Relative Risk Reduction Calculator with CI

“New drug reduces heart attack risk by 50%.” You stop scrolling because it’s a miracle. It is headline that sounds too good to be true. What you don’t realize, though, is that the number itself conceals an important story. This is the gap between absolute versus relative risk… A tiny difference magnified into giant headline.

Plug your numbers into the calculator above. It’ll do all the calculations for you (no need to convert things to percentages; no need to do coefficients).

Why Relative Risk Can Be Misleading

First, define the control event rate (CER). This is also known as your baseline risk. It ask how often this bad thing happens if nothing changes. Next, input the treatment event rate, or EER. If it’s less than the CER, then the intervention was effective. One minus the ratio of those two rates are called the relative risk reduction. It’s a tidy little fraction that isolates proportional drop.

Here’s where proportion gets us into trouble: it doesn’t account for scale. If your risk is one tenth of one percent to start with and then you reduce it by half, that still sounds like a lot. In reality, you’re changing things by practicaly nothing; you’ve only reduced something that wasn’t very likely to begin with.

It tell us the real, percentage-point difference between the control group and the treatment group. How do we get this? We subtract the two groups, control minus treatment. And although it looks smaller, it’s the truth.

The calculator will spit out an NNT (number needed to treat), which is inverse of the absolute risk reduction. For example, if the ARR is two percent, then you have to treat fifty people to prevent one event. That’s a concrete way to weigh cost against benefit. The lower the NNT, the more powerful and efficient the intervention; the higher the NNT, perhaps it’s too risky or expensive for everybodies. It’s a practical metric for decision makers who has to justify assigning resources.

You also need to be skeptical of confidence intervals. This are the range of plausible values around our best guess (the point estimate). If the bottom of that confidence interval touches zero, then maybe effect is nothing at all. When your event count is small, the confidence interval will remain accurate even as you go down to smaller numbers; this is because it apply a standard log transformation. You can adjust the confidence level if you want stricter certainty, though that usually widens the range.

The math for a division-by-zero is a headache. The formula for relative risk cannot divides by zero. The tool adds a tiny fraction to both counts and so provides a kind of continuity correction: the calculation continues and the result isn’t inflated so much. Flag this in your report; it indicates that either the real effect might have been unstable, or that it could even be larger.

Relative risk is beloved by humans. It can make something small look large. For individuals, it’s dangerous; for making treatment comparisons between groups of people, it’s handy. Combine it with absolute difference; and view the baseline risk. Even a massive relative drop in risk may result in a tiny absolute gain if your initial risk was low. Mathematically, there is no problem. Be careful with how you interpret it.

These measures arm you against hype. They turn marketing jargon into easy-to-understand figures. They makes you look past the percentage and ask what change really means. If you know what numbers to believe, then the numbers themselves won’t deceive you. But if not, they can mislead.

Pay attention to the number needed to treat and the absolute change. These are the metrics that would of stand up to inspection; they tell whether or not the intervention is worthwhile.

It’s little, I know. But it matters.