Survival Curves and Hazard Ratios in Longevity Research
Key Takeaways
- A survival curve estimates the proportion expected to remain event-free beyond each time point; it does not display the event rate at that moment. [1] [3]
- A hazard ratio compares conditional event rates between groups and should not be read as a risk ratio or as a percentage change in lifespan. [4] [5]
- A single Cox hazard ratio is easiest to interpret when the ratio of group hazards is reasonably stable over time. [2] [6]
- In longevity studies, median lifespan, late-life survival, and age-specific mortality can change differently, so the whole curve and its uncertainty matter. [9] [10]
Who This Is Useful For
This page is for readers evaluating human cohort studies, clinical trials with time-to-event outcomes, or animal lifespan experiments. It is especially useful when a paper reports a Kaplan–Meier curve, a log-rank test, or a Cox-model hazard ratio and the practical meaning of the result is unclear. [3] [9]
Longevity research often asks not only whether an event occurs, but when it occurs. The event may be death, disease onset, loss of function, or another defined endpoint. Survival analysis uses each participant's observed follow-up time and can retain partial information from participants whose event is not observed before follow-up ends. [1] [3]
Survival Probability and Hazard Are Different Quantities
The survival function, usually written as S(t), is the probability of remaining free of the defined event beyond time t. The hazard describes the conditional rate at which events occur around time t among individuals who have remained event-free up to that time. A survival curve therefore summarizes accumulated experience, whereas a hazard describes the event process locally in time. [1] [4]
This distinction is important in ageing research because mortality hazards commonly vary with age. Two groups can have similar survival probabilities at one age while having different hazards around that age, or have hazards that differ early but converge later. [5] [10]
How to Read a Kaplan–Meier Curve
| Feature | Meaning | Interpretation |
|---|---|---|
| Horizontal axis | Time since a defined origin, or sometimes attained age | The origin and time scale must be comparable across groups. [3] |
| Vertical axis | Estimated probability of remaining event-free | A value of 0.80 means an estimated 80% have not yet experienced the endpoint. [1] |
| Downward step | One or more observed events | The size of the step depends on the number still at risk at that time. [1] |
| Tick mark | A censored observation | The participant contributed information up to that time, but no later event time was observed. [3] |
| Confidence band | Sampling uncertainty around the curve | Bands usually widen late in follow-up as the risk set becomes smaller. [3] |
| Number at risk | Participants still observed and event-free | Late curve segments based on few individuals should be interpreted cautiously. [3] [9] |
What Censoring Does and Does Not Mean
Right censoring occurs when the event time is known only to be later than the last observed time. This can happen because follow-up ends, contact is lost, or an animal is removed for a reason defined by the protocol. Kaplan–Meier estimation can use these partial observations, but its interpretation relies on censoring being independent of the unobserved event time, conditional on relevant information. [1] [3]
Censoring is therefore not equivalent to survival. A censored animal or participant is known to have remained event-free only through the censoring time. Longevity reports should state why observations were censored and identify those observations in the data or curve. [1] [9]
What a Hazard Ratio Measures
In a two-group comparison, a hazard ratio is the hazard in one group divided by the hazard in the reference group. A hazard ratio of 0.70 describes a 30% lower hazard at a given time under the model's assumptions; it does not mean that 30% fewer participants will experience the event, that survival time is 30% longer, or that an individual's event probability fell by 30%. [4] [5] [8]
The Cox model can estimate hazard ratios while accounting for measured covariates without specifying the shape of the baseline hazard. An adjusted estimate is conditional on the variables and functional forms included in the model, so it should be read together with the study design, covariate list, confidence interval, and endpoint definition. [2] [3] [5]
The Proportional Hazards Assumption
A standard Cox proportional-hazards model represents a covariate effect as a multiplicative hazard ratio that is constant over time. The baseline hazard itself may change; proportional hazards requires the ratio between the compared hazards, rather than each hazard, to remain stable. [2] [6]
In longevity experiments, an exposure can affect early mortality but not late mortality, or its association can differ across phases of life. When effects vary substantially with age, one overall hazard ratio compresses that time pattern into a single summary and may conceal when the groups differ. Curves, age-specific estimates, and analyses that allow time-varying effects can make that pattern more visible. [6] [10]
When Curves Cross or Separate Unevenly
Crossing survival curves are a visible warning that the relative effect may change over time, although visual inspection alone is not a complete model check. Even without crossing, curves that separate early and then remain parallel can reflect a different time pattern from curves that gradually diverge. Both can produce a similar single-number summary. [6] [8]
When proportional hazards is doubtful, useful summaries can include survival probabilities at prespecified ages, differences in restricted mean survival time, or explicitly time-varying hazard ratios. Restricted mean survival time is the area under a survival curve up to a chosen horizon and can express an absolute difference in event-free time without requiring proportional hazards. [7] [8]
Median Lifespan Is Not the Whole Curve
Median lifespan is the time at which estimated survival reaches 50%. It is easy to locate when both curves pass that level, but it discards information before and after the midpoint. A study can change early deaths without moving the median, or move the median while leaving the longest-lived portion of the cohort largely unchanged. [8] [9]
Claims about mean, median, and maximum lifespan therefore concern different features of the survival distribution. In small animal cohorts, the far-right tail is supported by very few observations and is particularly sensitive to sample size, censoring rules, and the chosen definition of maximum lifespan. [9]
Hazard Ratios, Risk Ratios, and Lifespan Differences
| Measure | Question Answered | Does Not Directly Show |
|---|---|---|
| Survival probability at time t | What proportion is estimated to remain event-free beyond t? | The complete time pattern or instantaneous event rate. [1] |
| Hazard ratio | How do conditional event rates compare between groups? | The absolute probability difference or change in lifespan. [4] [5] |
| Median survival difference | How far apart are the times at which survival reaches 50%? | Differences elsewhere on the curves. [8] |
| Restricted mean survival-time difference | How much average event-free time separates groups through a specified horizon? | Survival beyond that horizon. [7] |
| Log-rank p-value | How compatible is the observed curve difference with the null hypothesis under the test? | The magnitude or practical importance of the difference. [3] [9] |
A Worked Interpretation
Suppose a mouse study reports a hazard ratio of 0.75 with a 95% confidence interval of 0.58 to 0.97. Under the fitted model, the comparison is consistent with a lower mortality hazard in the study group, with uncertainty represented by the interval. It does not by itself establish a 25% increase in median lifespan, a 25% reduction in the probability of death by a particular age, or a change in maximum lifespan. Those questions require the survival curves, estimates at relevant ages, median or restricted mean survival, tail analyses, and the study's censoring information. [4] [8] [9]
Questions to Ask When Reading a Study
- What was the event and when did follow-up begin? Different endpoints and time origins define different survival quantities. [3]
- How many remained at risk late in follow-up? Sparse late risk sets make tail estimates uncertain. [3] [9]
- Why were observations censored? Informative removal can undermine the usual Kaplan–Meier interpretation. [1]
- Are confidence intervals shown? A point estimate alone does not communicate precision. [3]
- Was proportional hazards plausible? Look for curve shape, model diagnostics, or time-varying estimates rather than assuming one ratio fits every age. [6]
- Are absolute summaries reported? Survival probabilities and restricted mean time can complement a relative hazard ratio. [7] [8]
- Was the comparison randomized? Adjustment in an observational Cox model does not by itself remove unmeasured confounding or justify a causal claim. [5]
What This Does Not Mean
- It does not mean hazard ratios are uninformative; they are compact summaries when their definition, model, and time pattern are understood. [2] [4]
- It does not mean visibly separated curves prove a meaningful or causal effect; uncertainty and study design still govern inference. [3] [5]
- It does not mean median lifespan should be replaced by one universal statistic; different summaries answer different questions. [7] [8]
- It does not mean every departure from exact proportionality invalidates a Cox analysis; it changes what the reported summary represents and how cautiously it should be interpreted. [6]
Related Reading
Summary
Survival curves and hazard ratios describe related but distinct features of time-to-event data. A curve displays estimated event-free survival over time, while a hazard ratio compares conditional event rates. Sound interpretation in longevity research requires the endpoint and time scale, censoring pattern, numbers at risk, uncertainty, proportional-hazards assumption, and absolute survival summaries to be considered together. [1] [2] [5] [8]
References
- Kaplan, E. L., & Meier, P. (1958). Nonparametric Estimation from Incomplete Observations. Journal of the American Statistical Association. https://doi.org/10.1080/01621459.1958.10501452
- Cox, D. R. (1972). Regression Models and Life-Tables. Journal of the Royal Statistical Society: Series B. https://doi.org/10.1111/j.2517-6161.1972.tb00899.x
- Zwiener, I., et al. (2011). Survival Analysis: Part 15 of a Series on Evaluation of Scientific Publications. Deutsches Ärzteblatt International. https://pmc.ncbi.nlm.nih.gov/articles/PMC3071962/
- Spruance, S. L., et al. (2004). Hazard Ratio in Clinical Trials. Antimicrobial Agents and Chemotherapy. https://pmc.ncbi.nlm.nih.gov/articles/PMC478551/
- Hernán, M. A. (2010). The Hazards of Hazard Ratios. Epidemiology. https://pmc.ncbi.nlm.nih.gov/articles/PMC3653612/
- Stensrud, M. J., & Hernán, M. A. (2020). Why Test for Proportional Hazards? JAMA. https://doi.org/10.1001/jama.2020.1267
- Royston, P., & Parmar, M. K. B. (2013). Restricted Mean Survival Time: An Alternative to the Hazard Ratio for the Design and Analysis of Randomized Trials with a Time-to-Event Outcome. BMC Medical Research Methodology. https://doi.org/10.1186/1471-2288-13-152
- Uno, H., et al. (2014). Moving Beyond the Hazard Ratio in Quantifying the Between-Group Difference in Survival Analysis. Journal of Clinical Oncology. https://doi.org/10.1200/JCO.2014.55.2208
- Han, S. K., et al. (2016). OASIS 2: Online Application for Survival Analysis 2 with Features for the Analysis of Maximal Lifespan and Healthspan in Aging Research. Oncotarget. https://pmc.ncbi.nlm.nih.gov/articles/PMC5302902/
- Jiang, N., et al. (2025). Deciphering the Timing and Impact of Life-Extending Interventions: Temporal Efficacy Profiler Distinguishes Early, Midlife, and Senescence Phase Efficacies. Nature Communications. https://doi.org/10.1038/s41467-025-65158-4
This content is provided for educational purposes only and does not constitute medical advice.