Survival, Risk, and Study Endpoints
Educational orientation
Expanded educational edition: 2026-09-10. This page explains statistics; it does not estimate personal prognosis.
By the End of This Chapter
Calculate absolute/relative risk and explain censoring and OS/PFS events.
Learning matrix and gaps in the short edition
Core Idea
Cancer statistics describe groups, not individual certainty. Incidence counts new cases. Mortality counts deaths. Relative survival compares people with cancer to similar people in the general population. Trial endpoints such as overall survival, progression-free survival, response rate, and quality of life answer different questions.
Numbers need source, year range, population, cancer type, stage, and method. A statistic from one country or era may not transfer to another.
Step-by-Step Explanation
1. First choose the event and the clock
Overall survival measures time to death from any cause from a defined starting point. Progression-free survival generally measures time to progression or death under the protocol. Response rate measures a proportion meeting response criteria. None automatically replaces quality of life or the others: event, time origin, and assessment method change the question.
2. Censoring means incomplete information
If a person is alive at the last observation, we know survival extends to that point but not the total time to the event. This is right censoring, not death or survival forever. Survival methods use this partial information under assumptions; losses related to deterioration can distort analysis. A group's median survival is not an individual deadline.
3. The denominator tells part of the story
A response rate must specify its denominator: everyone allocated, everyone treated, or only evaluable participants? Excluding people without assessments may change the result. Comparing event proportions without considering observation time can similarly favor the group followed for less time. A confidence interval describes statistical uncertainty under the methods and assumptions used; it does not automatically capture every bias. Complete interpretation combines definition, population, time, estimate, and uncertainty. Even a correct isolated number may answer a different question from the one suggested by a headline.
Supporting reading: FDA: Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics.
The Concept Illustrated
Solid line: observed period. Empty space: no later information.
Worked Example
Over the same period, two hypothetical groups of 1,000 people have 20 and 10 events. Observed risks are 2% and 1%. The absolute difference is 1 percentage point; relative reduction is (2% − 1%) ÷ 2% = 50%. Both descriptions are true but communicate different scales. Without appropriate design and uncertainty, this comparison does not establish a causal effect.
Essential Vocabulary
- Incidence: new cases over time.
- Mortality: deaths over time.
- Overall survival: time until death from any cause.
- Progression-free survival: time from a defined starting point to progression or death, under the study protocol.
What This Does Not Mean
Population statistics should not be read as a personal clock.
Builder/Data Lens
Survival analysis needs censoring, follow-up time, endpoint definition, and cohort entry date.
Check Your Understanding
Can we count person C as alive through month 6?
Answer and reasoning
No. Observation extends only to month 3. Filling later months as event-free invents follow-up and can bias the result.
Optional Deeper Exploration
A hazard ratio compares instantaneous event rates under a model; it is not simply a ratio of cumulative risks or the percentage of people benefiting. Check assumptions, confidence intervals, numbers at risk, and follow-up duration before interpreting curves.
Go Deeper
References
- National Cancer Institute. Cancer Statistics
- SEER. Cancer Stat Facts