Competing Risks in Ageing Research
Key Takeaways
- A competing risk is an event that prevents the outcome of interest from being observed, such as death occurring before dementia diagnosis or disability onset.
- Ageing studies are especially affected because mortality, multimorbidity, and functional decline often occur during the same follow-up period.
- Standard survival methods can overstate the probability of an outcome when competing events are treated as ordinary censoring.
- Cause-specific hazard models and Fine-Gray subdistribution models answer different questions, so the right interpretation depends on the study aim.
Who This Is Useful For
This page is useful for readers interpreting ageing studies where death, dementia, disability, cancer, cardiovascular disease, institutionalization, or loss to follow-up can occur in the same population. It is especially relevant for older-adult cohorts and trials in which many participants may die before the non-fatal endpoint under study can occur. [1] [4] [6]
Competing risks are a central interpretation problem in ageing research because older participants are often at risk of several mutually limiting outcomes at the same time. If a study asks who develops dementia over ten years, death before dementia is not simply missing information; it changes whether dementia can ever be observed for that participant during follow-up. [1] [4] [6]
What a Competing Risk Is
A competing risk is an event whose occurrence prevents the primary event of interest from occurring or being observed. In many ageing studies, death is the most common competing risk because it can preclude later diagnosis of a chronic disease, measurement of functional decline, or observation of admission to long-term care. [1] [2] [4]
This differs from ordinary censoring. Censoring assumes that a participant who leaves observation could still have the event later in a way that is compatible with the remaining risk set. A participant who dies before the event of interest is no longer at risk for many non-fatal endpoints in the same way. [2] [3]
Why It Matters More With Age
The competing-risk problem becomes more visible as baseline mortality rises. In younger cohorts, few participants may die before the endpoint of interest. In older cohorts, deaths during follow-up can be common enough to change both estimated incidence and the apparent effect of risk factors. [4] [6]
This is why analyses of dementia, frailty, cancer recurrence, cardiovascular events, and disability in later life often need to distinguish the risk of the event from the rate at which susceptible people experience it while still alive and observable. [3] [4] [6] [10]
Ageing-Relevant Examples
| Outcome of Interest | Competing Risk | Why It Changes Interpretation |
|---|---|---|
| Dementia incidence | Death before diagnosis | Deaths can reduce the observed probability of dementia even if underlying dementia susceptibility is high |
| Frailty onset | Death before a later frailty assessment | Later waves may overrepresent healthier survivors and miss decline that ended in death |
| Cancer recurrence in older adults | Death from other causes | Recurrence risk and non-cancer mortality both shape what patients are likely to experience |
| Institutionalization | Death while still living in the community | The probability of entering care depends partly on surviving long enough to enter care |
How Standard Survival Analysis Can Mislead
Kaplan-Meier estimates and standard Cox models are often used for time-to-event data, but they can be misleading when competing events are treated as non-informative censoring. In competing-risk settings, the Kaplan-Meier approach can estimate the probability of the event as if competing events did not permanently remove people from the event process, which can overstate cumulative incidence. [2] [5] [9]
The issue is not that standard survival models are always wrong; it is that they answer a different question. A cause-specific Cox model can describe how a covariate relates to the instantaneous rate of one event among people who are still event-free, while a cumulative incidence approach describes the probability of observing the event over time in the presence of competing events. [2] [3] [8]
Cause-Specific vs Subdistribution Models
Cause-specific hazard models and Fine-Gray subdistribution hazard models are both used in competing-risk analysis, but they should not be read as interchangeable. Cause-specific models are often useful for etiologic questions about event processes among those still at risk. Fine-Gray models directly model the cumulative incidence function and are often used for prognosis or absolute-risk prediction in the presence of competing risks. [2] [3] [7] [8]
The same exposure can therefore have different-looking associations depending on whether the analysis is asking about biological tendency toward the endpoint, observed probability of the endpoint before other events, or prediction of what happens first in a real population. [3] [8]
How to Read a Competing-Risk Result
| Question | Why It Matters | Warning Sign |
|---|---|---|
| What event was treated as competing? | The interpretation depends on whether death, another disease, or another endpoint prevents the outcome | The paper reports only censoring without explaining deaths or alternate events |
| Is the result a hazard ratio or cumulative incidence estimate? | Rates among event-free people and probabilities in the whole cohort answer different questions | A hazard ratio is described as if it were the absolute probability of disease |
| How common was the competing event? | Rare competing events may have little practical effect, while common deaths can reshape estimates | Older cohorts have high mortality but the analysis treats death as ordinary dropout |
| Are absolute risks shown? | Cumulative incidence curves or risk estimates make the practical effect easier to see | The paper reports only relative associations without event probabilities |
Relation to Composite Endpoints
Competing risks and composite endpoints are related but distinct. A competing-risk analysis separates mutually limiting events, while a composite endpoint combines several events into one outcome. In ageing research, a composite such as death, dementia, or persistent disability can avoid treating death as a nuisance event, but it creates a different interpretation problem because the components may differ in frequency and importance. [4] [11]
What This Does Not Mean
- It does not mean every ageing study must use a Fine-Gray model.
- It does not mean cause-specific hazard models are inferior; they answer a different question.
- It does not mean death should always be folded into a composite endpoint.
- It does not mean competing risks are only a statistical detail; they can change the substantive interpretation of ageing outcomes.
Practical Interpretation Examples
- If a biomarker predicts lower dementia incidence: check whether it also predicts mortality, because higher death rates can reduce observed dementia diagnoses.
- If a frailty study excludes people who died before follow-up: the remaining sample may describe survivors rather than everyone initially at risk.
- If a paper reports a cause-specific hazard ratio: read it as an event-rate comparison among those still event-free, not automatically as a population probability.
- If a study reports cumulative incidence: look for the competing event curve as well as the outcome-of-interest curve.
Related Reading
Summary
Competing risks matter in ageing research because older participants often experience several outcomes during the same follow-up period, and some events prevent others from being observed. Treating death or other competing events as ordinary censoring can distort estimates of disease, disability, or functional decline. Careful interpretation depends on identifying the competing event, separating hazards from cumulative incidence, and matching the statistical model to the study question. [1] [2] [3] [6]
References
- Wolbers, M., et al. (2014). Competing risks analyses: objectives and approaches. European Heart Journal. https://academic.oup.com/eurheartj/article/35/42/2936/784255
- Putter, H., et al. (2007). Tutorial in biostatistics: competing risks and multi-state models. Statistics in Medicine. https://pubmed.ncbi.nlm.nih.gov/17031868/
- Andersen, P. K., et al. (2012). Competing risks in epidemiology: possibilities and pitfalls. International Journal of Epidemiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC3396320/
- Berry, S. D., et al. (2010). Competing Risk of Death: An Important Consideration in Studies of Older Adults. Journal of the American Geriatrics Society. https://pmc.ncbi.nlm.nih.gov/articles/PMC2873048/
- Gooley, T. A., et al. (1999). Estimation of failure probabilities in the presence of competing risks: new representations of old estimators. Statistics in Medicine. https://pubmed.ncbi.nlm.nih.gov/10204198/
- Lee, S. J., et al. (2018). The Competing Risk of Death in Longitudinal Geriatric Outcomes. Journal of the American Geriatrics Society. https://pmc.ncbi.nlm.nih.gov/articles/PMC6367003/
- Fine, J. P., & Gray, R. J. (1999). A Proportional Hazards Model for the Subdistribution of a Competing Risk. Journal of the American Statistical Association. https://doi.org/10.1080/01621459.1999.10474144
- Austin, P. C., et al. (2016). Introduction to the Analysis of Survival Data in the Presence of Competing Risks. Circulation. https://pubmed.ncbi.nlm.nih.gov/26858290/
- Kim, H. T. (2007). Cumulative Incidence in Competing Risks Data and Competing Risks Regression Analysis. Clinical Cancer Research. https://aacrjournals.org/clincancerres/article/13/2/559/178178/Cumulative-Incidence-in-Competing-Risks-Data-and
- Koller, M. T., et al. (2012). Competing risks and the clinical community: irrelevance or ignorance? Statistics in Medicine. https://pmc.ncbi.nlm.nih.gov/articles/PMC3575691/
- McCoy, C. E. (2018). Understanding the Use of Composite Endpoints in Clinical Trials. Western Journal of Emergency Medicine. https://pmc.ncbi.nlm.nih.gov/articles/PMC6040910/
This content is provided for educational purposes only and does not constitute medical advice.