Incidence Rate vs Hazard: Understanding Average and Instantaneous Event Rates in Survival Analysis
On this page
Introduction
A concept-first guide with worked examples
Survival analysis and epidemiology often use two “rate-like” quantities that sound similar but answer different questions: incidence rate and hazard. They share the same units (e.g., events per person-year), yet their meaning is different because they treat time differently.
1) The core idea in one line
- Incidence rate = average speed over the whole follow-up period
- Hazard = instantaneous speed at a particular moment among those still event-free
2) Definitions
Incidence Rate (IR)
Incidence rate tells you:
“On average, how often did the event occur in this population during the entire observed follow-up?”
It is computed as:
It collapses the entire follow-up into one average number.
Hazard (h(t))
Hazard tells you:
“Right now at time (t), among people who have not yet had the event, how fast is the event occurring?”
Conceptually, hazard is time-specific and conditional on surviving up to time (t).
A practical approximation over a short time interval is:
3) The “speed” analogy (to lock it in)
| Quantity | Analogy | Meaning |
| Incidence rate | Average speed of the whole trip | Average event frequency over total follow-up |
| Hazard | Speedometer reading right now | Event rate at a specific moment among survivors |
Important: Hazard is not acceleration. It is still a “speed” (a rate), just evaluated at a specific time.
4) Worked example: same incidence rate, different hazards
Scenario
Two groups (A and B) have the same total follow-up: 100 person-years each.Each group has 10 cancer events total.
Incidence rate (both groups)
So the incidence rate says:
“Both groups have 0.10 cancers per person-year.”
But timing differs
Group A (events happen early)
- First 2 years: 10 events happen quickly
- Remaining years: few or none
This implies:
- High hazard early, low later
Group B (events happen late)
- First years: few events
- Events accumulate near the end
This implies:
- Low hazard early, higher later
✅ Same incidence rate ❗ Different hazard patterns (early vs late risk)
Clinical meaning: Group A tends to fail earlier, even if the total rate looks identical.
5) Worked example: calculating hazard in a time interval
Suppose during years 2–3:
- 10 people are at risk at the start of year 2
- During the year:
- 2 develop cancer
- 1 is censored halfway through the year
Step 1: approximate person-time
- 7 people complete the full year → 7 × 1 = 7 person-years
- 2 who develop cancer halfway → 2 × 0.5 = 1 person-year
- 1 censored halfway → 1 × 0.5 = 0.5 person-year
Total person-time ≈ 7 + 1 + 0.5 = 8.5 person-years
Step 2: compute hazard (interval approximation)
Interpretation:
“During years 2–3, among those still at risk, cancer occurred at about 0.235 per person-year.”
That is a time-local “speed”—not the average across the full study.
6) Why this matters for analysis
Incidence rate is great for
- Describing disease frequency in populations
- Comparing event frequency when timing is not the focus
- Poisson models / rate ratios (IRR)
Hazard is central for survival analysis because it captures timing
- Kaplan–Meier is built from risk sets (who is still at risk at each event time)
- Log-rank compares observed vs expected events over time (risk-set logic)
- Cox regression estimates hazard ratios (relative instantaneous event rates)
7) A clean takeaway sentence
Incidence rate summarizes how often events occur on average over follow-up, whereas hazard describes how fast events occur at a specific time among those still event-free.