The Derivative as Instantaneous Risk Rate

The Paradox of an "Instantaneous" Rate

Let's begin with a simple question: How fast is a car moving at a single, precise moment in time? It seems like a straightforward question, but if you think about it, it's a bit of a paradox. The very idea of "speed" or "rate of change" implies movement over a duration, comparing a start point and an end point. If you only look at a single, frozen instant, how can there be any change at all?

This apparent contradiction is at the heart of why the derivative was invented. The following video from the channel 3Blue1Brown does an exceptional job of exploring this paradox and setting up the solution.

As the video explains, to calculate any rate—be it speed, risk, or something else—we fundamentally need two points in time to measure the change in some quantity (like distance) divided by the change in time. The challenge is to describe a rate at a single instant using this framework.

Approximating an Instantaneous Rate

So, how does a real-world device like a car's speedometer solve this? It "cheats" by not actually measuring speed at a single instant. Instead, it measures the distance traveled over a very, very small time interval and calculates the speed from that.

For example, it might measure the distance covered between t = 3.00 seconds and t = 3.01 seconds. The change in time is tiny (0.01 seconds), but it's not zero. The speedometer then calculates:

$\text{Speed}\approx \frac{\text{Small change in distance}}{\text{Small change in time}}$

This is an approximation, but for a very small time interval, it's a very good one. The key insight from calculus is to ask: what happens as we make that small time interval smaller and smaller, approaching zero?

Let's return to the 3Blue1Brown video, which explains this process of approximation and refinement beautifully.

This brings us to the core definition of the derivative. The derivative of a function at a point is not the ratio of change for a specific tiny interval; it is the value that this ratio approaches as the interval shrinks towards zero. Visually, it is the slope of the line that is perfectly tangent to the curve at that single point.

The video makes an excellent point that I want to emphasize: because "change in an instant" is still paradoxical, it's best to think of the derivative as the best constant approximation for the rate of change around a point. This is the key interpretative idea you should hold on to.

From Speed to Risk: The Hazard Rate

Now, let's translate this concept from the speed of a car to the world of survival analysis. Instead of tracking distance over time, we are tracking survival. The "event" is no longer covering a meter of road, but rather an event of interest, such as disease recurrence or death.

The rate we are interested in is not speed, but risk. The hazard rate (often denoted $h(t)$) is the instantaneous rate at which the event occurs at time $t$, given that the individual has survived up to time $t$.

Just as the speedometer gives an instantaneous rate of travel, the hazard rate gives an instantaneous measure of risk.

As the article explains, the hazard rate is the limit of a ratio as the time interval Δt approaches zero. It represents the probability that an individual who is event-free at time $t$ will experience the event in the very next instant of time. A higher hazard rate means a higher instantaneous risk.

How to Interpret a Hazard Rate

Let's say a patient's hazard rate at $t=2$ years is 0.1. What does this number mean? The units are "events per unit of time," so in this case, it's 0.1 events per year.

This can sound strange, as a patient can only have the event once. The zedstatistics video you've seen before offers a helpful, if slightly whimsical, way to think about this.

The key is to remember that this is an instantaneous rate, not a probability over a whole year. A hazard of 0.1 events per year means that for the population of individuals who have survived to the 2-year mark, they are currently failing at a rate of 0.1 per year. It is the "propensity" to fail at that exact moment.

The textbook resource below provides a formal definition that neatly summarizes this idea.

This gives us the most important interpretation: the hazard function, $h(t)$, measures the amount of risk an individual is exposed to at a specific time $t$.

The Shape of Risk Over Time

Since the hazard is a function of time, $h(t)$, it can change. The way it changes tells a story about the nature of the risk.

Looking at the graph above, we can see several common patterns:

1. Bathtub Shape (Curve 1): This is common in human lifetimes and electronics. There's a high initial "infant mortality" risk, followed by a period of low, stable risk, and finally an increasing "wear-out" risk as time goes on.
2. Increasing Hazard (Curve 2): This represents processes of aging or wear-out. The risk of the event continuously increases with time. This is common for many chronic diseases.
3. Constant Hazard (Curve 3): The risk is the same at all points in time. The event is as likely to happen early as it is to happen late. This might model the risk of accidental death, which is independent of age.
4. Decreasing Hazard (Curve 4): The risk is highest at the beginning and declines over time. This could describe post-surgery recovery, where the risk of complications is highest immediately after the procedure. In your field of CIS, this pattern is also used to model the discovery of bugs in a new software release—many are found early, and the rate of discovery decreases as the software matures.

Conclusion

In this lesson, we tackled the seemingly paradoxical idea of an "instantaneous rate of change." By using the analogy of a speedometer and the visual of a tangent line, we built an intuitive understanding of the derivative.

Here are the key takeaways:

• The derivative measures an instantaneous rate of change by finding what a rate over a small interval approaches as that interval shrinks to zero.
• It's best interpreted as the best constant approximation for a rate of change around a single point.
• In survival analysis, this concept is called the hazard rate or hazard function, $h(t)$.
• The hazard rate represents the instantaneous risk of an event occurring at a specific time $t$, given that the individual has survived up to that point.

We have now defined risk at a single point in time. In the next lesson, we will explore the concept of the integral, which does the opposite: it accumulates a quantity over time. This will allow us to define the cumulative hazard and, from there, the all-important survival function, $S(t)$.