← Delphi pilots

Sample consensus record

GuidelineVahti · Delphi pilot deliverable

Synthetic example. Fictional statements, panel and votes, for illustration only — not a real guideline and not clinical advice. It shows the shape of what a pilot delivers.

Tip: print or save as PDF from your browser (⌘/Ctrl + P).

Round summary

Modified-Delphi / RAND–UCLA appropriateness, two rounds. Topic: surveillance of incidental pulmonary nodules (illustrative).

Panelists
12
Statements
5
Rounds
2
Consensus reached
4 of 5
Threshold
median 7–9, ≥75% in-region
Agreement (α)
0.71

Chance-corrected ordinal agreement is interval Krippendorff’s α across the statement set (0.71 — substantial). One statement did not reach consensus and is carried to a further round rather than forced to a recommendation.

Per-statement result

Median, IQR, the share of votes in the appropriate region, the RAND/UCLA classification, and the resulting GRADE recommendation.

S4. Acquire nodule CT at slice thickness ≤ 1.5 mm. Consensus · for

Median
9
IQR 0.5
% for (7–9)
100%
Certainty
high
Votes
12
1–3 · 0
4–6 · 0
7–9 · 12
GRADERecommend (strong, high certainty)

S1. Use volumetry (not diameter) as the primary growth metric for solid nodules 6–15 mm. Consensus · for

Median
8
IQR 1
% for (7–9)
92%
Certainty
moderate
Votes
12
1–3 · 0
4–6 · 1
7–9 · 11
GRADERecommend (strong, moderate certainty)

S5. Apply a validated baseline malignancy-risk model (e.g. Brock). Consensus · for

Median
8
IQR 1.5
% for (7–9)
83%
Certainty
low
Votes
12
1–3 · 0
4–6 · 2
7–9 · 10
GRADESuggest (conditional, low certainty) — the panel agrees, but low certainty holds it to conditional

S2. Report incidental micro-nodules < 4 mm without measuring them. Consensus · for

Median
7
IQR 2
% for (7–9)
75%
Certainty
low
Votes
12
1–3 · 0
4–6 · 3
7–9 · 9
GRADESuggest (conditional, low certainty)

S3. Discharge a stable solid nodule after 12 months of follow-up. No consensus · re-vote

Median
5
IQR 6
% for (7–9)
50%
Certainty
Votes
12
1–3 · 6
4–6 · 0
7–9 · 6
GRADENo recommendation — consensus not reached — the panel is split (extreme-disagreement rule); carried to a further round

Implementation priority

Importance × feasibility (median, cut at 7). The split statement (S3) is omitted until it resolves.

1 5 9 1 5 9 Feasibility → Importance ↑ S4 S1 S5 S2
Implement first — high importance · feasible
S4 — imp 9 · feas 8 — slice thickness ≤ 1.5 mm
High importance · lower feasibility — enable, then do
S1 — imp 8 · feas 4 — volumetry as the primary growth metric
Feasible · lower importance — optional quick win
S2 — imp 6 · feas 8 — report micro-nodules without measuring

Final statement table

#StatementMedian (IQR)% 7–9ClassificationGRADE
S4Slice thickness ≤ 1.5 mm9 (0.5)100%Consensus · forRecommend (strong, high)
S1Volumetry as primary growth metric8 (1)92%Consensus · forRecommend (strong, moderate)
S5Validated baseline risk model8 (1.5)83%Consensus · forSuggest (conditional, low)
S2Report micro-nodules without measuring7 (2)75%Consensus · forSuggest (conditional, low)
S3Discharge stable nodule at 12 months5 (6)50%No consensusNo recommendation — re-vote

Methods appendix (draft text)

Panel consensus was established using a modified-Delphi / RAND–UCLA appropriateness method. Twelve panelists independently rated each statement from 1 (strongly inappropriate) to 9 (strongly appropriate) across two rounds. Between rounds, each panelist received their own prior rating and the anonymized group distribution; no panelist was identified at any stage.

A statement was classified as consensus (for) when the group median fell in 7–9 with at least 75% of ratings in that region, consensus (against) for the symmetric 1–3 case, and no consensus otherwise, with an extreme-disagreement rule for bimodal distributions. Ordinal agreement across the statement set was summarised with interval Krippendorff’s α (α = 0.71). Recommendations were graded by combining the panel’s consensus direction with the certainty of evidence using GRADE / Evidence-to-Decision, yielding strong (“recommend”) or conditional (“suggest”) recommendations; statements without consensus received no recommendation and were carried to a further round.

Analyses were computed in GuidelineVahti, in the browser, on aggregate ratings only, with a versioned data file retained as the source record. The panel and its chairs are responsible for the recommendations; the tool records what the panel decided and does not certify that a recommendation is correct.

Run a pilot on your own statements

This is a synthetic example. To produce a record like this from your panel, start with a scoped pilot — from €1,500.

See the pilot tiers →