Cause-specific hazards regression
Cause-specific hazards regression links covariates to the rate of one event among people still free of every competing event.
Fit
When to use it, and when not to
Use it when
- You want an adjusted association with the rate of a specific first event.
- You can record each event type and when it happens.
Not when
- You want absolute event probability by a horizon. Estimate cumulative incidence, using all relevant cause-specific hazards if you need adjustment. Read: Choosing an analysis for competing risks
- Your question concerns repeated events within people. Use a recurrent-event model. Read: Choosing an analysis for recurrent events
Before you start
What you need
- One target eventcompetingEventsCode all competing first events and censoring separately.
- A justified adjustment setChoose baseline covariates for the question, rather than selecting them by P values.
- Time origin and follow-upAlign eligibility, exposure measurement and the start of outcome follow-up.
- Loss to follow-up is unrelated to future events among people with the same recorded factors.How to check: Record why follow-up ended. Assess how results change if those lost have a worse outlook.
- Different people supply independent observations.How to check: For patients grouped within clinics or families, use inference that accounts for those groups.
- Missing covariates are handled under a stated assumption.How to check: Under missing at random, impute using observed predictors of missingness and outcome. If missingness is not at random, use delta or tipping-point sensitivity analyses.
- A reported constant hazard ratio fits the cause-specific model.How to check: Inspect residuals and time interactions. Allow changing effects when needed.
- A causal interpretation requires adequate control of confounding.How to check: An adjusted association alone does not establish a causal effect.
Plain language
How it works
Fit a Cox model for the chosen event type. Remove people from its risk set when they have any first event. For this model, a competing event is coded as an end to risk without the target event. That coding estimates the cause-specific hazard; it does not justify reading a Kaplan-Meier complement as cumulative incidence. Prentice and colleagues define the observable cause-specific rates.
The fitted coefficient describes a contrast among people still event-free at that time. Lunn and McNeil show how Cox regression can model competing event types. To obtain adjusted cumulative incidence, combine models for every cause with event-free survival, or use another valid risk model. One cause-specific hazard ratio alone cannot supply a patient probability.
Worked example
A worked example
- Setting
- Patients are followed for relapse, with death as a competing first event.
- Question
- How does one fitted coefficient change the relapse hazard?
- Approach
- Assume a fitted log hazard coefficient of 0.693147 and a reference hazard of 0.02 per month at the chosen time.
- Result
- The cause-specific hazard ratio is 2.00. The fitted hazard at that time becomes 0.04 per month.
- What it means
- This is twice the instantaneous relapse rate among people still alive and relapse-free. It does not imply twice the relapse probability.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A cause-specific hazard ratio describes an event rate among people still event-free, not a ratio of cumulative probabilities. A Kaplan-Meier curve that censors a competing death or another event overstates cumulative incidence when read as one minus survival.
- Calling removal at a competing death an assumption that the latent causes are independent.Instead: Cause-specific hazards are observable rates. Independent latent failure times are not required to estimate them.
- Treating every adjusted coefficient as a causal effect.Instead: Only a contrast tied to the causal question and its assumptions can have that interpretation.
Inside Mallard
How Mallard uses this
Learn search lists Cause-specific hazards regression. The method registry links this method to its single-method article and code.
cause-specific-cox).analysis.cause_specific_hazards, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- The analysis of failure times in the presence of competing risks. (opens in a new tab) · Biometrics
- Applying Cox regression to competing risks. (opens in a new tab) · Biometrics