Sys.setenv(LIVINGGLP1_ROOT = normalizePath(".."))
suppressMessages(pkgload::load_all("..", quiet = TRUE))
tab <- function(file, cols = NULL) {
  x <- utils::read.csv(file.path("..", "protocol", file), check.names = FALSE)
  if (!is.null(cols)) x <- x[, cols]
  knitr::kable(x, row.names = FALSE)
}

Status of this document

Draft protocol, written before any screening, data extraction or analysis, and revised after ten rounds of peer review (protocol/peer-review/). Items marked TO BE COMPLETED need information from the review owner before registration and are listed in the administrative checklist below. After registration the protocol is frozen and every change is recorded in protocol/amendments.csv with its date, rationale and whether it was made before or after any result for the affected outcome was seen.

The machine-readable parts of the protocol are authoritative where the text summarizes them: populations.csv, outcomes.csv, windows.csv, agents.csv, nodes.csv, node_classes.csv, effect_modifiers.csv, harms_nrs.csv, priors.csv, reference_sd.csv, turner_map.csv, amendments.csv and ../search/strategies/concepts.yaml. Appendices A to C reproduce them.

Administrative information

Registration. PROSPERO and OSF, before title and abstract screening begins. Registration numbers: TO BE COMPLETED.

Review type. Living systematic review with network meta-analysis of randomized trials; a separate living synthesis of nonrandomized studies for rare and long latency harms; and a public inventory of all trials, observational studies, registrations and published systematic reviews of GLP-1 receptor agonist containing drugs.

Reporting guidance. PRISMA-P 2015 for this protocol (moher2015prismap; shamseer2015prismap); PRISMA 2020 (page2021prisma), PRISMA-NMA (hutton2015prismanma), PRISMA-LSR (akl2024prismalsr) and PRISMA-S (rethlefsen2021prismas) for reports.

Contributors and roles (PRISMA-P items 3a and 3b). Guarantor and corresponding author: Ahmad Sofi-Mahmudi (affiliation and postal address TO BE COMPLETED). Second and third reviewers, information specialist, clinical advisors (endocrinology, cardiology, nephrology, hepatology, obesity medicine, pediatrics, psychiatry and addiction medicine, neurology) and patient partners: TO BE COMPLETED. Screening, extraction of outcome data, risk of bias assessment and certainty ratings are done in duplicate by named human reviewers; no stage begins until the team is named here.

Amendments (item 4). Logged in protocol/amendments.csv and summarized in each release note.

Support (items 5a to 5c). Funder, sponsor and their roles: TO BE COMPLETED. No funder or manufacturer has any role in design, conduct, analysis, interpretation or reporting. Competing interests of every contributor, including any relationship in the previous 36 months with a manufacturer of an incretin based drug or of a comparator: TO BE COMPLETED.

Administrative checklist before registration: team, affiliations and postal address; funding and competing interests; PRESS peer review of the search by an information specialist; confirmation of Embase access (section sources); verification of the Chinese terms by a Chinese-reading team member; confirmation and locking of the provisional thresholds by clinical advisors and patient partners; an OpenAlex API key; and a logged harvest of all four API sources with the registered concepts version, with every count reconciled, before the baseline search.

Introduction

Rationale

Glucagon-like peptide-1 (GLP-1) receptor agonists began as glucose lowering drugs for type 2 diabetes. Members of the class are now approved in several jurisdictions for chronic weight management, for cardiovascular risk reduction, for kidney outcomes in type 2 diabetes with chronic kidney disease, for obstructive sleep apnea with obesity and for steatohepatitis, and are studied in heart failure, osteoarthritis pain, type 1 diabetes and neurodegenerative, psychiatric and substance use disorders. The class now includes daily and weekly injectable peptides, an oral peptide, oral small molecules, and molecules that also act on GIP, glucagon or amylin pathways. Trials are published and registered faster than conventional reviews can be updated, and published network meta-analyses differ in scope, node definitions, time points and certainty methods, so their conclusions are hard to compare and quickly outdated.

A living network meta-analysis compares every regimen with every other within a population, including regimens never compared head to head, and is updated as evidence accrues (elliott2017lsr; nikolakopoulou2018livingnma). Publishing the full inventory of studies and documents with all code and data lets others verify, reuse and extend each step.

Objectives

Primary objectives, for each population in populations.csv that meets the connectivity rule (section connectivity):

1. To estimate the comparative benefits and harms of all GLP-1 receptor agonist containing regimens, defined by molecule, route, formulation, maintenance dose, frequency and dosing strategy, against each other and against placebo, standard care, lifestyle intervention and other active treatments.
2. To rate the confidence in every network estimate for critical and important outcomes, and to draw conclusions with the GRADE minimally contextualized framework.

Secondary objectives:

3. To maintain a public, continuously updated inventory of every randomized and nonrandomized study of these drugs, linking each study to all of its publications, registrations, protocols, statistical analysis plans, regulatory documents, results postings, supplements and shared code and data.
4. To synthesize nonrandomized evidence on rare and long latency harms.
5. To inventory, appraise and compare published systematic reviews and meta-analyses of these drugs.
6. To keep all of the above current, with dated, citable versions.

Methods

Eligibility criteria

Study designs

Infusions of native GLP-1, or of receptor antagonists, in physiological experiments are outside scope: they are not therapeutic drug regimens and do not inform the decisions this review addresses. On 2026-10-12 the endogenous hormone MeSH heading retrieved 6,988 PubMed records not otherwise found, of which 484 were human infusion experiments; their exclusion is a scope decision, not a search omission.

There is no minimum sample size or duration. Trials contribute to each outcome whose window (windows.csv) they meet. Single arm trials, pharmacokinetic studies in healthy volunteers and mechanistic studies are inventoried and excluded from both syntheses.

Participants

All indications are eligible; each population in populations.csv forms its own networks, which are never pooled in a primary analysis.

• A trial is assigned to a population when at least 80 percent of participants meet its definition at baseline, or when results are reported for a subgroup defined at baseline that meets it. A trial may enter more than one population and enters any single network once.
• Type 2 diabetes (P01) glycemic and weight networks are built within background therapy strata S0 to S3; the 80 percent rule also governs stratum assignment. Trials with no stratum reaching 80 percent form S4, which enters only the pooled type 2 diabetes networks for patient important outcomes and a pooled sensitivity analysis with stratum as a covariate. Weight management trials in type 2 diabetes are analyzed in their stratum and also report percent weight outcomes (O10, O11).
• When a background drug is stopped in one arm only (replacement or switch designs), each arm's node is the regimen it was randomized to; drugs continued in all arms are the background.
• Heart failure with LVEF above 40 percent (P07A) and with LVEF of 40 percent or lower (P07B) are separate populations regardless of connectivity and are never pooled. Chronic kidney disease with diabetes (P08A) and without (P08B) are separate populations; pooling them is a sensitivity analysis only.
• Randomized withdrawal and maintenance trials after a GLP-1 receptor agonist containing run in (P02W) are analyzed separately from initiation trials.
• Conditions not listed receive a population code by amendment when their first eligible trial is included (P99 rule).

Interventions

Any therapeutic drug whose mechanism includes GLP-1 receptor agonism, alone or in fixed combination, at any dose, route, formulation and duration (agents.csv, node_classes.csv). Withdrawn or discontinued agents (albiglutide, taspoglutide) are included because they inform the network; their exclusion is a sensitivity analysis.

Comparators

Any comparator is eligible: placebo; standard care or no added treatment; a randomized lifestyle intervention arm; another GLP-1 receptor agonist containing regimen (including another dose of the same drug); any other active drug; and bariatric or metabolic surgery. Clinical compatibility, not eligibility, decides whether a comparator enters a given network (surgery, for example, enters a network only when transitivity can be supported).

A trial enters a network only when at least two of its arms map to different nodes of that network. Trials whose randomized contrast does not change the GLP-1 receptor agonist containing regimen (for example, exercise versus no exercise on top of semaglutide in both arms) are inventoried only.

Trials comparing two non GLP-1 regimens are not eligible. Their evidence on comparators would enter only indirectly, including them would extend screening to the whole literature of every comparator class, and it is synthesized in dedicated reviews; the consequence (less precision for comparator versus comparator contrasts and fewer closed loops) is reported for each network.

Outcomes and time points

Outcomes, definitions, model scales, directions, importance, thresholds, absolute effect horizons, an a priori rare outcome flag and mapping rules are in outcomes.csv (77 outcomes at version 0.11.0); time windows in windows.csv; primary outcomes per population in populations.csv. Critical outcomes for a population are those rated critical whose population list includes it; primary outcomes are the subset named in populations.csv. Rules are in section outcomes.

Other criteria

No restriction by language, publication date, publication status or setting. Non-English reports are translated. Conference abstracts and registry results are eligible data sources; a study reported only as an abstract is included and flagged, and its exclusion is a sensitivity analysis. Retracted studies are excluded from all syntheses and listed with the retraction notice.

Information sources

If Embase cannot be accessed, its omission is justified by CENTRAL, which incorporates trial reports identified from Embase, by Europe PMC and OpenAlex coverage of the journal literature, and by the trial registries, and is reported as a limitation. Every source, date, access route and count is logged per run in search/runs/<date>/run_log.json; manual searches in search/manual/.

Search strategy

One versioned concept file (search/strategies/concepts.yaml, version 1.2.0) lists every class term, generic name, brand name and development code, a mechanism block, PubMed controlled vocabulary and the Chinese terms; R/search_strategy.R renders it into each source's syntax, and every run stores the exact query sent (Appendix A reproduces the current rendering). No design, language, date or outcome filters are applied.

• PubMed: title and abstract terms for every name and code (PubMed supports truncation inside quoted phrases: on 2026-10-12 "GLP-1 receptor agonist*"[tiab] returned 8,109 records, the same as the singular and plural phrases combined); the MeSH heading and pharmacological action "Glucagon-Like Peptide-1 Receptor Agonists"; drug headings and supplementary concepts (iGlarLixi has none and is covered by free text); and the mechanism block (terms such as dual agonist, triple agonist and co-agonist, combined with GLP-1, GIP, glucagon, incretin or amylin). On 2026-10-12 the mechanism block retrieved 190 PubMed records missed by the version 1.0.0 terms, including trials of UBT251, zenagamtide (amycretin) and DR10624, which were then added as agents; with version 1.2.0 it still adds 185 records. Development codes are kept even when they return no records today, because a living search must find their first report. The query is partitioned by Entrez date so that every partition stays below the 9,999 record cap of ESearch, and the union is reconciled against the total count (book records are counted and listed).
• Europe PMC: TITLE_ABS terms with explicit word forms (no truncation inside phrases), all sources except patents.
• OpenAlex: the search.title_and_abstract parameter with the same terms in chunks plus the mechanism block, unioned on work identifier; the fielded query sent is stored.
• ClinicalTrials.gov: intervention search for every name and code, and a general term search for class terms and the mechanism block, all statuses.

An information specialist will review the strategy with the PRESS 2015 checklist before the baseline search (mcgowan2016press) (TO BE COMPLETED). Sensitivity is checked against a benchmark set assembled independently of the harvest: pivotal trials listed in FDA and EMA labels and assessment reports for every approved agent and indication; publications listed in ClinicalTrials.gov results records; and trials included in published systematic reviews (section reviews). Relative recall is reported per source and overall; every benchmark trial not retrieved leads to a documented correction of the concept file.

Study records

Data management

All data are plain text tables (UTF-8 CSV, compressed with gzip when large) described by a Frictionless Data Package (datapackage.json) giving each column's type, constraints and keys; validation runs on every change. Copyrighted material (full texts, supplements, abstracts) is kept in a local, non-public store; the repository holds a manifest with source URL, retrieval date, SHA-256 checksum, size, media type and license for every file. Only content whose license permits it is redistributed.

The inventory has three linked levels: record (one bibliographic or registry item with every identifier: PMID, PMCID, DOI, OpenAlex ID, Europe PMC ID, registry ID); study (one investigation, with a stable internal ID, its primary registry number and registry aliases); and study_record (a many to many link with a role: primary report, secondary report, protocol, analysis plan, registry entry, results posting, regulatory review, erratum, retraction, commentary). Records are deduplicated on normalized DOI, PMID, PMCID and OpenAlex ID, then on an exact match of normalized title, year and first author surname; uncertain matches are reviewed by a person; every merge is logged. Records are linked to studies by registry numbers in PubMed data bank fields, abstracts, full texts and registry reference lists, then by acronym, sponsor, sample size and arm set; uncertain links are adjudicated by two reviewers. Overlapping observational data sources and periods are linked in a cohort overlap table so that no population is counted twice.

Selection process

1. Classification. Every record is classified automatically, without exclusion, into candidate randomized trial, candidate nonrandomized study, systematic review, protocol or registration, and other, to order the work.
2. Calibration. Before independent screening the reviewers screen a random calibration set of 300 records and discuss every disagreement; independent screening starts when Cohen's kappa is at least 0.8, with percent agreement also reported.
3. Title and abstract screening. Two reviewers screen every record independently; disagreements go to a third reviewer.
4. Full text screening. Two reviewers assess every potentially eligible record independently, recording reasons for exclusion from a fixed list.
5. Prioritization. A prioritization model implemented in R orders records so that likely eligible ones are seen first; every record is still screened by two people.

Decisions are stored with reviewer, date and reason in data/screening/.

Data collection

Outcome data used in any synthesis are extracted independently by two reviewers and compared; study and arm characteristics are extracted by one reviewer and verified by a second. The source location (page, table or figure) of every value is recorded. Discrepancies are resolved by discussion and logged. Missing information is requested from investigators or sponsors, with two reminders over six weeks.

When sources disagree, the value used is, in order: the result of the prespecified analysis for the estimand of interest (section outcomes) with the longest follow-up for that estimand; the regulatory review of that analysis; the registry results posting; the journal publication. All values are kept, the choice is recorded, and the alternative is a sensitivity analysis. Completers, per protocol and other non treatment policy results are flagged.

Full text and supplementary material

For every record of a study classified as a candidate trial, nonrandomized study or systematic review, the pipeline retrieves, in order: the Europe PMC full text and supplementary files; the PMC open access package; the Unpaywall and OpenAlex best open access locations; the publisher page for supplements; the registry's attached protocol, analysis plan and consent documents; and any data or code deposit cited in a data availability statement (Zenodo, Figshare, Dryad, OSF, GitHub, Mendeley Data, Vivli, YODA, ClinicalStudyDataRequest), frozen by version or commit. Items not openly available are obtained through institutional access. The manifest records each item's status: retrieved, restricted, unavailable or requested.

Data items

Study level: identifiers and linked records; design; registration date versus start date; countries and sites; sponsor and funding; dates; eligibility criteria; run in; blinding; randomization ratio; planned and actual duration; estimands and the strategy for each intercurrent event (treatment discontinuation, rescue medication, death); analysis population; missing data method; cluster or crossover features; conflicts of interest.

Arm level: node (molecule, route, formulation, maintenance dose, frequency, dosing strategy), titration schedule, background therapy, lifestyle co-intervention category (EM11), numbers randomized, analyzed and completing, and baseline characteristics (the effect modifiers in effect_modifiers.csv plus age, sex, race and ethnicity, weight and waist).

Outcome level: definition as reported and its mapping to outcomes.csv; time point; estimand; statistic (adjusted mean change with standard error, mean and standard deviation, events and denominator with follow-up or exposure time, hazard or rate ratio with confidence interval); survival estimator and competing risk hazard type; number with known status at $t$ by arm; observation period; competing deaths without a preceding target event and person years at risk by arm (time until the first target event, competing event, censoring or cutoff, as defined in section outcomes); any reported hazard ratio for the competing event; digitized reference arm cumulative incidence curves for both the target event and the competing first event (or event free survival), with the estimator of each curve (Kaplan-Meier with the other event censored, or Aalen-Johansen), their reported uncertainty and the published numbers at risk, where published (a target curve alone is recorded only as a cumulative incidence); source location; extractors.

Nonrandomized studies: the items in harms_nrs.csv (design, data source and period, exposure and comparator definitions, time zero, latency, follow-up and censoring, confounders adjusted for and method, handling of immortal time and competing risks) and adjusted and unadjusted estimates.

Outcomes and prioritization

Importance. Ratings in outcomes.csv follow GRADE and are confirmed by clinical advisors and patient partners before any outcome data are extracted.

Time windows. windows.csv gives each outcome a primary window (and for some outcomes secondary windows) per population, with a selection rule, a target week and limits (lower inclusive, upper exclusive; a blank upper limit means no upper limit). A window applies to an outcome and population only when the outcome's population_ids include that population. Windows for the same outcome and population never overlap, so every observation belongs to at most one window, and each outcome and population has exactly one primary window. Under nearest_target, the observation closest to the target week within the limits is used, and ties go to the later observation; under longest_follow_up (W00 for event outcomes and harms; W25 for eGFR slope), the longest randomized follow-up at or above the lower limit is used. Outcomes measured per procedure (O36A, O36B) or per pregnancy (O120) have no week window. Each window is analyzed as its own network, with trial duration within the window as a meta-regression covariate. The narrow window sensitivity analysis keeps only observations within 4 weeks of the target. For type 2 diabetes the primary window for HbA1c and weight in kilograms (O01, O02) is 36 to 64 weeks (target 52), because newer agents reach their maintenance doses only after 16 to 24 weeks of escalation, so 26 week values understate their maintenance effects; the 20 to 36 week window (target 26) is a prespecified secondary analysis. Percent weight change (O10) uses 40 to 104 weeks (target 68); its short trial version O10B, used in P12 and P15, uses 12 to 52 weeks (target 24).

Composite definitions. Within any outcome identifier, trials are pooled only when their component sets and thresholds are identical; otherwise the outcome is analyzed in definition strata, each its own network. Kidney composites illustrate the rule: O22 (50 percent eGFR decline), O22B (40 percent), O22C (kidney failure, sustained doubling of serum creatinine, or kidney death), O22D (any composite including cardiovascular death) and O22E (composites including albuminuria or smaller declines without cardiovascular death) are never pooled with each other; a composite that includes cardiovascular death is O22D even when it also includes albuminuria or a smaller eGFR decline. The pre 2017 hypoglycemia mapping in O04 is a prespecified exception to this rule; mapped trials are flagged and excluded in a sensitivity analysis.

Event outcomes and harms. Time to event outcomes use reported (or validly reconstructed) log hazard ratios over the longest randomized follow-up. For an outcome with a competing event, complete status counts (rule below) enter only its subdistribution network and are never pooled with cause specific estimates. For an outcome with no competing event (at present only all cause mortality, O20), reported hazard ratios, Kaplan-Meier contrasts and complete status counts have hazard type not applicable and enter the same network, which is primary; for an outcome with a competing event a Kaplan-Meier contrast remains a cause specific input. A trial with more than one eligible input to a network contributes one, in this order: the reported hazard ratio, the cumulative incidence or Kaplan-Meier contrast, the arm counts; the alternatives are logged.

Binary harms and adverse events (model_scale log_hazard_cloglog) are analyzed in two prespecified analyses, each rated separately.

Known status. Status at time $t$ is known when the target event occurred by $t$, a documented competing death occurred by $t$ without a preceding target event (counted as event free; complete status counts are therefore subdistribution inputs for an outcome with a competing event, and not applicable inputs for O20), or observation confirms absence of the target event through $t$. Loss or administrative censoring before $t$ without an established target or competing event leaves status unknown. Participants with competing deaths remain in the randomized denominator.

• Hazard ratio analysis. A reported hazard ratio enters as a normal likelihood on the log hazard ratio. A reported cumulative incidence at a common time $t$ in each arm enters as the contrast of $\log(-\log(1 - p))$ between arms, a normal likelihood with a delta method standard error from the confidence intervals, only when $0 < p < 1$ with finite uncertainty. Arm counts enter a binomial likelihood with a complementary log log link and an offset of $\log t$ (years) only when status is known for every randomized participant at $t$, the longest follow-up at which that holds. In a sensitivity analysis, $t$ is instead the longest follow-up at which status is known for at least 95 percent of randomized participants in every arm; unknown participants remain in the denominator, count as event free, and the event counts by that $t$ enter the same likelihood as a sensitivity analysis of the subdistribution network (or of the not applicable network for O20), never of a cause specific network. There is no general bound or direction of the resulting bias, even with equal missing fractions, because missingness may depend on the outcome. In an illustrative calculation with missingness independent of the outcome and occurring at baseline, equal 5 percent loss changes a hazard ratio of 2.0 (risks 0.50 and 0.75) to about 1.93, while 5 percent loss from the control arm only changes 1.20 (risks 0.010 and 0.012) to about 1.26.
• Recorded harms analysis. All trials' counts in the same complementary log log model with offset $\log t$, where $t$ is the nominal observation period in years. The reported quantity is the exponentiated contrast $Q = {-\log(1 - r_B)/t_B}/{-\log(1 - r_A)/t_A}$, where $r$ is the recorded event proportion; $Q$ is interpreted descriptively, without an incidence rate or hazard interpretation. No pooled absolute effect is derived from it; each trial's recorded proportions and within trial risk difference are displayed with its observation period and follow-up completeness. Magnitude dependent certainty domains use the range 0.8 to 1.25 on $Q$, and the rating states the scale. Incomplete follow-up is extracted by arm and judged in RoB 2 domain 3, and a sensitivity analysis excludes trials with more than 20 percent incomplete follow-up or a difference above 10 percentage points between arms.

Conclusions about a binary harm or a time to event outcome use the primary hazard type network when it meets the connectivity rule; otherwise the other hazard type network, labeled with its hazard type; otherwise, for binary harms, the recorded harms analysis, labeled as such, and for time to event outcomes, descriptive reporting. For O20 the not applicable network is the only network: it is used when it meets the connectivity rule, otherwise results are reported descriptively.

Denominators are the number randomized; the safety population is used only when the randomized number is not reported, and those trials are flagged.

Competing risks. For every input to a hazard scale analysis, the estimator and the hazard type are recorded: cause specific (a Cox model with competing deaths censored, or a contrast between arms of one minus Kaplan-Meier estimates with competing deaths censored, which enters as a normal likelihood on the contrast of $\log(-\log S)$ at a common $t$, the difference in log cumulative cause specific hazards, where $S$ is the Kaplan-Meier estimate, with a delta method standard error from the reported confidence intervals or the Greenwood variance, only when $0 < 1 - S < 1$), subdistribution (a Fine and Gray model, or a contrast derived from cumulative incidence that counts competing deaths as event free, including arm counts with complete status), or not applicable (only for outcomes with no competing event, at present only O20; for every other outcome the hazard type follows the estimator even when no competing event occurred in the trial). Hazard type is part of the network key; cause specific and subdistribution estimates never share a network; a trial reporting inputs of both types contributes each to its own network. For a time to event outcome with a competing event the cause specific network is primary, because outcome trials report Cox models; for O20 the not applicable network is the only network and is primary; for binary harms the subdistribution network is primary, because counts and cumulative incidence estimate it. Subdistribution networks assume proportional subdistribution hazards: reported proportionality assessments are extracted and available curves inspected, and where material nonproportionality is found, horizon specific cumulative incidence contrasts are reported separately and no constant subdistribution hazard ratio is used to convert effects across horizons. One minus a Kaplan-Meier estimate with competing deaths censored is never used as an absolute or reference risk for an outcome with competing events.

Absolute risks for outcomes with competing events are cumulative incidences that account for competing deaths; a single event rate is never sufficient. Let $h_{1a}(u)$ and $h_{2a}(u)$ be the cause specific hazards of the first target event and of the competing event in arm $a$. Survival free of both events is $S_a(u) = \exp[-\int_0^u {h_{1a}(v) + h_{2a}(v)},dv]$ and the cumulative incidence of the target event is $F_{1a}(t) = \int_0^t S_a(u),h_{1a}(u),du$. For the treated arm the target hazard is multiplied by its cause specific hazard ratio and the competing hazard by the cause specific hazard ratio for the competing event when one is reported; otherwise the competing hazard is unchanged, which assumes no treatment effect on it and is stated. The reference hazards come from the covering reference arms of section absolute by one of two routes.

• Piecewise integration requires both cause specific hazards over the analyzed horizon with valid joint uncertainty, from every covering reference arm: on each reported interval either the events and person time at risk for both the target and the competing first event, or digitized curves from the same arm and risk set that identify both hazards. Curves qualify only as cumulative incidences of mutually exclusive first events (Aalen-Johansen curves of both events, or of the target event with event free survival), or as Kaplan-Meier curves of each first event with the other censored; a target curve alone does not qualify, and a mortality curve that includes deaths after a nonfatal target event cannot substitute for the competing first event curve. The intervals form one grid fixed before pooling: the union over covering arms of the ends of each arm's reported intervals, which are the intervals of its reported events and person time or, for digitized curves, the intervals between the times of its published numbers at risk; an arm's hazards are constant within each of its own reported intervals and are carried onto every grid interval that the reported interval contains. On a reported interval from $t_s$ to $t_e$ of length $\Delta$, an arm's hazards are: from counts, the event count of each type over the person time at risk; from Aalen-Johansen curves, with $S = 1 - F_1 - F_2$ (or $F_2 = 1 - S - F_1$ when event free survival is published instead), $H = -\log{S(t_e)/S(t_s)}/\Delta$ and $h_k = H,\Delta F_k/(\Delta F_1 + \Delta F_2)$; from a Kaplan-Meier curve of event $k$ with the other event censored, $h_k = -\log{S_k(t_e)/S_k(t_s)}/\Delta$. These forms are exact under the stated constancy and require $0 < S(t_e) < S(t_s)$. The variance of each log hazard is $1/d_k$, where $d_k$ is the reported event count or, for curves, the count reconstructed from the increment and the number at risk at the start of the interval. An arm interval with a zero count for either event, or without a published number at risk, makes this route unavailable; no constants are added and no interval or arm is omitted. Log hazards are pooled per grid interval across arms with the random effects model of the reference risks (median of the predictive distribution).

• Exponential approximation uses constant rates $\lambda_1$ and $\lambda_2$ for both events with valid person time, each pooled across covering arms on the log scale with the random effects model of the reference risks (variance $1/d_k$; a zero count for either event makes the route unavailable), giving $F(t) = \lambda_1/(\lambda_1 + \lambda_2),(1 - e^{-(\lambda_1 + \lambda_2)t})$, only when constant hazards are supported, and is labeled as such.

• Cause specific hazard ratios: the reference risk shown with the estimate is $F_{1a}(t)$ of the reference node computed by the same route as the treated risk (piecewise integration, otherwise the exponential approximation), so that both come from one source; when neither route is available, only relative effects are reported.

• Subdistribution hazard ratios: the treated risk is $1 - (1 - F_{\mathrm{ref}})^{\mathrm{HR}}$, where $F_{\mathrm{ref}}$ is the reference risk at the analyzed horizon: the reported Aalen-Johansen cumulative incidence; otherwise the piecewise integration; otherwise the exponential approximation under supported constant hazards. Conversion requires supported proportional subdistribution hazards; constant cause specific hazards are required only when $F_{\mathrm{ref}}$ comes from the exponential approximation, which then carries its label. When no $F_{\mathrm{ref}}$ is available, the relative only rule of section absolute applies.

Rates are events per person year at risk: the total time until the first target event, competing event, censoring or analysis cutoff, whichever comes first, over the same interval as the events. The number randomized times the mean follow-up is used only when that mean is exactly this time at risk; general study follow-up, treatment exposure and median follow-up are not substitutes. When valid person time is unavailable, rates are not estimated by division, and another supported route or the relative only rule applies.

Both judgments (material nonproportionality, and support for constant hazards) are made by two reviewers before relative effects are modeled and logged with reasons, with the alternative as a sensitivity analysis: constant hazard support from the reference arm cumulative incidence curves where published, otherwise from cumulative incidences or event rates reported at two or more time points (when neither exists the exponential approximation is unavailable; any other eligible route is kept, including a reported Aalen-Johansen cumulative incidence for subdistribution conversion, and relative effects only are reported when no applicable route qualifies), and proportionality from reported tests and the published arm curves; nonproportionality is material when a reported test rejects proportionality at the 5 percent level or the arms' cumulative incidence curves cross after the first quarter of follow-up.

Procedure related outcomes (O36A, O36B) are risks per procedure on the odds ratio scale. Recurrent hypoglycemia (O04B) uses only reported log rate ratios with their standard errors.

Estimands. Each reported result is classified by its strategy for every intercurrent event (treatment policy, hypothetical, while on treatment, composite, principal stratum) following ICH E9(R1) (ich2019e9r1). Primary networks use treatment policy results only. Hypothetical (trial product) results form separate networks when enough trials report them; estimands are never substituted for one another in the same network.

Effect measures. Mean differences in natural units; standardized mean differences for O24B (physical functioning) and O24C (total HRQoL), each in its own network; hazard, odds and rate ratios as set by model_scale. Absolute effects are computed for every binary and time to event outcome where a valid conversion exists (section absolute).

Risk of bias in individual studies

• Randomized trials: RoB 2 for every result in a synthesis, with the cluster and crossover variants (sterne2019rob2). The effect of interest follows the causal question of the network: the effect of assignment for treatment policy networks and the effect of adhering for hypothetical networks. Two reviewers assess independently and reach consensus.
• Nonrandomized studies: ROBINS-I V2 for cohort designs, recording the exact release used (robinsi2025v2); ROBINS-I 2016 with documented adaptations for case control and self controlled designs (sterne2016robinsi). ROBINS-E (higgins2024robinse) is not used because these studies estimate effects of interventions. Results at critical risk of bias are excluded from quantitative synthesis and described.
• Missing evidence: ROB-MEN for network estimates (chiocchia2021robmen) and ROB-ME for pairwise syntheses (page2023robme), informed by registered trials without results and by registered outcomes not reported.

Judgments, signaling question answers and supporting quotations are stored as data.

Data synthesis

Networks and nodes

A network is defined by population, population stratum, outcome, definition stratum, window, estimand and competing risk hazard type; the definition stratum is the component set and threshold when an outcome identifier allows more than one. Nodes are defined by rule; nodes.csv lists the regimens known at registration.

• GLP-1 receptor agonist containing nodes are molecule by route by formulation by randomized maintenance dose by frequency. Titration schedules are recorded and do not define nodes. Arms are never classified by achieved dose, which is post randomization.
• Flexible dose and maximum tolerated dose arms form their own node, defined by molecule, route, formulation, frequency, permitted maintenance dose set and adjustment algorithm (for example TIRZ_SC_MTD_QW: 10 or 15 mg by tolerability); a different set or algorithm is a different node, and a sensitivity analysis assigns such arms to the highest permitted dose.
• Reformulated oral semaglutide tablets (1.5, 4 and 9 mg) are separate nodes; a sensitivity analysis pools 1.5 mg with 3 mg, 4 mg with 7 mg and 9 mg with 14 mg only.
• Free combinations randomized as a unit form a node named by both components; fixed ratio combinations with insulin are nodes of their own.
• Comparator drugs form molecule level nodes; arms of different doses of the same comparator within a trial are combined into one arm. Placebo, standard care, lifestyle and surgery are separate nodes; merging placebo and standard care is a sensitivity analysis.
• A regimen not yet listed is added to nodes.csv by applying these rules at extraction, which is not a protocol amendment; a new rule is. Rows marked is_template = yes are templates and never enter a network: dose templates (agents whose doses are not yet known) until a numeric randomized maintenance dose replaces the placeholder; the comparator template until an identified molecule or procedure and its class replace it; and the iGlarLixi switching strategy template until a strategy identifier, permitted pens, switching threshold and titration algorithm are recorded.
• iGlarLixi forms three node types: the 2:1 pen used throughout, the 3:1 pen used throughout, and pen switching strategies (one node per set of permitted pens, switching threshold and titration algorithm); they are combined only in a sensitivity analysis. registry_dose_candidates.csv lists the doses found in registration arm descriptions on 2026-10-12 as an aid, not as nodes.

Connectivity rule

For each population, population stratum, outcome, definition stratum, window, estimand and hazard type: a network meta-analysis is performed when at least two eligible trials form a connected network of at least three nodes; with two connected nodes, a pairwise meta-analysis with the same models; a single trial is reported descriptively. Binary outcome networks with fewer than 20 events in total are reported descriptively with arm level counts. Disconnected components are analyzed separately and never bridged by assumption.

Bayesian network meta-analysis (primary)

The primary analysis is a Bayesian random effects network meta-analysis in a generalized linear model framework (dias2011tsd2; dias2018book), fitted with multinma, which compiles models in Stan through rstan (phillippo2025multinma; phillippo2020mlnmr). Models outside its scope (dose response, bias adjusted and some sensitivity models) are written in Stan, fitted with cmdstanr, versioned in analysis/stan/ and tested by recovering known parameters from simulated networks.

For trial $j$ and arm $k$ receiving treatment $t_{jk}$,

$$ g(\theta_{jk}) = \mu_j + \delta_{jk}, \qquad \delta_{j1} = 0, \qquad \delta_{jk} \sim \mathrm{N}!\left(d_{t_{jk}} - d_{t_{j1}}, \tau^2\right), $$

with consistency $d_{ab} = d_b - d_a$, a common heterogeneity variance $\tau^2$ and covariance $\tau^2/2$ between relative effects within multi-arm trials. Likelihoods: normal with identity link for continuous outcomes (arm level means and standard errors, or the contrast level likelihood with the multi-arm covariance); binomial with logit link for fixed time binary outcomes; binomial with complementary log log link and log follow-up offset for binary harms and for complete status counts of time to event outcomes (with reported cumulative incidence and Kaplan-Meier contrasts entering as described in section outcomes); normal on the log hazard ratio for time to event outcomes; normal on reported log rate ratios for recurrent events (a count model for recurrent events would require a logged amendment). Arm level and contrast level data are combined in one model.

Priors. The complete specification is in priors.csv (Appendix C): weakly informative normal priors for trial baselines and relative effects, scaled by a reference standard deviation per outcome (reference_sd.csv) that only sets prior scales, with tighter log scale priors for rare outcomes; half normal priors on $\tau$ for continuous outcomes; and, for binary, time to event and count outcomes, the log normal predictive distributions of Turner and colleagues (their Table IV) for the mapped outcome type (turner_map.csv) (turner2015priors), applied to log hazard and log rate ratios as an approximation. The comparison type is pharmacological versus placebo or control unless more than half of the direct comparisons in the network are between two active drugs, in which case the pharmacological versus pharmacological distribution is used, where a direct comparison is a distinct pair of nodes with direct evidence (not a trial or an arm); the choice follows from the network geometry and is recorded before outcome data are extracted. The tighter relative effect prior (PR06) and the relative threshold apply to the outcomes flagged rare_outcome = yes in outcomes.csv, a flag fixed before extraction and never recomputed from observed risks. For the complementary log log model the offset is log follow-up in years. Prior predictive checks confirm plausible outcome values. Sensitivity analyses use the alternative priors in priors.csv.

Computation and diagnostics. Four chains of 2,000 warmup and 2,000 sampling iterations (more when needed), with seeds recorded. Convergence requires rank normalized $\hat{R} < 1.01$, bulk and tail effective sample sizes of at least 400 for every reported parameter (vehtari2021rhat), no divergent transitions after warmup (adapt_delta raised to 0.99, then reparameterization) and E-BFMI above 0.3. Monte Carlo standard errors are reported. Fit is assessed by total residual deviance against the number of data points, leverage plots, posterior predictive checks and DIC; random and fixed effect models are compared by DIC. Posterior summaries and the draws of reported quantities are kept with the checksums of their inputs.

Frequentist network meta-analysis (secondary)

The same networks are fitted with netmeta (balduzzi2023netmeta; rucker2012graph): random effects with a common heterogeneity variance estimated by REML, with 95 percent confidence and prediction intervals (riley2011prediction). Outcomes with reported hazard or rate ratios are analyzed on those scales. Binary outcomes are analyzed on the log odds ratio scale: the inverse variance random effects model without continuity correction when every arm has at least one event and one non-event, and otherwise the Mantel-Haenszel network method (efthimiou2019mh), explicitly labeled as a common effect model, with the change in contributing trials and connectivity reported. For harms whose primary Bayesian model is on the hazard scale, this frequentist odds ratio analysis is restricted to trials whose follow-up is 20 to 104 weeks and is a sensitivity analysis, not a replicate of the primary model. The frequentist analysis supplies the study contribution matrix (netcontrib, shortest path method, as used by CINeMA) (papakonstantinou2018contribution) and a check of robustness to the framework.

Missing statistics and conversions

Standard deviations are derived from standard errors, confidence intervals, t statistics or P values; medians with ranges or interquartile ranges are converted (wan2014sd) and flagged; when no dispersion can be derived, the median standard deviation of the same outcome and window in the same population is imputed and its exclusion is a sensitivity analysis. Values read from figures are digitized twice independently and flagged. Adjusted estimates are preferred; change from baseline is preferred to final values, and the two are not mixed for standardized mean differences. WOMAC scores are rescaled to 0 to 100; HbA1c in mmol/mol is converted to percent.

Cluster and crossover trials

Cluster trials enter with cluster adjusted estimates; otherwise contrast variances are multiplied by the design effect $1 + (m - 1)\rho$, with $m$ the mean cluster size and $\rho$ the trial's intracluster correlation, or 0.05 when unavailable (flagged, with 0.01 and 0.10 in sensitivity analyses). Crossover trials enter with a paired contrast only when its standard error is reported and carryover and period effects were addressed by washout or analysis; otherwise the first period is used if reported, and the trial is described if not.

Transitivity

Before any treatment effect is examined, the distributions of the effect modifiers in effect_modifiers.csv are tabulated and plotted across comparisons for each network. Imbalance is addressed by network meta-regression or restriction. When important intransitivity remains unresolved for a comparison, indirect and network estimates involving it are not calculated, and only direct evidence is reported.

Heterogeneity

$\tau$ with its credible interval, and 95 percent prediction intervals for every comparison, interpreted against the outcome thresholds; $I^2$ from the frequentist model as a secondary descriptive measure.

Incoherence

Global: consistency versus unrelated mean effects model by DIC and residual deviance (dias2011tsd2); the design by treatment interaction model in the frequentist analysis (higgins2012designbytreatment) with the net heat plot (krahn2013netheat). Local: node splitting for every comparison with direct and indirect evidence (dias2010nodesplit) and SIDE in netmeta. A nonsignificant test is not evidence of coherence in sparse networks; incoherence is judged by the size of discrepancies against the outcome threshold and investigated for data errors, effect modifiers and design differences.

Network meta-regression and subgroups

Covariates are the clinical effect modifiers in effect_modifiers.csv, in their stated priority order. Each is fitted one at a time: continuous covariates centered at the network mean and scaled to 1 SD; categorical covariates with prespecified indicator coding and the most common category as reference. The interaction is common across treatments within a class (exchangeable class interactions as an alternative), with the prior in priors.csv, and is fitted only when at least ten trials report the covariate, the covariate overlaps across the compared nodes, and each class coefficient is identifiable (at least three trials with covariate variation in that class). Trials missing the covariate are excluded from that model; when more than 20 percent of trials miss it, the model is not fitted and intransitivity is judged instead. A multivariable model with at most two covariates is fitted only when each coefficient has at least ten trials. Funding, blinding, year and estimand are examined in bias sensitivity analyses, not as clinical modifiers. Subgroup results reported within trials are extracted for aggregate subgroup meta-regression; multilevel network meta-regression (phillippo2020mlnmr) needs individual participant data from at least one trial and is planned only if such data are obtained.

Additional models

• Drug level network: doses of each molecule, route and formulation pooled into one node.
• Class effects model: exchangeable effects within the classes of node_classes.csv (multinma class effects).
• Dose response: Emax and linear models within one molecule, route and formulation (mawdsley2016mbnma), in Stan; fixed ratio combinations, flexible and maximum tolerated dose nodes are excluded.
• Component network meta-analysis for free combinations and fixed ratio insulin combinations, coding basal insulin and the GLP-1 receptor agonist component as present or absent, with additive and interaction models (rucker2020cnma).

Sensitivity analyses

For every primary network: excluding results at high risk of bias; excluding abstract only or registry only trials; excluding imputed standard deviations; narrow windows; fixed effect model; alternative priors; merging placebo and standard care; excluding withdrawn or discontinued agents; excluding trials in exclusively East Asian populations; flexible and maximum tolerated dose arms at their highest dose; reformulated oral semaglutide pooled with its bioequivalent doses; cluster design effects of 0.01 and 0.10; iGlarLixi nodes combined; counts from trials with at least 95 percent known event status added to the subdistribution network (binary harms and time to event outcomes; the not applicable network for O20); frequentist odds ratios for harms within 20 to 104 weeks of follow-up; recorded harms excluding trials with more than 20 percent incomplete follow-up or a difference above 10 percentage points between arms; the O04 hypoglycemia mapping excluded.

Ranking

SUCRA (Bayesian) and P scores (frequentist) are secondary summaries (salanti2011sucra; rucker2015pscore), always shown with absolute effects and certainty; rankings are never used alone to draw conclusions.

Absolute effects

The reference node of each network is declared before fitting: placebo if it is connected, otherwise standard care, otherwise the active node that appears in the most trials. Reference risks come from a random effects meta-analysis of the reference arms of the network's trials, using the median of its predictive distribution, with the 25th and 75th percentiles of trial reference risks as low and high risk scenarios (for outcomes with competing events, each covering arm's own cumulative incidence from its own hazards by the route in use); reference arm risks are logged before relative effects are modeled.

• Fixed time outcomes (odds ratios in a time window): the horizon is the target week of the analyzed window and the reference risk comes from that window; odds ratios are never extrapolated across horizons.
• Hazard based outcomes: absolute_horizon_weeks is the uncapped horizon; the analyzed horizon is the smaller of that value and the median follow-up of the network's reference arms (so heart failure networks, whose trials last about a year, use about 52 weeks). The reference risk at the analyzed horizon is estimated only from reference arms whose follow-up covers it. For outcomes without a competing event it uses reported Kaplan-Meier cumulative incidence at that time when available and otherwise their event rate under a constant hazard within their own follow-up. For an outcome with competing events a single event rate is never used: for a cause specific hazard ratio the reference risk shown is computed by the same route as the treated risk (piecewise integration of the reference hazards, otherwise the exponential approximation under supported constant hazards, otherwise the relative only rule); for a subdistribution hazard ratio $F_{\mathrm{ref}}$ is the reported Aalen-Johansen cumulative incidence, otherwise that piecewise integration, otherwise the exponential approximation, otherwise the relative only rule (section outcomes). Shorter arms are excluded from the baseline estimate, so nothing is extrapolated. When only one reference arm covers the analyzed horizon, its risk is used and no between arm heterogeneity is estimated. Competing risks follow section outcomes.
• Per procedure and per pregnancy outcomes (O36A, O36B, O120): absolute effects per 1,000 procedures or pregnancies.

Relative effects are converted to absolute differences per 1,000 (per 10,000 for rare outcomes) at the horizon, propagating uncertainty, and thresholds are applied to these absolute differences. For outcomes flagged rare_outcome = yes, the threshold is relative: the range of little or no difference is a hazard or odds ratio from 1/1.25 to 1.25, and absolute effects are still reported. Recurrent hypoglycemia (O04B), synthesized from reported rate ratios without a modeled baseline rate, uses the same relative range (rate ratio 1/1.25 to 1.25) and prior PR14. When a valid absolute conversion is unavailable (for example a cause specific hazard ratio without the competing hazard, or the recorded harms analysis), absolute effects and absolute importance categories are marked unavailable, magnitude dependent certainty domains are judged against the relative range 1/1.25 to 1.25 with the reason stated (never "no concerns" merely because conversion is unavailable), and rankings that need absolute effects are omitted; this overrides the general requirement for absolute effects.

Synthesis of nonrandomized studies for harms

Outcomes with nrs_synthesis = yes are specified in harms_nrs.csv (population, exposure, comparators, eligible designs, time zero and latency, estimand, minimum confounders, effect measures). Adjusted estimates are pooled by random effects meta-analysis separately by population, exposure and comparator contrast (active comparator versus non users), estimand (initial treatment versus on treatment), latency and risk window, effect measure and design; estimates that do not adjust for the listed minimum confounders are described and excluded from quantitative synthesis. Pooling uses REML (Sidik and Jonkman if REML fails to converge) and the Hartung-Knapp adjustment to the interval, and in a Bayesian model with a half normal prior with scale 1 on $\tau$ on the log scale. Where data sources overlap, one estimate per source, outcome and comparison is chosen: active comparator new user design first, then the longest follow-up, then the largest sample. Results at critical risk of bias are excluded. These results are presented beside the randomized evidence and never pooled with it in a primary analysis; a bias adjusted synthesis is exploratory, with bias priors published before it is run.

Meta-biases

Risk of bias due to missing evidence is judged with ROB-MEN for network estimates and ROB-ME for pairwise ones (chiocchia2021robmen; page2023robme), using registered trials without results, registered but unreported outcomes, discrepancies between registered and published primary outcomes, and comparison adjusted funnel plots when at least ten trials inform a comparison with a common reference (chaimani2013graphical).

Confidence in cumulative evidence

Every network estimate of every critical and important outcome is rated with CINeMA in six domains (nikolakopoulou2020cinema; papakonstantinou2020cinema), separately for the Bayesian and the frequentist estimates of the same evidence; where the two ratings differ, the lower is reported. Contribution matrices must represent the studies and network behind the estimate being rated. For hazard scale estimates the contribution network uses, for each trial, the input that entered the primary model: the reported log hazard ratio; or the cumulative incidence or Kaplan-Meier contrast with its delta method variance (no continuity correction); or a complementary log log estimate from the arm counts, adding 0.5 to every cell of a trial with a zero event or all event arm. The correction is used only for contributions, which depend on precision weights and network geometry, and never enters an effect estimate; every trial in the primary estimate is kept and shared arm covariance is preserved. When valid contributions still cannot be computed, within study bias and indirectness are judged qualitatively across the full evidence set, with the supporting studies and reasons recorded. Restricted sensitivity analyses (for example odds ratios within 20 to 104 weeks of follow-up) are rated separately and never substitute their rating for that of the primary estimate. Within study bias, reporting bias and indirectness are judged once per evidence set from these contributions; imprecision, heterogeneity and incoherence are judged on each framework's own estimates. Numeric rules:

• Within study bias and indirectness: risk of bias (low 1, some concerns 2, high 3) and indirectness (low 1, moderate 2, high 3) are averaged over studies weighted by their contribution to the estimate; no concerns below 1.5, some concerns from 1.5 to below 2.0, major concerns at 2.0 or above.
• Reporting bias: the ROB-MEN judgment (low risk: no concerns; some concerns; high risk: major concerns).
• Imprecision: against the range of little or no difference set by the threshold (absolute scale at the horizon; the ratio range 1/1.25 to 1.25 for rare outcomes, for rate outcomes synthesized from reported rate ratios and for estimates under the relative only rule of section absolute; the range 0.8 to 1.25 on $Q$ for the recorded harms analysis): no concerns when the 95 percent interval lies entirely within the range or entirely beyond one threshold; some concerns when it crosses one threshold; major concerns when it crosses both.
• Heterogeneity: the same rules applied to the 95 percent prediction interval, counting only crossings that the confidence or credible interval does not already make.
• Incoherence: the 95 percent interval of the direct minus indirect difference (node split or SIDE; for ratio scales, the difference of log effects compared with plus or minus log 1.25, or with the threshold converted to the relative scale at the reference risk) is compared with plus or minus the threshold: no concerns when it lies entirely within, major concerns when it lies entirely outside, some concerns otherwise. When incoherence cannot be estimated (no closed loops or no identifiable inconsistency parameters), the limitation is recorded and the judgment is made from transitivity and network structure, with the reason stated.

Each domain lowers confidence by one level for some concerns and two for major concerns, starting from high, without double counting across domains. Thresholds in outcomes.csv are provisional; clinical advisors and patient partners confirm them, with their sources, and they are locked before any outcome data are extracted. Every judgment and its reason are stored as data.

Conclusions follow the GRADE minimally contextualized framework (brignardello2020minimally; brignardello2018grade; puhan2014grade). Nonrandomized harms evidence is rated with GRADE using ROBINS-I, starting at high certainty and rating down for risk of bias (schunemann2019grade18).

Living review methods

Cadence. Searches run automatically every week; new records are screened monthly; models are refitted and a version released every quarter.

Triggers for an earlier release (within four weeks): results of a trial with at least 1,000 participants or of a dedicated cardiovascular, kidney or heart failure outcome trial; the first trial of a new node in a primary network; a regulatory safety communication about the class from FDA, EMA, MHRA, Health Canada, PMDA or TGA; a retraction or major correction of an included study.

Versions. Each release is a tagged, immutable snapshot (version as year and quarter, for example 2027.1) with the search date, data and code commit, software versions, seeds, a change log separating new evidence, corrections and methods changes, a PRISMA-LSR flow diagram with cumulative and update specific counts, and a Zenodo DOI. A release with no new eligible evidence is published as a dated notice.

Updating. Every release refits all models to the cumulative data with the original priors; posteriors are never recycled as priors. No sequential testing is done; intervals are estimates, not tests; no release is triggered by statistical significance (simmonds2017lsr). A conclusion has changed when the GRADE minimally contextualized category or the certainty level of an estimate changes; release notes highlight such changes.

Human gates. Automation retrieves, deduplicates, classifies and prioritizes; inclusion, extraction, risk of bias, certainty and conclusions are human decisions (thomas2017lsr).

Retirement. The review moves from living to periodic updating when, for four consecutive releases, no conclusion changes and no registered ongoing trial could change one, or when funding ends; the decision and its reasons are published.

Comparison with published systematic reviews

Systematic reviews and meta-analyses of GLP-1 receptor agonist containing drugs are identified from the same harvest and from PROSPERO and OSF. For each we retrieve the full text, supplements, shared data and code, and the prospective protocol (PROSPERO, OSF or a journal protocol); when none exists we record "no prospective protocol located" and keep the methods section only as reported methods. We extract scope, search dates, synthesis and certainty methods and main results; appraise quality with AMSTAR 2 (shea2017amstar2) and risk of bias with ROBIS (whiting2016robis); record discrepancies between protocol and publication; quantify the overlap of included trials with the corrected covered area (pieper2014cca); and compare estimates that overlap in scope with ours.

Reporting and dissemination

• Paper. A classic article per population or group of populations, following PRISMA 2020, PRISMA-NMA and PRISMA-LSR, generated from the same outputs.
• Website. A static website on GitHub Pages built in TypeScript, with the inventory as an evidence map, every network with interactive plots (network graph, forest plots, league tables, CINeMA profiles, funnel and contribution plots) made in R with the ggextreme package, study pages with all linked documents, the review inventory, this protocol with its peer review history, release notes and downloads. Every figure has a text and table alternative, and the site targets WCAG 2.2 level AA.
• Data and code. All R code and all data produced by the review are public, with a versioned JSON interface for the website described by JSON Schemas.

Reproducibility and open science

The repository is an R package (documented and tested functions) with an renv lockfile and a pinned CmdStan version; every stage runs from scripts in pipeline/. Searches run on continuous integration; model fitting runs on a workstation or self hosted runner. Inputs are checksummed, and every output records the inputs, code commit and seed that produced it. Licenses: code GPL 3 or later; data produced by the review CC BY 4.0; third party content keeps its license and is redistributed only where permitted.

Use of automation and large language models

Searches, deduplication, classification, prioritization and retrieval are deterministic code in this repository. Large language models assist with code, with drafting, and with peer review of this protocol (the full record is published in protocol/peer-review/); they may also act as an advisory, non binding additional reader during screening and extraction. They never replace a human decision, and when they are used in screening or extraction their agreement with human reviewers is measured and reported. Each release names the tools and versions used.

Patient and public involvement

People living with type 2 diabetes, obesity and the other conditions studied will be invited to confirm outcome importance and thresholds, to review plain language summaries and to comment on the website (TO BE COMPLETED).

Ethics

The review uses published and publicly registered aggregate data and needs no ethics approval. Individual participant data, if obtained, will be used under the provider's agreement and never published.

Timeline

TO BE COMPLETED after the team is named: registration; PRESS review; baseline search with the registered concepts version; calibration and screening; extraction; baseline analysis; first release. A pipeline test harvest ran on 2026-10-12 with concepts version 1.0.0 against PubMed, Europe PMC and ClinicalTrials.gov (OpenAlex awaits an API key); a logged harvest of all four API sources with the registered concepts version, with every count reconciled, precedes the baseline search.

References {.unnumbered}

Appendix A. Search strategies {.unnumbered}

Rendered from search/strategies/concepts.yaml (version r read_concepts()$version) by R/search_strategy.R when this document is built.

PubMed

#| results: asis
cat("```", strwrap(pubmed_query(read_concepts()), 100), "```", sep = "\n")

Europe PMC

#| results: asis
cat("```", strwrap(europepmc_query(read_concepts()), 100), "```", sep = "\n")

OpenAlex (search.title_and_abstract, one query per chunk, unioned)

#| results: asis
q <- openalex_queries(read_concepts())
cat("```", unlist(lapply(seq_along(q), function(i) c(paste0("Chunk ", i, ":"), strwrap(q[i], 100), ""))), "```", sep = "\n")

ClinicalTrials.gov (query.intr, then query.term, unioned)

#| results: asis
ct <- ctgov_queries(read_concepts())
cat("```", "query.intr:", strwrap(ct$intr, 100), "", "query.term:", strwrap(ct$term, 100), "```", sep = "\n")

Appendix B. Populations, outcomes and windows {.unnumbered}

tab("populations.csv", c("population_id", "label", "strata", "primary_outcome_ids", "network_status"))

tab("outcomes.csv", c("outcome_id", "label", "model_scale", "direction", "importance", "nrs_synthesis", "threshold_scale", "threshold", "threshold_unit", "absolute_horizon_weeks"))

tab("windows.csv")

Appendix C. Agents, nodes, priors, effect modifiers and harms specifications {.unnumbered}

tab("agents.csv", c("agent_id", "name", "class_ids", "class_verification"))

tab("node_classes.csv", c("class_id", "label", "role"))

tab("nodes.csv", c("node_id", "class_id", "formulation", "dosing_strategy", "label"))

tab("priors.csv", c("prior_id", "parameter", "model_scale", "applies_to", "prior", "sensitivity_prior"))

tab("reference_sd.csv")

tab("turner_map.csv")

tab("effect_modifiers.csv", c("modifier_id", "label", "populations", "kind", "priority"))

tab("harms_nrs.csv")