied: the parent trial is named and the paper gives registration numbers, so D1 is judged at the parent-trial level. Concealment is PRESUMED from the large registered industry phase 3 character of PIONEER 1, not verified — the retrieved sources describe no sequence generation or IVRS/IWRS mechanism, and the parent report (Aroda 2019, Diabetes Care, reference 19) was not retrieved. Arm-level baseline characteristics are not reported (Table 1 pools three trials and splits by hs-CRP cutoff, not by arm), so baseline balance cannot be checked to test the presumption.
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
PIONEER 1: randomized clinical trial of the efficacy and safety of oral semaglutide monotherapy in comparison with placebo in patients with type 2 diabetes
D2 · Deviations from intended interventions
High risk of bias		Same trial-product-estimand problem as SUSTAIN 3: the reported estimate deliberately excludes post-rescue and off-treatment observations, so it does not estimate the effect of assignment, and no ITT estimate is available. Numbers excluded, and rescue and discontinuation rates by arm, are not reported. Blinding is claimed (and PIONEER 1 is genuinely placebo-controlled, so the open-label contradiction that affects SUSTAIN 3 does not apply here), but the estimand issue is independent of blinding and is sufficient for High.
Data were evaluated for subjects receiving trial drug without rescue medication (trial product estimand), to avoid confounding effects from other antihyperglycemic drugs.
once-daily oral semaglutide 3, 7, or 14 mg or placebo (PIONEER 1)
D3 · Missing outcome data
Some concerns about risk of bias		As O1_SUSTAIN3: 99.6% baseline availability pooled across all four trials, but no per-trial or per-arm end-of-treatment availability is reported, so neither the 20% nor the 5-point prong can be evaluated. MMRM MAR assumed, no MNAR sensitivity analysis. NI on the outcome-availability question rather than a demonstrated problem, hence Some concerns not High.
A total of 2471 of 2482 randomized subjects (99.6%) had an hsCRP measurement at baseline and were included in the analyses.
In an MMRM, the missing values are automatically accounted for using a missing-at-random assumption.
D4 · Measurement of the outcome
Low risk of bias		Central-laboratory hs-CRP, identical in both arms, no assay change within the included trials (the assay bias affected only the excluded SUSTAIN 2/4/5). Objective biomarker; guide D4 rule gives Low.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Item 7: the authors self-describe the analysis as exploratory and state that no multiplicity adjustment was performed — affirmative evidence of non-prespecification, which reaches High rather than the Some concerns floor. Additionally specific to PIONEER 1: the paper reports two semaglutide dose arms (7 mg and 14 mg) separately for O1 but pools them for the mediation analysis, with no stated rule for when doses are pooled; and PIONEER 1 is the one trial whose tertile interaction was significant (p = 0.0074) out of a large family of unadjusted interaction tests across cutoffs, tertiles, sex, statin use and eGFR, which is what an unadjusted multiplicity environment predicts.
A p-value of < 0.05 was considered significant; no adjustment for multiplicity was performed.
the ETRs for semaglutide versus comparators were not affected by hsCRP tertile at baseline, as indicated by the interaction p-values for each trial, which were all nonsignificant, except for PIONEER 1 (p interaction : 0.0074; Additional file 1 : Fig. S1)

hs-CRP, ETR to baseline, oral semaglutide 14 mg vs comparator · week 52 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Item 6 presumption applied at parent-trial level; PIONEER 2 is named and the paper supplies registration numbers. Concealment PRESUMED from the trial's large registered industry phase 3 character, not verified: no retrieved source states the sequence generation or concealment mechanism, and the parent report (Rodbard 2019, Diabetes Care, reference 20) was not retrieved. Arm-level baseline balance is not reportable from Table 1.
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
Oral semaglutide versus empagliflozin in patients with type 2 diabetes uncontrolled on metformin: the PIONEER 2 trial
D2 · Deviations from intended interventions
High risk of bias		Trial-product estimand again — the reported ETR excludes off-treatment and post-rescue observations and so does not estimate the effect of assignment; no ITT estimate is reported. As with SUSTAIN 3, the paper's blanket "double-blind" description sits uneasily with the design (oral semaglutide vs an SGLT-2 inhibitor), and no arm-level discontinuation or rescue figures are given to bound the impact.
Data were evaluated for subjects receiving trial drug without rescue medication (trial product estimand), to avoid confounding effects from other antihyperglycemic drugs.
once-daily oral semaglutide 14 mg or once-daily sodium–glucose cotransporter-2 (SGLT-2) inhibitor empagliflozin 25 mg (PIONEER 2)
D3 · Missing outcome data
Some concerns about risk of bias		As above — end-of-treatment hs-CRP availability is not reported per trial or per arm, so neither D3 prong is evaluable; NI on availability. PIONEER 2 has the longest follow-up among the PIONEER trials here (52 weeks), which makes the unreported attrition more consequential, not less. MMRM MAR, no MNAR sensitivity analysis.
A total of 2471 of 2482 randomized subjects (99.6%) had an hsCRP measurement at baseline and were included in the analyses.
hsCRP was measured at weeks 0 and 56 in SUSTAIN 3; weeks 0 and 26 in PIONEER 1; weeks 0, 26, and 52 in PIONEER 2; and weeks 0, 8, and 26 in PIONEER 5.
D4 · Measurement of the outcome
Low risk of bias		Objective central-laboratory hs-CRP, same assay and procedure in both arms, no within-trial assay change.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Item 7: self-declared exploratory analysis with no multiplicity adjustment is affirmative evidence of non-prespecification. The emphasised ratio-to-baseline result is significant while the corresponding absolute-change analysis for this trial is not and is placed in the supplement — the transformation-selection pattern the guide's D5 rule names, here on top of the exploratory statement.
A p-value of < 0.05 was considered significant; no adjustment for multiplicity was performed.
In the other trials, there was a trend for a negative absolute hsCRP change from baseline with oral semaglutide 14 mg versus placebo (PIONEER 5) and with semaglutide (s.c. 1.0 mg or oral 14 mg doses) versus active comparators (SUSTAIN 3 and PIONEER 2, respectively), although these differences did not reach statistical significance (p > 0.05; Additional file 1 : Fig. S4).

hs-CRP, ETR to baseline, oral semaglutide 14 mg vs comparator · week 26 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Item 6 presumption applied at parent-trial level; PIONEER 5 is named and registered. Concealment PRESUMED from the trial's large registered industry phase 3 character, not verified — no mechanism is described in any retrieved source and the parent report (Mosenzon 2019, Lancet Diabetes Endocrinol, reference 21) was not retrieved. PIONEER 5 baseline characteristics are given separately (Table 1) but still split by hs-CRP cutoff rather than by arm, so arm-level balance cannot be checked.
Efficacy and safety of oral semaglutide in patients with type 2 diabetes and moderate renal impairment (PIONEER 5): a placebo-controlled, randomised, phase 3a trial
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
D2 · Deviations from intended interventions
High risk of bias		Trial-product estimand, not an assignment-effect analysis; no ITT estimate reported. This matters more here than elsewhere: PIONEER 5 is the CKD population with the longest diabetes duration (mean 14 years) and the highest background medication burden, so differential rescue and discontinuation are most likely precisely where the estimand excludes them, and none of those numbers are reported.
Data were evaluated for subjects receiving trial drug without rescue medication (trial product estimand), to avoid confounding effects from other antihyperglycemic drugs.
PIONEER 5 data were analyzed separately because they included subjects with CKD (baseline mean eGFR 48 mL/min/1.73 m 2 ) who had longer disease duration (mean 14 years) [ 21 ].
D3 · Missing outcome data
Some concerns about risk of bias		Table 1 shows 324 in the full analysis set and 321 with baseline hs-CRP, so baseline availability is near-complete for this trial. End-of-treatment availability is again not reported, per arm or overall, so the D3 prongs are not evaluable. MMRM MAR, no MNAR sensitivity analysis. Note the eGFR subgroup analysis is reported with "data not shown" for its interaction, which is a reporting-completeness signal in the same direction.
Full analysis set, N 2158 324
Randomized with hsCRP measured at baseline, n 2150 394 750 1006 321 45 118 158
D4 · Measurement of the outcome
Low risk of bias		Objective central-laboratory hs-CRP, same assay both arms, no within-trial assay change.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Item 7: self-declared exploratory, no multiplicity adjustment. Specific to PIONEER 5, the headline result is null and the paper repeatedly reframes it as directional — "a nonsignificant trend of reduction" in Results, and "hsCRP ratio to baseline was reduced at end-of-treatment" in Discussion, which states a within-arm change as though it answered the between-arm question. That is selective emphasis on the reported result and belongs in D5.
In PIONEER 5, there was a nonsignificant trend of reduction, as the ETR at week 26 for oral semaglutide versus placebo was 0.83 [95% CI, 0.67–1.03]; p = 0.08; Fig. 1 .
Mean baseline hsCRP in PIONEER 5 was similar to that in other trials, and hsCRP ratio to baseline was reduced at end-of-treatment.

hs-CRP, estimated treatment ratio (ETR) to baseline, semaglutide s.c. 1.0 mg vs comparator · week 56 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Item 6 presumption applied. The parent trial is identified by name and the paper supplies registration numbers, so D1 is judged on the parent trial's randomization. Allocation concealment is PRESUMED from the character of a large registered industry phase 3 programme — it is NOT verified: no retrieved source describes the randomization sequence generation or any IVRS/IWRS mechanism, and the parent report (Ahmann 2018, Diabetes Care) is not among the retrieved sources. Baseline data are reported only pooled across three trials and split by hs-CRP cutoff, not by treatment arm, so the presumption's rebuttal check (baseline imbalance between arms) cannot be performed. That unverifiable arm-level baseline balance, on top of a presumed rather than described mechanism, keeps this at Some concerns rather than Low.
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
ClinicalTrials.gov identifiers: NCT01885208 (first registered June 2013), NCT02906930 (first registered September 2016), NCT02863328 (first registered August 2016), NCT02827708 (first registered July 2016).
D2 · Deviations from intended interventions
High risk of bias		Two problems. (a) The analysis is explicitly not an effect-of-assignment analysis: data were restricted to the trial-product estimand ("on-treatment without rescue medication"), which excludes observations after discontinuation or rescue. Since the effect of interest here is assignment (guide rule 4; no ITT hs-CRP estimate is reported anywhere in the paper), an appropriate analysis to estimate the assignment effect was NOT used, and the exclusion is tied to a post-randomization event (rescue/discontinuation) that plausibly differs by arm for a GLP-1 RA. No count of excluded participants or of rescue use per arm is given, so the impact cannot be bounded. (b) Blinding is internally contradicted for this trial: the paper calls all four trials double-blind, while its own cited parent report describes SUSTAIN 3 as open-label. For a central-lab biomarker, awareness matters little in itself (guide D2 rule), but it bears on differential rescue and discontinuation, which is exactly what the estimand excludes. D2 is High on (a).
Data were evaluated for subjects receiving trial drug without rescue medication (trial product estimand), to avoid confounding effects from other antihyperglycemic drugs.
Observed values from the on-treatment without rescue medication period were included in the MMRM.
D3 · Missing outcome data
Some concerns about risk of bias		Baseline availability is near-complete (2471/2482, 99.6%), but the guide's D3 thresholds are applied to availability of the OUTCOME, i.e. at end-of-treatment, and that is never reported: the paper gives no per-arm count of participants with an end-of-treatment hs-CRP value, no completion or withdrawal numbers, and no arm-by-arm missingness comparison. Neither threshold prong can therefore be evaluated, so this is NI on "data available for nearly all", not a Low. MMRM under MAR handles missingness under an assumption that is stated but not tested; no sensitivity analysis for value-dependent (MNAR) missingness is reported. Under calibration ruling 2 a missing MNAR sensitivity analysis alone would be a limitations note — but here the prior question of how much is missing is unanswerable from the retrieved sources, which is what drives Some concerns.
A total of 2471 of 2482 randomized subjects (99.6%) had an hsCRP measurement at baseline and were included in the analyses.
In an MMRM, the missing values are automatically accounted for using a missing-at-random assumption.
D4 · Measurement of the outcome
Low risk of bias		hs-CRP is an objective central-laboratory assay, measured identically in both arms, and the paper states a single central laboratory analyzed the samples. The guide treats central-lab hs-CRP as Low unless the assay changed partway or measurement differed between arms; neither applies to SUSTAIN 3. Note that an assay bias problem is disclosed, but it is confined to SUSTAIN 2/4/5, which were excluded from this analysis for that reason — a disclosure that, if anything, supports the assay integrity of the included trials. A prespecified LLoQ imputation rule was applied and affected a trivial fraction of values.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
hsCRP data collected in SUSTAIN 2, 4, and 5 trials [ 22 – 24 ] were not used due to technical issues during sample collection — the hsCRP analysis assay was found to have a negative bias (an under-recovery of up to –25% at hsCRP concentrations below 5 mg/L), which affected a proportion of the samples.
D5 · Selection of the reported result
High risk of bias		Item 7 applies directly and on the strongest footing available: the authors affirmatively describe the whole analysis as exploratory — in the title, the background, the aim, and the limitations — and state that no multiplicity adjustment was performed. Under item 7 an author statement that the assessed analysis was exploratory or unadjusted for multiplicity is affirmative evidence of non-prespecification, and D5 is High; it is not the "protocol merely unretrievable" case that would sit at the Some concerns floor. Reinforcing this: hs-CRP was analyzed in multiple competing framings — ratio-to-baseline (emphasised), absolute change (relegated to the supplement, and largely null), clinical cutoffs, tertiles, sex, statin use, eGFR — with the transformation that yields significance foregrounded and the authors explicitly discussing why the relative-change results were significant while absolute changes were not. Under the guide's D5 domain rule multiple CRP transformations with one chosen for emphasis is itself at least Some concerns; here it compounds an already-High finding. One LLoQ imputation rule is described as prespecified, but that is a handling rule for a fraction of a percent of values and does not touch the prespecification of the outcome or its analysis; and under item 7 a paper's own claim of prespecification is not verification.
Impact of semaglutide on high-sensitivity C-reactive protein: exploratory patient-level analyses of SUSTAIN and PIONEER randomized clinical trials
The aim of this exploratory analysis was to evaluate the effect of semaglutide on hsCRP, and the contribution of glucose control and weight loss to this effect, in subjects at different stages of type 2 diabetes.

O4 — hs-CRP effect not explained by weight change; reported ONLY as proportion mediated by change in body weight, 10.9%. Residual direct effect not reported. · week 26 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Ruling 4: D1 stays trial-level. Concealment PRESUMED from the named, registered parent trial PIONEER 1 under item 6, not verified — no mechanism in any retrieved source, parent report not retrieved, arm-level baseline table unavailable. The post-randomization character of the mediation contrast is carried in D5/overall.
PIONEER 1: randomized clinical trial of the efficacy and safety of oral semaglutide monotherapy in comparison with placebo in patients with type 2 diabetes
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
D2 · Deviations from intended interventions
High risk of bias		Trial-product estimand rather than an assignment-effect analysis, plus conditioning on a post-randomization mediator (change in body weight) which is itself an effect of treatment; the resulting direct-effect contrast is not randomized and rests on an unstated and unexamined no-unmeasured-confounding assumption. PIONEER 1 additionally pools two dose arms for mediation that are reported separately for O1, with no justification given.
Mediation analyses included changes in HbA 1c and body weight (BW) as mediators, separately and combined, for pooled semaglutide arms (7 mg and 14 mg in PIONEER 1) versus comparators.
Mediation analyses examine associations among known, measured variables and outcomes, but do not necessarily identify causality.
D3 · Missing outcome data
Some concerns about risk of bias		No analysis population or missingness figure is reported for the mediation models; availability of hs-CRP, body weight and the covariate set jointly is NI. Mediator values were limited to the two hs-CRP visits (weeks 0 and 26).
Mediator data were only taken from the visits where hsCRP was assessed for estimation of the direct effect.
hsCRP was measured at weeks 0 and 56 in SUSTAIN 3; weeks 0 and 26 in PIONEER 1; weeks 0, 26, and 52 in PIONEER 2; and weeks 0, 8, and 26 in PIONEER 5.
D4 · Measurement of the outcome
Low risk of bias		Objective central-lab hs-CRP outcome and objective body-weight mediator, measured identically in both arms.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Item 7: exploratory self-description plus the explicit statement that the trials were not powered for mediation. Plus, as with all four mediation results, the direct effect is defined in Methods but never reported — only the proportion mediated is given. Ruling 4 carry: the derived non-randomized character of the mediation contrast is recorded here.
Finally, these trials were not powered for mediation analyses specifically, and hence there was high variability in the statistics used.
Limitations of this analysis include its exploratory nature, lack of hard endpoints for evaluation, low hsCRP testing frequency, lack of data on other markers of inflammation, and the relatively short follow-up time (26–56 weeks).

O4 — hs-CRP effect not explained by weight change; reported ONLY as proportion mediated by change in body weight, 5.7%. Residual direct effect not reported. · week 52 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Ruling 4: D1 stays trial-level. Concealment PRESUMED under item 6 from the named, registered parent trial PIONEER 2; not verified — no mechanism described in any retrieved source, parent report not retrieved, no arm-level baseline table.
Oral semaglutide versus empagliflozin in patients with type 2 diabetes uncontrolled on metformin: the PIONEER 2 trial
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
D2 · Deviations from intended interventions
High risk of bias		Trial-product estimand, not assignment; and conditioning on post-randomization change in body weight. A comparator-specific wrinkle sharpens the problem here: empagliflozin also causes weight loss, so the mediator differs far less between arms than in a placebo design, which mechanically shrinks the indirect effect and inflates the apparent weight-independent remainder. The 5.7% figure is therefore driven partly by the choice of comparator rather than by any property of semaglutide; the paper does not flag this.
once-daily oral semaglutide 14 mg or once-daily sodium–glucose cotransporter-2 (SGLT-2) inhibitor empagliflozin 25 mg (PIONEER 2)
with change in BW playing a lesser role in PIONEER 1 (10.9%) and PIONEER 2 (5.7%) but a similar or greater role in SUSTAIN 3 (35.8%) and PIONEER 5 (57.0%; Fig. 3 b)
D3 · Missing outcome data
Some concerns about risk of bias		No mediation-model denominators or missingness reported; joint availability of outcome, mediator and covariates is NI. Longest follow-up of the PIONEER trials here (52 weeks) with attrition unreported.
Mediator data were only taken from the visits where hsCRP was assessed for estimation of the direct effect.
D4 · Measurement of the outcome
Low risk of bias		Objective central-lab hs-CRP and objective body-weight mediator, identically measured in both arms.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Item 7: exploratory self-description and the explicit not-powered-for-mediation statement. Direct effect defined but never reported. Ruling 4 carry recorded here. Note also that the discussion generalises across trials ("HbA 1c seemed to contribute slightly more than BW") in a way that smooths over a 5.7%-to-57.0% spread in the weight fraction across the four trials, which is selective emphasis in the framing of the reported result.
Finally, these trials were not powered for mediation analyses specifically, and hence there was high variability in the statistics used.
the effect on hsCRP was partially mediated via the effect of semaglutide on HbA 1c and BW, although HbA 1c seemed to contribute slightly more than BW

O4 — hs-CRP effect not explained by weight change; reported ONLY as proportion mediated by change in body weight, 57.0% — the largest of the four trials. Residual direct effect not reported. · week 26 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Ruling 4: D1 stays trial-level. Concealment PRESUMED under item 6 from the named, registered parent trial PIONEER 5; not verified — the mechanism is described in no retrieved source and the parent report (reference 21) was not retrieved. The post-randomization conditioning is carried in D5 and the overall judgment.
Efficacy and safety of oral semaglutide in patients with type 2 diabetes and moderate renal impairment (PIONEER 5): a placebo-controlled, randomised, phase 3a trial
ClinicalTrials.gov identifiers: NCT01885208 (first registered June 2013), NCT02906930 (first registered September 2016), NCT02863328 (first registered August 2016), NCT02827708 (first registered July 2016).
D2 · Deviations from intended interventions
High risk of bias		Trial-product estimand rather than assignment, in the population where differential rescue and discontinuation are most likely (CKD, 14-year mean diabetes duration, insulin use prevalent enough to require its own covariate); plus conditioning on post-randomization change in body weight, which breaks randomization for the direct-effect contrast. No numbers are reported that would let the impact be bounded.
Data were evaluated for subjects receiving trial drug without rescue medication (trial product estimand), to avoid confounding effects from other antihyperglycemic drugs.
insulin treatment (PIONEER 5 analyses only), metformin treatment (SUSTAIN 3 and PIONEER 5 analyses only)
D3 · Missing outcome data
Some concerns about risk of bias		Smallest trial in the set (full analysis set 324), and no mediation-model denominator or missingness is reported; joint availability of hs-CRP, body weight and the expanded covariate set is NI. Mediator values limited to the hs-CRP visits.
Full analysis set, N 2158 324
Mediator data were only taken from the visits where hsCRP was assessed for estimation of the direct effect.
D4 · Measurement of the outcome
Low risk of bias		Objective central-lab hs-CRP and objective body-weight mediator, identically measured in both arms.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Item 7 applies as elsewhere (exploratory; not powered for mediation; direct effect defined but never reported). PIONEER 5 adds a distinct and more serious D5 problem: a proportion mediated is being reported for a TOTAL EFFECT THAT IS NULL. The total effect in this trial is ETR 0.83 (95% CI 0.67-1.03, p = 0.08) — a quantity not distinguishable from zero. The product method defines proportion mediated as indirect divided by total; when the denominator is statistically indistinguishable from zero the ratio is unstable and uninterpretable, and it can take any value including values outside 0-100% with small changes in the estimates. Reporting 57.0% as if it were a meaningful decomposition, without flagging that the effect being decomposed is null, is a selective-reporting problem in the presentation of the assessed result, not merely an analytical weakness. The authors' own general caveat about "high variability in the statistics" is the closest the paper comes to acknowledging it, and it does not name the null-denominator problem.
In PIONEER 5, there was a nonsignificant trend of reduction, as the ETR at week 26 for oral semaglutide versus placebo was 0.83 [95% CI, 0.67–1.03]; p = 0.08; Fig. 1 .
with change in BW playing a lesser role in PIONEER 1 (10.9%) and PIONEER 2 (5.7%) but a similar or greater role in SUSTAIN 3 (35.8%) and PIONEER 5 (57.0%; Fig. 3 b)

O4 — hs-CRP effect not explained by weight change. Reported ONLY as the proportion of the semaglutide effect on hs-CRP mediated by change in body weight: 35.8%. The residual direct effect (the O4 quantity) is not reported. · week 56 (end-of-treatment)

Assessed, not settled · RoB 2 · Two reviewers reached different judgments and nothing has settled it

This result has no overall judgment, because the two independent reviews did not reach one. Both readings are shown. The domains they did agree on are below.

Where they differThe two readings
D1 · Randomisation processOne reviewer: Some concerns about risk of bias
The other: Low risk of bias
D2 · Deviations from intended interventionsOne reviewer: High risk of bias
The other: Some concerns about risk of bias
DomainJudgment and reasoning
D1 · Randomisation process
Some concerns about risk of bias		Calibration ruling 4 is applied: D1 stays trial-level and is NOT downgraded for the derived, non-randomized character of a mediation analysis. So D1 here is the same judgment as O1_SUSTAIN3 — concealment PRESUMED, not verified, from the named and registered parent trial SUSTAIN 3, with no mechanism described in any retrieved source and no arm-level baseline table available to test the presumption. The post-randomization conditioning is carried in D5 and the overall judgment below, with reasons, as ruling 4 requires.
Study designs for these double-blind, randomized controlled trials have been published previously [ 18 – 21 ].
ClinicalTrials.gov identifiers: NCT01885208 (first registered June 2013), NCT02906930 (first registered September 2016), NCT02863328 (first registered August 2016), NCT02827708 (first registered July 2016).
D2 · Deviations from intended interventions
High risk of bias		Inherits the trial-product-estimand problem from O1 (not an assignment-effect analysis, no ITT alternative reported), and adds a problem specific to the mediation design: the analysis conditions on change in body weight, a variable measured after randomization and itself affected by treatment. Conditioning on a post-randomization mediator breaks the randomization for the direct-effect contrast, so the direct effect is an observational quantity that requires no-unmeasured-confounding of the mediator-outcome relationship. The paper adjusts for a covariate list but nowhere states or examines that assumption; it warns only in general terms that mediation does not identify causality. Additionally, the pooling of dose arms for mediation (while O1 is reported by dose) is an analysis choice with no stated justification.
Mediation analyses examine associations among known, measured variables and outcomes, but do not necessarily identify causality.
Mediation analyses included changes in HbA 1c and body weight (BW) as mediators, separately and combined, for pooled semaglutide arms (7 mg and 14 mg in PIONEER 1) versus comparators.
D3 · Missing outcome data
Some concerns about risk of bias		The mediation analysis requires complete data on hs-CRP AND body weight AND the full covariate set at the hs-CRP visits; no analysis population, denominator, or missingness figure is reported for the mediation models in any retrieved source. Availability is therefore NI and is necessarily no better than for O1, which is itself unreported at end-of-treatment. The paper additionally discloses that mediator values were restricted to hs-CRP visits, which for SUSTAIN 3 means two timepoints only (weeks 0 and 56) — information loss rather than missingness, but it compounds the same problem.
Mediator data were only taken from the visits where hsCRP was assessed for estimation of the direct effect.
Another limitation is that the mediation analysis did not use all mediator values obtained during the trial that may have a direct relationship with hsCRP levels, as may be done in more complex longitudinal mediation analyses.
D4 · Measurement of the outcome
Low risk of bias		Both the outcome (hs-CRP, central laboratory) and the mediator (body weight) are objectively measured with the same procedure in both arms. Measurement of the mediator is not described in detail in this paper, but body weight in a phase 3 programme is a routine objective measure and there is no indication of differential measurement. Measurement bias is not the issue with this result; identification is.
Another strength is that the same central laboratory analyzed hsCRP and other laboratory results.
D5 · Selection of the reported result
High risk of bias		Two independent routes to High, plus the ruling-4 carry. (a) Item 7: the analysis is self-described as exploratory with no multiplicity adjustment, and the authors state outright that the trials "were not powered for mediation analyses specifically, and hence there was high variability in the statistics used" — an explicit admission that the mediation analyses were not planned for in the trial design. Under item 7 that is affirmative evidence of non-prespecification and reaches High, not the Some concerns floor. (b) Selective reporting WITHIN the assessed result: of the three quantities the method section defines — direct effect, indirect effect, proportion mediated — only the proportion mediated is reported. The direct effect, which is the O4 quantity, is defined in Methods and in the Fig. 3 legend but never given a value, a confidence interval, or a p-value. The paper then makes a qualitative claim about it ("there may also be a direct semaglutide effect") with no reported estimate behind it. (c) Ruling 4 carry: the derived, post-randomization character of the mediation contrast is recorded here rather than in D1 — the direct effect is not a randomized comparison.
Finally, these trials were not powered for mediation analyses specifically, and hence there was high variability in the statistics used.
The direct effect is the effect of semaglutide versus comparator on change in hsCRP, independent of changes in HbA 1c and/or BW.
02Funding

Funding and conflicts

Funding
not reported in abstract
Industry funded
Unclear
Manufacturer
Novo Nordisk, Eli Lilly, Lilly, AstraZeneca, Sanofi, Boehringer Ingelheim, Pfizer, Hanmi, GlaxoSmithKline
Sponsor role
not reported in abstract
Author conflicts
OM reports personal fees for advisory board consultancy and speaker’s bureau from AstraZeneca, Boehringer Ingelheim, Eli Lilly, MSD, Novo Nordisk, and Sanofi; and research grant support from AstraZeneca and Novo Nordisk. MC reports being a partner at Clifton Medical Centre, a director at RIO Weight Management, Ltd, and a consultant for LighterLife and McDonald’s; he also reports research funding from Abbott, Boehringer Ingelheim/Lilly Alliance, Janssen, MSD, and Novo Nordisk; advisory board consultancy, consultancy and honoraria from Abbott, Boehringer Ingelheim/Lilly Alliance, and Novo Nordisk; and meeting support from Boehringer Ingelheim/Lilly Alliance and Novo Nordisk. ADR, SR, and PW are full-time employees of Novo Nordisk; SR also owns shares in Novo Nordisk. JR reports research funding from Applied Therapeutics Inc., Boehringer Ingelheim, Eli Lilly, Genentech, GlaxoSmithKline, Hanmi, Intarcia, Janssen, Lexicon, Merck, Metacrine, Novo Nordisk, Novartis, Oramed, Pfizer, and Sanofi; and advisory board consultancy, consultancy, and honoraria from Applied Therapeutics Inc., Boehringer Ingelheim, Eli Lilly, Hanmi, Intarcia, Janssen, Novo Nordisk, Oramed, Sanofi, and Zealand.
Independent replication
unknown
Notes
Authors declare relationships with the drug's manufacturer: Novo Nordisk, AstraZeneca (originally Amylin/Eli Lilly)

Funding is shown on every study and never used to score it.

03Claims

Claims this study bears on

04Source

The source, as retrieved

Abstract

[BACKGROUND] Exploratory analysis to determine the effect of semaglutide versus comparators on high-sensitivity C-reactive protein (hsCRP) in subjects with type 2 diabetes. [METHODS] Trials of once-weekly subcutaneous (SUSTAIN 3) and once-daily oral (PIONEER 1, 2, 5) semaglutide with hsCRP data were analyzed. Subjects with type 2 diabetes (N = 2482) received semaglutide (n = 1328) or comparators (placebo, n = 339; exenatide extended-release, n = 405; empagliflozin, n = 410). hsCRP ratio to baseline at end-of-treatment was analyzed overall, by clinical cutoff (< 1.0, ≥ 1.0 to ≤ 3.0, or > 3.0 mg/L), by tertile, and by estimated glomerular filtration rate in PIONEER 5 (a trial which was conducted in a population with type 2 diabetes and chronic kidney disease [CKD]). Mediation analyses assessed the effect of change in glycated hemoglobin (HbA1c) and/or change in body weight (BW) on hsCRP reductions. [RESULTS] Geometric mean baseline hsCRP was similar across trials (range 2.7-3.0 mg/L). Semaglutide reduced hsCRP levels by clinical cutoffs and tertiles from baseline to end-of-treatment in all trials versus comparators (estimated treatment ratios [ETRs] versus comparators: 0.70-0.76; p < 0.01) except versus placebo in PIONEER 5 (ETR [95% CI]: 0.83 [0.67-1.03]; p > 0.05). The effect of semaglutide on hsCRP was partially mediated (20.6-61.8%) by change in HbA1c and BW. [CONCLUSIONS] Semaglutide reduced hsCRP ratios-to-baseline versus comparators in subjects with type 2 diabetes (not significant with CKD). This effect was partially mediated via reductions in HbA1c and BW and potentially by a direct effect of semaglutide. Semaglutide appears to have an anti-inflammatory effect, which is being further investigated in ongoing trials. [TRIAL REGISTRATIONS] ClinicalTrials.gov identifiers: NCT01885208 (first registered June 2013), NCT02906930 (first registered September 2016), NCT02863328 (first registered August 2016), NCT02827708 (first registered July 2016).

Where this record came from

SourceRetrievedIdentifier
pubmedSep 13, 202636056351
first ingestion