Survival Hazard Ratios: Ask for the Absolute Comparison
Separate a survival hazard ratio from event risk, percentage-point benefit and time gained. Preserve the endpoint, direction, horizon and uncertainty.
Which quantity does the survival headline compare?
A readout reports a hazard ratio of 0.75. That number alone does not establish that 25% more participants are alive at a particular visit or that patients live 25% longer. Before translating the headline into a clinical or valuation assumption, ask which event, treatment direction, analysis and time horizon the estimate represents.
The National Cancer Institute's hazard-ratio definition describes a comparison of events between groups over time. A hazard comparison concerns event rates among those still at risk; a proportion experiencing an event by a specified time is a different summary. For death as the event, also verify whether the reported ratio is treatment over control or the reverse.
Choose a company and prepare a question. Run it manually after confirming the question and reviewing the required credits.
Prepared research question
How reliable is the trial evidence for a company I choose?
Traceable citations · Unknowns marked
Separate the endpoint from its effect measure
The FDA's December 2018 oncology-endpoints guidance, section III.A, defines overall survival using death from any cause. A result for progression or a composite outcome is not automatically a result for death. The outcome definition must survive the summary.
Keep these quantities in different fields:
| Reported quantity | What the source must identify | What the number alone cannot supply |
|---|---|---|
| Hazard ratio | Event, direction, model, analysis population and interval | Absolute event probability at a chosen time |
| Event risk at time t | Specified time and the appropriate cumulative probability estimate | A hazard ratio without event-time information |
| Survival probability at time t | Event-free probability, horizon and uncertainty | Mean lifetime or a benefit outside that horizon |
| Median survival | The reported time-to-event distribution and estimation | A fixed percentage increase in each patient's lifespan |
An OS hazard ratio and a PFS hazard ratio answer different questions. Even for the same endpoint, an adjusted model estimate, an unadjusted curve and a subgroup estimate may not target the same comparison.
Work through a risk calculation without renaming it a hazard ratio
Hypothetical examples, not trial results: assume all 100 participants in each arm are followed through a stated 12-month time point, with death recorded completely and no censoring. These simplifying assumptions allow a direct count-based risk calculation.
| Illustration | Control deaths by 12 months | Treatment deaths by 12 months | Treatment/control risk ratio | Absolute reduction in event risk |
|---|---|---|---|---|
| A | 20/100 = 20% | 15/100 = 15% | 0.75 | 5 percentage points |
| B | 40/100 = 40% | 30/100 = 30% | 0.75 | 10 percentage points |
Both examples have a 25% relative reduction in the specified event risk, because 1 - 0.75 = 0.25. They have different absolute differences. In A, the assumed survival proportions are 80% and 85%: a five-point difference, not 25 percentage points.
These counts do not establish either example's hazard ratio. The event times and risk sets are missing. Do not take the numerical 0.75 risk ratio from this illustration and label it the 0.75 hazard ratio from an unrelated press release. The examples also supply no confidence intervals, significance result or assessment of clinical importance.
Check whether one hazard ratio is an adequate summary
In their 2014 methods paper, Uno and colleagues discuss the proportional-hazards assumption behind a conventional single hazard-ratio summary and the interpretation problems when relative hazards change over time. Their abstract also argues for a prespecified, clinically meaningful comparison and robust inference. This is methodological evidence, not a finding about a current drug.
Ask for the protocol's analysis, curves, interval and any assessment of the model assumptions. Do not assume proportional hazards merely because one ratio appears in a headline, or choose a more favorable time point after seeing the curves. A visually different curve is not a complete analysis of uncertainty or safety.
Where follow-up is incomplete, the simple deaths-divided-by-randomized-count calculation above may not reproduce a published survival estimate. Preserve the actual method and risk table. If a requested fixed-time comparison is not reported and cannot be appropriately estimated from the available evidence, its value remains UNKNOWN.
Preserve a comparison that can be revisited
For the next readout, retain the trial identifier, population, endpoint, treatment direction, analysis method, data cutoff, reported estimate and interval. Record an absolute comparison at its stated horizon when supported, alongside unresolved assumptions. A ratio alone does not establish regulatory approval, complete benefit-risk evidence or an investment outcome.
Sources were reviewed October 16, 2026. FDA guidance is dated December 2018; the methods paper appeared online June 30, 2014 and in the August 2014 journal issue. These dates are separate from any company's clinical event. The NCI dictionary entry supplies a definition, not company evidence.
Use the oncology-endpoint guide, estimand checklist and p-value guide before selecting a company through the existing general trial-design entry. The illustrative counts are not automatically loaded as research evidence. Review the question and required credits before any manual run.
Get Pivotal Phase 2/3 Clinical Readout Alerts
Monitor topline readout windows, ClinicalTrials.gov protocol updates, and sponsor disclosures with traceable citations.
Related Articles
Crossover Confounding in Phase 3 Trials: ITT, RPSFT, and IPCW Survival Math
Audit treatment switching and crossover confounding in Phase 3 oncology survival trials across ITT, RPSFT, and IPCW methods under FDA ODAC review standards.
FDA Post-CRL Resubmissions: Type A Meetings and Class 1 vs Class 2 Clocks
Model FDA post-CRL NDA and BLA resubmission timelines: Type A vs Type B meetings, 2-month Class 1 vs 6-month Class 2 PDUFA clocks, and SEC 8-K verification.
Minimal Residual Disease (MRD) Negativity as an FDA Surrogate Endpoint
Evaluate MRD negativity as an accelerated approval surrogate endpoint in hematologic malignancies: FDA ODAC precedent, 10^-5 vs 10^-6 sensitivity, and PFS.