Absolute risk shows what relative risk leaves out

The bottom line

Relative effects answer a proportional question: how much larger or smaller was the reported risk or rate compared with the reference group? Absolute effects answer a scale question: how many percentage points, or how many events per stated denominator and time, separated the groups? Both can be correct. Neither is complete without the other.

A headline can make a modest absolute change sound dramatic by reporting only a large relative percentage. The reverse is possible too: a small relative change can matter when the baseline event rate is high. The useful reading habit is to reconstruct the two group numbers before interpreting the percentage.

Start with a fictional example

Imagine an intervention that produces the same fictional 25% relative reduction in two populations. If 20 of 100 people would have the event without it, a one-quarter reduction corresponds to 15 of 100: five fewer events per 100, or a 5-percentage-point absolute reduction. If 4 of 100 would have the event, the same relative reduction corresponds to 3 of 100: one fewer event per 100, or 1 percentage point.

The intervention did not become weaker in relative terms. The absolute effect changed because the starting risk changed. That is why applying a trial’s relative estimate to a person or population with a different baseline risk requires additional evidence and assumptions.

Percent is not percentage points

If an event falls from 20% to 15%, the absolute reduction is 5 percentage points. Relative to the starting 20%, the reduction is 25%. Saying “five percent lower” is ambiguous: it might mean five percentage points, or five percent of the starting value. Good reporting names the unit.

Counts can make the same comparison easier to see: 20 versus 15 events per 100 people over the stated period. But counts still need a time window and an outcome definition. Five fewer mild laboratory abnormalities is not interchangeable with five fewer deaths.

Rates are not cumulative risks

Trials often report events per person-year because participants are followed for different amounts of time. A rate of 10 events per 1,000 person-years is not the same statement as 1% of participants having an event. A single person can contribute part or all of a person-year, follow-up varies, and some outcomes can recur.

Do not multiply a person-year rate by the average follow-up and call the result a cumulative risk unless the study provides a valid method for that conversion. This explainer keeps SPRINT’s annualized percentages and ASPREE’s events per 1,000 person-years exactly on their reported scales.

What the trial examples show

The table is not a treatment ranking. SPRINT and ASPREE tested different interventions, populations, outcomes and follow-up. Their rows teach reading, not which treatment is “better.”

A hazard ratio is not automatically a risk ratio

Both trials used time-to-event analyses. A hazard ratio compares the event hazard between groups across follow-up under a statistical model. It uses when events occurred and how long people remained under observation. A risk ratio usually compares cumulative proportions by a specified time.

Those quantities can look similar when events are uncommon and modeling assumptions fit, but they are not definitions of the same number. Subtracting a hazard ratio from one does not yield an absolute risk reduction. Converting an HR into a number needed to treat requires a baseline survival estimate, a time point and assumptions that this page does not supply.

Uncertainty belongs beside the estimate

A 95% confidence interval shows statistical uncertainty around the estimate under the analysis. Narrower intervals are more precise; wider intervals leave more effects compatible with the data. When an interval for a ratio includes 1, the study has not demonstrated a difference at the conventional threshold. That is not proof that the true effects are identical.

Precision is only one part of certainty. Eligibility, adherence, measurement, early stopping, missing data, endpoint choice and applicability can matter even when an interval is narrow. SPRINT’s selective population and intensive measurement protocol limit transfer. ASPREE’s result applies to initiation for primary prevention in selected older adults, not to every aspirin indication.

Benefits and harms need separate rows

SPRINT’s cardiovascular benefit came with more serious hypotension, electrolyte abnormalities, acute kidney injury or failure and syncope. ASPREE measured a cardiovascular composite and major hemorrhage in the same participants; its bleeding increase must not disappear behind a cardiovascular headline.

Do not subtract unlike outcomes into an invented net score. People value outcomes differently, and trial definitions differ. Report each benefit and harm with its own denominator, time, uncertainty and clinical meaning.

What this explainer does not show

It does not calculate a personal baseline risk, recommend a blood-pressure target, or tell anyone to start, stop or change aspirin or another medicine. It does not establish that SPRINT slowed aging or that ASPREE settles secondary prevention. It also avoids a number needed to treat or harm because the necessary cumulative-risk time point is not supplied consistently across these examples.

Practical rule: When a headline gives one percentage, look for two group numbers, the outcome, the denominator, the time window and the confidence interval. If any are missing, the headline is not enough for a decision.