Fine-Gray regression
Fine-Gray regression models a particular event through a subdistribution hazard, which is linked to its cumulative incidence despite competing events.
Fit
When to use it, and when not to
Use it when
- You want covariate associations with cumulative incidence of one first event.
- Baseline covariates, event types and censoring times are available.
Not when
- Your target is the event rate among people still event-free. Use a cause-specific hazards model. Read: Choosing an analysis for competing risks
- Covariates change after baseline. Use a method that can define their values after a competing event. Read: Choosing an analysis for time-varying treatment
Before you start
What you need
- First-event typescompetingEventsDistinguish the outcome of interest, competing events and censoring.
- Baseline covariates and codingDefine units, reference groups and adjustment factors before fitting.
- Prediction horizonTo report probability, also estimate the baseline cumulative incidence at the time of interest.
- 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 constant subdistribution hazard ratio fits over the relevant follow-up.How to check: Inspect time patterns and model residuals. If it fails, allow time-varying effects or use a different risk model.
Plain language
How it works
Fine-Gray uses a special risk set that retains people after a competing event. They cannot have the target first event, but retaining them links the model to cumulative incidence. The subdistribution hazard is therefore not the event rate among living event-free patients. Geskus derives weighted risk-set representations and fitting methods.
With proportional subdistribution hazards, the target-event-free probability for an exposed profile is the reference target-event-free probability raised to its hazard ratio. Subtract from one to get cumulative incidence. This calculation needs the baseline curve as well as the coefficient. Zhou and colleagues develop residual-based checks of proportionality. Adjustment alone does not give a causal interpretation, because exposure may also affect the competing event.
Worked example
A worked example
- Setting
- A baseline exposure is linked to relapse, while death can prevent relapse.
- Question
- Does doubling the subdistribution hazard double relapse probability?
- Approach
- Assume a fitted log subdistribution hazard coefficient of 0.693147. The reference cumulative incidence at 12 months is 0.20.
- Result
- The subdistribution hazard ratio is 2.00. Predicted relapse probability at 12 months is 36 in 100 (36%), versus 20 in 100 (20%).
- What it means
- The predicted risk ratio is 1.80. Twice the subdistribution hazard is not twice the probability.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A subdistribution hazard ratio is neither a cause-specific hazard ratio nor a risk ratio. A Kaplan-Meier curve that censors a competing death or another event overstates cumulative incidence when read as one minus survival.
- Describing the subdistribution risk set as only people who could still relapse.Instead: People with a competing event remain in an artificial risk set, with censoring weights when needed.
- Reporting a subdistribution hazard ratio as an absolute prognosis.Instead: Provide predicted cumulative incidence at a stated horizon with uncertainty.
Inside Mallard
How Mallard uses this
Learn search lists Fine-Gray regression. The method registry links this method to its single-method article and code.
fine-gray-subdistribution).analysis.fine_gray, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Cause-specific cumulative incidence estimation and the fine and gray model under both left truncation and right censoring. (opens in a new tab) · Biometrics · doi 10.1111/j.1541-0420.2010.01420.x
- Goodness-of-fit test for proportional subdistribution hazards model. (opens in a new tab) · Stat Med · doi 10.1002/sim.5815