Delphi consensus methodology.

Purpose and output classification

The protocol governs consensus exercises using two anonymous online Delphi rounds followed by in-person Congress ratification, with a project-specific protocol published before panel participation.

Outputs must be labelled accurately: a "consensus statement" records agreement reached by the protocol; a "position statement" is a timely organisational position developed under a separately described process; a "clinical guideline" is used only where the project includes sufficiently rigorous evidence review, recommendation development, certainty/strength assessment, external review and an update plan. Consensus is not evidence of efficacy by itself, and the certainty and strength of a recommendation must reflect the underlying evidence, not merely the percentage agreement.

Governance and roles

Each project runs on defined roles, each with a minimum responsibility:

• Steering Group (normally 5–9): scope, protocol, panel composition, statements, conflict management, deviations, interpretation and report.
• Methods Lead: independent method oversight; thresholds; survey design; anonymity; analysis; ACCORD reporting.
• Evidence Lead/team: search, appraisal, evidence summaries, certainty and citation integrity.
• Patient/public contributor: patient-important outcomes, acceptability and wording for patient-facing topics, where relevant and practicable.
• Secretariat/data custodian: invitations, reminders, coded data, version control, audit trail and access restriction.
• Panel: independent anonymous rating, comments, conflict disclosure and confidentiality.

Project protocol and registration

Before Round 1 opens, a protocol is published or time-stamped containing at least:

• the aim, intended audience, geographic scope and exact output type;
• the Steering Group, funder/sponsor role, conflicts and independence controls;
• the evidence-review method, databases/time period, appraisal approach and update date;
• panel eligibility, recruitment, target composition and planned sample;
• rounds, platform, anonymity, rating scale, thresholds, feedback and stopping rule;
• management of new/revised statements, missing ratings, attrition and deviations;
• the Congress ratification method, collaborator/authorship rules, data handling and publication plan.

Panel composition

Panel size and diversity defaults are set prospectively and may only be varied with prospective justification:

• default target: at least 30 participants accept and start Round 1, with at least 20 valid respondents completing Round 2;
• for an international clinical topic: at least five countries, no single professional discipline above 50%, and no single institution above 20%;
• individuals whose primary current role is industry/commercial should not exceed 20% of the panel; their votes remain equal and anonymous, and interests are reported in aggregate;
• for patient-facing recommendations, include at least two patient/public contributors where practicable, or explain why not;
• panel membership is based on relevant expertise or lived experience and declared eligibility; membership fees, title or Congress payment are not scientific eligibility criteria by themselves.

Rating scale and consensus thresholds

A 1–9 rating scale applies: 1–3 = disagree; 4–6 = uncertain/neutral; 7–9 = agree; "unable to rate" is excluded from that statement’s denominator. The denominator is the number providing a valid 1–9 rating; missing and "unable to rate" responses are not imputed, and statement-level response counts must be reported.

• Consensus in: at least 80% rate 7–9 and no more than 10% rate 1–3.
• Consensus out: at least 80% rate 1–3 and no more than 10% rate 7–9.
• Borderline: 75–79% in the relevant agreement band — revise and re-rate if comments identify a remediable issue.
• No consensus: below 75% in the relevant agreement band, or a conflicting distribution.
• A project may use a different threshold only if it is justified and published before ratings are seen — it may not lower the threshold after results are known.

Rounds 1 and 2

Round 1 is anonymous to other panel members; participants receive the protocol summary, evidence pack, definitions, declarations, confidentiality notice and rating instructions, and the response window is normally 10–14 days with neutral reminders. Each statement receives a 1–9 rating and an optional comment.

Between rounds the Steering Group reviews the aggregate distribution and anonymised themes without changing the pre-specified rules, and every wording change carries a version record and rationale. Round 2 shows each participant their prior rating where technically possible, the group distribution and a balanced anonymised summary of reasons; participants re-rate independently. A statement created after Round 2 is deferred or requires an additional anonymous round and cannot be declared consensus at Congress.

Congress ratification

The in-person Congress presents the evidence, panel composition, response/attrition, statement-level results and uncertainty, with facilitated discussion and conflicts reiterated.

The meeting may clarify wording without changing meaning but may not overturn an anonymous threshold by applause, show of hands, seniority or office — a material change returns to an anonymous ballot using the published threshold. Completing both online rounds and attending the Congress ratification qualifies a participant for collaborator status for that project; remote attendance is not offered.

Reporting, authorship and update

Reporting uses the ACCORD checklist and CREDES principles where applicable: invitations, acceptances, completion and attrition per round; panel characteristics and conflicts (protecting anonymity); statement wording at every round, response counts, percentages in the 1–3/4–6/7–9 bands, final classification and reasons for changes; and protocol deviations, missing data, limitations and sponsor role.

The project keeps a contribution log from inception; authors satisfy the publication policy/ICMJE criteria, CRediT roles are recorded, and collaborators are listed separately. Each output states an expiry or review point — default review is within three years, or earlier on major change — and an out-of-date statement remains archived and visibly marked.