Ask Mallard
Core concept

Choosing an analysis for competing risks

4 min read

A competing risk, like death, prevents the outcome. Censoring it makes one minus Kaplan-Meier overstate cumulative incidence if it precedes later outcomes. Cause-specific hazard ratios still answer their own question.

Fit

When it matters, and when it does not

It matters when

  • Death or another event can happen first and make the outcome impossible, such as death before a readmission.
  • You want to know how many patients actually have the outcome over time, when some die first.
  • Patients are older or sicker, so competing deaths are common and ignoring them would mislead most.

It matters less when

  • Nothing can prevent the outcome, such as when the outcome is death from any cause. Use Kaplan-Meier curves and Cox regression. Read: Cox proportional hazards regression
  • Counting the competing event as part of the outcome, such as death or readmission, answers your question. Analyze that composite as one event. Read: Composite outcomes

Before you start

What to check in your study

Decide these first
  1. Which events can prevent the outcomecompetingEventsName them before the analysis, such as death from other causes or a transplant.
  2. Which question you are askingtargetHow many patients will have the outcome, or how a factor changes its rate among those still at risk. Each needs a different model.
  3. The outcome, and the date of every eventoutcomeTypeRecord the date and type of each patient's first event: the outcome, a competing event, or neither.
  4. Time zero, and the time points to reportSuch as the share readmitted by 30 days and by one year after discharge.
It rests on these
  1. Competing events are defined before the analysis.How to check: List them in the protocol, not after looking through the data.
  2. A competing event ends the chance of the outcome for good. It is not censoring.How to check: Code each patient's end as the outcome, a competing event, or censored, which means still at risk when follow-up stopped.
  3. Ordinary censoring, such as leaving the study, is uninformative.How to check: Record why patients left, and compare those censored early with those followed longer.
  4. The regression model matches the question.How to check: Choose cause-specific hazards for event-free rates. Choose Fine-Gray for associations with cumulative incidence; its ratio is not a risk ratio.

Plain language

How it works

Choose the question before the model. For the chance of an outcome by a set time, use cumulative incidence from Aalen-Johansen. Treat a competing event as an observed event, not as censoring.

What competing states look like
A state diagram for competing risks. Everyone starts at risk. From there a person can move to the event of interest or to a competing event that prevents it; neither state has a way back.Competing risks: two ways to leave the at-risk state, and no way backAt riskEvent of interestCompeting event thatprevents itCause-specific hazardCause-specific hazardCensoring the competing event and reading one minus a Kaplan-Meier curve treats those people as thoughthey could still have the event of interest, which overstates its cumulative incidence.
Conceptual diagram. It shows the structure of the idea, not data from any study.
Which question each method answers

For the event rate among patients still free of every event, choose cause-specific hazards regression. For associations with cumulative incidence, consider Fine-Gray regression. Neither hazard ratio is a risk ratio. Absolute risk depends on all causes, so report the competing event too. A yes-or-no outcome at a fixed time still needs this choice.

Worked example

A worked example

Invented for illustration
Setting
1,000 frail adults over 80 discharged from medical wards, followed for one year through linked records.
Question
How many are readmitted within a year, when some die first?
Approach
Nobody is lost, so nobody is censored. Each patient's first event is coded as readmission, death, or neither.
Result
Within a year, 300 of 1,000 were readmitted, 30 in 100. Another 200 died before any readmission, most of them in the first three months, while readmissions went on all year.
What it means
With no censoring, the cumulative incidence of readmission is simply 30 in 100. Many deaths came before readmissions that happened later. So one minus a Kaplan-Meier curve that censors the deaths would give a higher figure, as if the dead could still be readmitted.

In practice

Recent studies PubMed search · not vetted

Searching PubMed…

Watch out

Common mistakes

The one to remember

The complement of a Kaplan-Meier curve OVERSTATES cumulative incidence when a competing event exists, because it treats the competing death as though the patient could still have the outcome later. Cause-specific and subdistribution hazard models answer different questions and their coefficients are widely misreported as interchangeable.

  • Leaving out patients who died before the outcome could happen.Instead: Keep them, and code their death as a competing event. Dropping them inflates how common the outcome looks.
  • Reporting the outcome alone and saying nothing about the competing event.Instead: Report the cumulative incidence of both. A treatment can lower readmissions simply because more patients die first.
  • Using a cause-specific hazard ratio to say how many patients will have the outcome.Instead: Use the cumulative incidence function for that. The absolute risk also depends on how often the competing event happens.
  • Analyzing a yes-or-no outcome by a fixed time, such as readmission by 30 days, while ignoring patients who died first.Instead: Estimate the cumulative incidence by that time, treating death as a competing event. Or count death as part of the outcome, and say which you chose.

Inside Mallard

How Mallard uses this

Learn search lists Choosing an analysis for competing risks. Competing-event plan cues link to the concept; its related methods explain each estimator.

No Mallard code example points to this page.
Atlas node concept.choosing_competing_risks, the id a plan cue records.

Related Learn pages

Related designs and methods

Evidence

Sources & further reading

  1. Introduction to the Analysis of Survival Data in the Presence of Competing Risks (opens in a new tab) · Circulation · doi 10.1161/CIRCULATIONAHA.115.017719
  2. Practical recommendations for reporting Fine-Gray model analyses for competing risk data (opens in a new tab) · Stat Med · doi 10.1002/sim.7501
Source records checked against PubMed on 2026-11-07: record verified, confirm it supports this use. These educational summaries do not claim independent professional approval.
Last source review: 2026-11-07
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.