Draft synthesis: Cardiovascular outcomes with glucagon-like… — submissions
The 13 submissions received, published in full with declared interests and secretariat responses.
§2Submissions and responses
13 submissions were received. Each is published in full below with its declared interest, the secretariat response and the disposition. The Institute publishes submissions it did not accept in the same form as those it did.
The conclusion is stated more strongly than the certainty rating supports
The draft rates the anchor outcome at low certainty and then states in the conclusion that the intervention is effective. The respondent states that the two sentences cannot both be true as written.
The respondent proposes that conclusion language be tied mechanically to the certainty rating, so that a low rating cannot produce an unqualified claim.
The secretariat accepts this submission. The mismatch is the failure mode the certainty framework exists to prevent.
Conclusion wording is now drawn from a fixed set of formulations tied to the certainty rating, so that a low certainty rating produces a statement that the evidence may suggest an effect and that the estimate is likely to change with further research.
Studies using different outcome definitions are pooled into one estimate
The contributing studies define the outcome in at least two ways, and the draft pools them. The respondent states that the resulting estimate is an average across definitions rather than an estimate of any one quantity.
The respondent proposes that studies be pooled only within a definition, and that the definitions be reported separately with their own certainty ratings.
The secretariat accepts this submission. Pooling across definitions produces a figure with no referent, and the draft did it without saying so.
Studies are now pooled only within an outcome definition, each definition is reported with its own estimate and certainty rating, and the summary of findings states which definition each row concerns.
A single-trial result is presented in the visual form of a pooled estimate
An outcome contributed by one trial is presented in the same forest plot format as outcomes contributed by several. The respondent states that the format carries an implication of replication that a single trial does not have.
The respondent proposes that single-trial outcomes be presented differently and labelled as unreplicated.
The secretariat accepts this submission. The form of a graphic is part of what it asserts.
Outcomes contributed by a single trial are now reported without a pooled diamond, labelled as unreplicated, and downgraded for imprecision or inconsistency according to the framework rather than presented as a synthesis.
A sortable table implies a comparison the underlying data do not support
The draft presents a sortable table whose columns are drawn from sources of differing quality. The respondent states that sorting on such a column produces an ordering that looks like a ranking and is not one.
The respondent proposes that sorting be disabled on any column whose values are not commensurable.
The secretariat notes this submission and records that the point is correct in principle.
No amendment arises here because every sortable table in the document set already carries a standing statement above it that the ordering is not a ranking and that the values in each column are commensurable only where the column header says so. The proposal to disable sorting was considered and not adopted, because a reader who cannot sort a table generally sorts it elsewhere and without the statement.
A funnel plot is presented for a set too small to interpret it
The draft includes a funnel plot for an outcome contributed by fewer than ten studies. The respondent states that asymmetry cannot be assessed reliably at that number and that presenting the plot invites a conclusion the data cannot support.
The respondent proposes that the plot be removed and replaced with a statement that publication bias could not be assessed.
The secretariat accepts this submission. Presenting an uninterpretable graphic is not a neutral act.
The funnel plot is removed for every outcome contributed by fewer than ten studies, and replaced by a statement that small-study effects could not be assessed at that number, together with the count of registered trials identified without posted results.
The document should state what a reader ought to do
The draft assesses evidence and stops. The respondent, a practising clinician, states that a reader arriving at the document with a decision to make is left to convert an assessment into an action without help, and proposes that each document close with a recommendation.
The respondent argues that other evidence bodies issue recommendations and that declining to do so transfers the difficult part of the work to the reader.
The secretariat does not accept this submission, and records that the point is a reasonable one rather than a misunderstanding.
The Institute assesses evidence and does not issue recommendations, because a recommendation embeds values and a resource context that the Institute does not hold and cannot state. That constitutional limit is published on the methodology page and is not varied by consultation. The submission remains published in full.
Point estimates are given without an interval
Several estimates in the draft appear as single figures. The respondent states that a point estimate without an interval invites a precision the underlying data do not support, and that the effect is worst where the estimate is drawn from a small contributing set.
The respondent proposes that no point estimate appear anywhere in the document set without its interval, including in summary tables and in the abstract.
The secretariat accepts this submission in part. Intervals are added wherever the source reports one. The proposal is declined for figures the source published without an interval, because the Institute will not compute an interval a source did not report.
Every estimate now carries its interval where the source reported one, and where it did not, the estimate is annotated as reported without an interval rather than left to appear as a precise figure.
Intention-to-treat and efficacy-estimand results are combined without distinction
Several contributing trials report both a treatment-policy result and an efficacy result that censors at discontinuation. The draft draws from whichever is reported first in each paper.
The respondent, a trial statistician, states that the two answer different questions, that the difference is substantial where discontinuation is common, and that a synthesis should choose one and say which.
The secretariat accepts this submission. Mixing estimands within a single pooled estimate is a defect of the synthesis rather than of the trials.
The treatment-policy estimand is used throughout as the primary analysis, the efficacy estimand is reported as a secondary analysis where available, and the estimand used is stated in every row of the summary of findings.
A surrogate outcome is used as the anchor without validation evidence
The anchor outcome in the draft is a surrogate. The respondent states that the relationship between the surrogate and the outcome a decision turns on is itself an evidential question, and that the draft assumes it.
The respondent proposes that no surrogate serve as an anchor.
The secretariat accepts this submission in part. The anchor is retained where the surrogate is the only outcome the contributing trials measured, and the validation question is addressed rather than assumed.
Where the anchor is a surrogate, the synthesis now states the evidence for the surrogate relationship, rates it separately, and downgrades the anchor rating for indirectness accordingly rather than carrying the surrogate as though it were the outcome of interest.
The choice of effect measure is not justified and changes the appearance of the result
The draft reports relative effects for benefits and absolute effects for harms. The respondent states that the combination flatters the intervention and that the choice should be justified or made uniform.
The respondent proposes that both relative and absolute effects be reported for every outcome.
The secretariat accepts this submission in part. Both measures are reported for every outcome where the baseline risk needed for the absolute effect can be stated. Where it cannot, the relative effect is reported alone with the reason.
Every outcome now reports the relative effect and, where an assumed baseline risk can be stated and sourced, the corresponding absolute effect, with the baseline risk and its source given in the same row.
Subgroup findings are reported that were not registered in the protocol
The draft reports differences between subgroups that do not appear in the registered protocol. The respondent states that unregistered subgroup analysis is hypothesis-generating and that the draft presents it in the same form as the pre-specified results.
The respondent proposes that unregistered analyses be removed.
The secretariat accepts this submission in part. The analyses are retained and relabelled rather than removed, because removing an analysis that was conducted leaves no record that it was.
Every subgroup analysis is now labelled as pre-specified or post hoc against the registered protocol, post hoc analyses are reported in a separate subsection without a certainty rating, and the protocol version against which the labelling was made is stated.
Studies not published in English were excluded without assessment
The respondent states that a language restriction was applied at screening and that for some of the compounds in scope a substantial literature is published in other languages.
The respondent proposes that the restriction be removed and the review re-run.
The secretariat accepts this submission in part. The restriction is removed prospectively and the records already excluded on language grounds have been retrieved and screened on title and abstract in translation. The full re-run proposed is not undertaken, and the limitation is recorded rather than concealed.
Language restriction is removed from the protocol for the series, the records previously excluded on that ground are listed with their screening outcome, and the residual limitation is stated in the abstract of this review.
Efficacy outcomes are rated for certainty and harms are not
The draft assigns certainty ratings to the efficacy outcomes and reports harms narratively without ratings. The respondent states that the asymmetry implies harms are less amenable to assessment when they are simply less well measured.
The respondent proposes that harms carry certainty ratings on the same scale, with the reasons for downgrading stated.
The secretariat accepts this submission. Rating one side of the balance and not the other produces a document that cannot be used to weigh them.
Every reported harm now carries a certainty rating on the same scale as the efficacy outcomes, with the downgrade reasons stated, and discontinuation for adverse events appears in the summary of findings rather than in an annex.
References cited on this page
References are numbered in order of first citation in this document. Each superscript in the text links to its entry below.
- International Organization for Standardization. ISO/IEC 17025:2017 General Requirements for the Competence of Testing and Calibration Laboratories. ISO/IEC Standard 2017;3rd edition. identifier not held by the Institute
Identifiers are reproduced only where the Institute holds them. Where a digital object identifier or PubMed identifier is not shown, the Institute has recorded the journal and year and has not constructed an identifier.