How to Build a Composite Outcome with Delphi
Agreeing that an event belongs in an outcome leaves two practical questions: how much should it count, and when will the study look for it?
One composite outcome can require several decisions
The PrECiSIOn study used Delphi to develop a composite outcome for postoperative pulmonary complications. Forty-four panelists contributed expert judgments, and five patients or patient representatives contributed perspectives through interviews or surveys. Panelists assigned severity weights to components that reached consensus.
The panel supported a standard approach to monitoring during the hospital stay. The best follow-up period and whether to continue surveillance after discharge remained unresolved. The study illustrates why component selection, severity, and measurement deserve separate reporting.
Define what the composite will measure
A composite outcome combines information about several events into one endpoint. A core outcome set specifies a minimum set of outcomes to measure and report. A core set may include a composite, but the composite needs its own definitions and scoring rule.
Start with the population or unit being measured, the purpose of the endpoint, and what counts as an event. Write one operational definition per candidate component. Check for overlapping events and decide when they represent the same occurrence.
Specify which participants will judge each decision. Record how literature, stakeholder interviews, and panel suggestions shaped the list. Define the inclusion rule and valid-rating denominator before voting. Give important exclusions and unresolved items a place in the record.
Then specify the severity task. A rank orders events; numerical weights determine how much each contributes to a score. If the study will use weights, explain how they will be elicited and combined. Ask whether the resulting endpoint will flag any event, use the most severe event, or combine events under another prespecified rule.
A worked example from a document delivery pilot
In a fictional study, the unit is one document request, observed for seven days from intake. Twenty eligible panelists provide valid votes on each candidate event. The example uses at least 75% support to retain a component. A separate review assigns grades to the retained events.
| Candidate event | Votes to include | Decision | Grade if included |
|---|---|---|---|
| C01: Handoff completed late on the scheduled day | 18/20 (90%) | Include | Minor |
| C02: Handoff deferred to a later day because of a processing problem | 17/20 (85%) | Include | Major |
| C03: Typo corrected before handoff | 11/20 (55%) | Exclude | Not applicable |
The proposed composite uses the highest observed grade during the window. If a request has a minor delay on one handoff and a major disruption on a later handoff, it receives the major grade. A request with no qualifying event and complete observation receives the no-event category. An incomplete record needs its own missing-data rule.
Run the draft rule on test records before using it. Include a request with two events, a request with an event on the cutoff, and a request whose tracking stops early. Check how each is classified and keep the component results available.
Choose the measurement window before interpreting results
Define when observation starts, the duration, and whether events at the exact cutoff count. Name the data source and the person or system that will confirm each event. Ask what happens when a request ends early, an event recurs, or an event cannot be verified.
For each open measurement decision, show what the panel assessed and what evidence or testing is still needed. The team may need to settle the window in a later protocol. Attribute that choice to the group that made it and retain the earlier unresolved result.
Report the parts and the rule that combines them
- Publish the candidate list and definitions, who rated them, the valid counts, and inclusion or exclusion decisions.
- Show who assigned grades or weights and how the team turned those judgments into the final scoring rule.
- Specify the observation window, source of event data, and treatment of overlapping events and incomplete follow-up.
- Report component results alongside the composite when the endpoint is used. Distinguish panel judgments from later evidence on how the measure performs.
In Surveylet, separate inclusion and severity questions can share stable component IDs. Export the judgments with the item version, round, and relevant counts. Plan the endpoint calculation and its verification as part of the analysis scope.
Source and scope
PrECiSIOn investigators. Consensus on the definition, components, and grading of composite postoperative pulmonary complication outcomes. The Lancet Respiratory Medicine. Published online November 11, 2026. DOI: 10.1016/S2213-2600(26)00296-1.
The document delivery example is fictional. Its vote threshold, severity grades, and seven-day window are illustrative choices for that example.
Plan the endpoint and its reporting
Define what the panel will judge, how the endpoint will be calculated, and what later testing needs to establish.