Network meta-analysis

Percent body weight change after randomization in maintenance trials

Weight maintenance after a GLP-1 receptor agonist containing run in

weeks 20 to 104 (target week 52)Treatment policyDoses of a drug pooledMean difference

Randomized trials

Treatments

1,473

Participants

Trial arms

In plain language

Across 2 trials with 1,473 participants, 2 of 2 treatments had a clearly better result than placebo on percent body weight change after randomization in maintenance trials. The most favorable estimate was for tirzepatide injection: mean difference -19.09 percentage points (95% credible interval -25.47 to -10.37).

Tirzepatide injection: better than placebo but not among the most effective (certainty pending).
Semaglutide injection: better than placebo but not among the most effective (certainty pending).

Lower values favor the treatment. "Clearly" means the 95% credible interval excludes no difference. These are averages across trials; they do not say how any one person will respond. Not medical advice.

Network of trials

Each circle is a treatment; each line joins treatments that were randomized against each other in at least one trial.

Node area shows participants; line width shows trials comparing two treatments directly. Blue: GLP-1 receptor agonist containing treatments; grey: comparators. Hover over or tab to a node or line for details.

Effects compared with Placebo

Mean difference in percentage points. Lower values favor the treatment.

Versus Placebo. Blue: Bayesian random effects posterior median with 95% credible interval (thick) and 95% prediction interval (dashed). Amber diamond: frequentist estimate with 95% confidence interval. Hover over or tab to a row for its values.

Effects of each treatment versus Placebo
Treatment	Bayesian MD (95% CrI)	95% prediction interval	Frequentist MD (95% CI)	Verdict	Confidence
Tirzepatide injection TIRZ_SC	-19.09 (-25.47 to -10.37)	-28.71 to -7.08	-19.50 (-21.27 to -17.73)	Better	pending
Semaglutide injection SEMA_SC	-14.55 (-21.00 to -6.42)	-24.44 to -3.26	-14.80 (-15.95 to -13.65)	Better	pending

Confidence in the results

How much each comparison can be trusted, rated with CINeMA in six domains and summarized with GRADE.

Ratings follow the protocol's numeric rules from risk of bias and indirectness judgments made by language model reviewers (amendment A028); no human has verified them yet. Reporting bias (ROB-MEN): no concerns.

Confidence in each comparison with Placebo
Comparison	Within study bias	Reporting bias	Indirectness	Imprecision	Heterogeneity	Incoherence	Confidence
Semaglutide injection versus Placebo	pending	none	pending	none	none	some	pending
Tirzepatide injection versus Placebo	pending	none	pending	none	none	some	pending

Every other comparison (1)

Confidence in the other comparisons
Comparison	Within study bias	Reporting bias	Indirectness	Imprecision	Heterogeneity	Incoherence	Confidence
Semaglutide injection versus Tirzepatide injection	pending	none	pending	major	none	some	pending

Ranking

How likely each treatment is to be among the best. Rankings ignore the size of differences and their certainty, so read them with the effects above.

Rank probabilities

Probability of each rank for each treatment
Treatment	Rank 1	Rank 2	Rank 3	Mean rank
TIRZ_SC	87%	13%		1.1
SEMA_SC	13%	87%		1.9
PLACEBO			100%	3.0

League table

Every treatment against every other: mean difference of the row treatment versus the column treatment, posterior median and 95% credible interval. Green: row clearly better; red: row clearly worse.

Every treatment in a row compared with every treatment in a column
Row versus column	PLACEBO	TIRZ_SC	SEMA_SC
PLACEBO	PLACEBO	19.09 10.37 to 25.47	14.55 6.42 to 21.00
TIRZ_SC	-19.09 -25.47 to -10.37	TIRZ_SC	-4.54 -14.62 to 6.04
SEMA_SC	-14.55 -21.00 to -6.42	4.54 -6.04 to 14.62	SEMA_SC

Heterogeneity, fit and convergence

Bayesian random effects (primary)

Heterogeneity (tau): 2.393 (95% CrI 0.121 to 7.690)
Residual deviance: 3.9 for 4 data points
DIC: 7.8 (pD 3.9)
Largest R-hat: 1.001
Smallest ESS: bulk 2,587, tail 2,510
Divergent transitions: 0
Convergence: All criteria met
Priors: PR01 N(0, 35^2); PR03 N(0, 17.5^2); PR07 half normal(0, 3.5^2) on tau

Frequentist check (netmeta)

Method: inverse variance, REML, MD
Tau: 0.000
Q (total): 0.00 on 0 df
Q (between designs): 0.00

Inconsistency

Do trials comparing two treatments directly agree with the indirect evidence through other treatments?

No comparison here has both direct and indirect evidence, so inconsistency cannot be checked by node splitting; the between designs Q above is the global check.

Small study effects

A comparison adjusted funnel plot: each point is one trial comparison, centered on its own network estimate. Asymmetry can signal that small trials with unfavorable results are missing.

Fewer than ten trial comparisons, too few to judge funnel plot asymmetry.

Contribution of direct evidence

Percent contribution of each direct comparison (columns) to each network estimate (rows), from netcontrib.

Contribution of direct comparisons to network estimates
Network estimate	PLACEBO:SEMA_SC	PLACEBO:TIRZ_SC
PLACEBO:SEMA_SC	100
PLACEBO:TIRZ_SC		100

Baseline characteristics of the trials

The baseline value of each effect modifier or prognostic factor that applies here, per trial: mean or percent over arms, weighted by the number randomized. A note under a value names a stand in (a median for a mean, 100 minus the male share, Asian subgroup rows, an all Asian site list); "not reported" means no source of the trial prints it.

Baseline effect modifiers and prognostic factors per trial
Trial	Baseline BMI or body weight (kg/m2) effect modifier, prognostic factor; 2 of 2
SURMOUNT-4 S000242	30.5
STEP 4 S000521	34.37

Comparability of the trials

Checked across designs (the set of treatments each trial compares) for eligibility, outcome definition and assessment, and the baseline factors above. Flags point to possible intransitivity; notes describe variation that is not tied to design.

FlagBaseline, Baseline BMI or body weight: design means range from 30.5 (PLACEBO vs TIRZ_SC) to 34.4 (PLACEBO vs SEMA_SC) kg/m2; threshold 3
FlagOutcome assessment, definition: confounded with design: percent change in body weight from randomization (week 36) to week 88 1, percent change in body weight from randomization at week 20 to week 68 1
FlagOutcome assessment, intercurrent events (discontinuation): confounded with design: not stated 1, treatment_policy 1
NoteOutcome assessment, time point: selected time points span 48 to 52 weeks

Comparability items per trial
Trial	time weeks	analysis set	definition	instrument	adjudicated	proportion method	intercurrent events (discontinuation)	intercurrent events (rescue)	blinding	funding	start year	Eligibility
SURMOUNT-4	52	full_analysis	percent change in body weight from randomization (week 36) to week 88	not stated	not stated	not stated	treatment_policy	not stated	double_blind	industry	2021	age 18 or more; BMI 30 or more, or 27 or more with weight related complication; no diabetes; MTD 10 or 15 mg reached by week 36
STEP 4	48	full_analysis	percent change in body weight from randomization at week 20 to week 68	not stated	not stated	not stated	not stated	not stated	double_blind	industry	2018	age 18 or more; BMI 30 or more, or 27 or more with comorbidity; no diabetes; reached 2.4 mg at week 20

Trial data

What each trial contributes. Arm rows give the arm value with its standard error, or events over participants; contrast rows give the mean difference on the analysis scale against the trial's first row. Flags mark converted, combined or approximated inputs.

Trial data entering this analysis
Trial	Arm	Treatment	Week	Data	Value	SE	Events / N	Source
SURMOUNT-4	A1	TIRZ_SC	52	arm, percent_change	-5.500	0.663	335	R0018941:Table 2
SURMOUNT-4	A2	PLACEBO	52	arm, percent_change	14.000	0.612	335	R0018941:Table 2
STEP 4	A1	SEMA_SC	48	arm, percent_change	-8.300	0.350	535	registry:outcomes
STEP 4	A2	PLACEBO	48	arm, percent_change	6.500	0.470	268	registry:outcomes

Rerun this analysis yourself

Remove treatments or trials and refit the frequentist and Bayesian models in your browser, with the same methods as the published analysis.

Open in the NMA lab