🚨 Why Hazard? The Limits of Traditional Risk Measures

Most measures (Risk, Odds, Rate) ask:

“Did the event happen?”

But Time-to-Event asks:

“When did it happen — if at all?”

In long-term studies, many patients may:

Be followed for different lengths
Drop out or die from other causes (censoring)
Survive beyond study period

That’s where hazard functions shine — they handle variable follow-up, censoring, and time-dependent risk.

🧾 Section 1: Definitions You Must Master

Term	Meaning
Event	The outcome of interest (e.g., death, relapse)
Time-to-event	Time from baseline to event (or censoring)
Censoring	The event is unknown because patient left or study ended
Survival function S(t)	Probability of surviving beyond time t: P(T > t)
Hazard function h(t)	Instantaneous rate of event at time t, given survival until t
Failure function F(t)	Probability that event has occurred by time t: 1 − S(t)

⚙️ Section 2: Formal Definition of Hazard

Intuitive First:

“Among those who survived up to time t, what’s the instantaneous risk of failure right after t ?”

Mathematical Form:

Understanding the Hazard Function

\( h(t) = \lim_{\Delta t \to 0} \frac{P(t \leq T < t + \Delta t \mid T \geq t)}{\Delta t} \)

This definition may look intimidating at first, but once you grasp what each part means, it becomes beautifully intuitive.

⚙️ Breaking Down the Hazard Function

      \( h(t) \)

      The hazard function at time t. It tells you the instantaneous rate at which events are occurring, given survival up to time t.

      Think of it as:

      "At this exact moment, how fast are events happening for those still at risk?"

      \( \Delta t \to 0 \)

      This means we're zooming in on an infinitesimally small time window — essentially a snapshot — making the hazard function reflect an instantaneous rate.

      \( P(t \leq T < t + \Delta t \mid T \geq t) \)

      The conditional probability that an event (e.g., failure, death) occurs shortly after time t, given the individual was still at risk at time t.
      T: the time of the actual event
t ≤ T < t + Δt: the event happens right after t
T ≥ t: the individual has survived to at least time t

    

The full expression:

        \( h(t) = \lim_{\Delta t \to 0} \frac{\text{Probability of event in } [t, t+\Delta t) \text{ given survival to } t}{\text{Width of time window } (\Delta t)} \)
      
      This gives a rate per unit time — similar to how speed is "distance per time", this is "event probability per time".

This is the instantaneous incidence rate at time t .

Contrast this with:

Incidence rate → average over time
Hazard → moment-by-moment risk

🔍 Section 3: Hazard vs Rate vs Risk

Concept	Unit	Accounts for Time?	Deals with Censoring?	Best For
Risk	Probability (0–1)	❌	❌	Short-term fixed cohorts
Incidence Rate	Events per time	✅	❌	Longitudinal with fixed follow-up
Hazard	Rate at time ttt	✅✅	✅	Dynamic, censored, long follow-up

📈 Section 4: Shapes of Hazard Functions

Constant Hazard → exponential survival (e.g., radioactive decay)
Increasing Hazard → aging-related mortality (e.g., cancer)
Decreasing Hazard → surgical recovery (e.g., stroke rehab)

This shape tells the story of risk over time — essential for modeling.

⏳ Section 5: Kaplan-Meier & Survival Function

To visualize survival:

Kaplan-Meier Survival Function

Kaplan–Meier Survival Function

\( S(t) = \prod_{i: t_i \leq t} \left(1 - \frac{d_i}{n_i} \right) \)

This formula estimates the probability of surviving past time t by sequentially applying the probability of surviving each interval.

Where:

\( d_i \): number of events (e.g., deaths, failures) that occur at time \( t_i \)
\( n_i \): number of individuals "at risk" just before time \( t_i \)
\( \prod \): product taken over all event times \( t_i \) less than or equal to \( t \)

Kaplan-Meier is a non-parametric estimator of the survival function.

📊 Section 6: Cox Proportional Hazards Model

Used to model hazard without assuming baseline hazard shape:

Cox Proportional Hazards Model

\( h(t \mid X) = h_0(t) \cdot \exp(\beta_1 X_1 + \beta_2 X_2 + \dots) \)

Where:

\( h_0(t) \): the baseline hazard — the hazard when all covariates are zero.
\( \exp(\beta_j) \): the Hazard Ratio (HR) for the covariate \( X_j \).

Example:
\( HR = 2.0 \Rightarrow \) At any time \( t \), the exposed group has double the hazard compared to the baseline group.

🧪 Section 7: Time-to-Event Analysis Outputs

You may see:

Median survival time:t when S(t) = 0.5
Disease-free survival (DFS): No relapse or event
Overall survival (OS): Any-cause mortality
3-year survival: S(3)

🔚 Section 8: Key Takeaways

Hazard is the time-specific, instantaneous risk of event — not just “if” but “when.”

It captures the speed of deterioration over time.
Survival function tells us probability of being alive beyond t.
Use Cox models to adjust for covariates without assuming rate shapes.
Hazard Ratio ≠ Risk Ratio. HR reflects ongoing risk; RR is fixed in time.

Hazard Functions: Why Time-to-Event Beats Traditional Risk Measures

🚨 Why Hazard? The Limits of Traditional Risk Measures

🧾 Section 1: Definitions You Must Master

⚙️ Section 2: Formal Definition of Hazard

Intuitive First:

Mathematical Form:

Understanding the Hazard Function

⚙️ Breaking Down the Hazard Function

🔍 Section 3: Hazard vs Rate vs Risk

📈 Section 4: Shapes of Hazard Functions

⏳ Section 5: Kaplan-Meier & Survival Function

Kaplan–Meier Survival Function

Where:

📊 Section 6: Cox Proportional Hazards Model

Cox Proportional Hazards Model

Where:

🧪 Section 7: Time-to-Event Analysis Outputs

🔚 Section 8: Key Takeaways

Comments