Absolute vs Relative Risk: A Complete Nursing Evidence Calculation
Absolute risk describes how often a defined event occurs in a group over a stated period. Relative risk compares two risks by division; absolute risk difference compares them by subtraction. Report both with the baseline, follow-up, and uncertainty. A 25% relative reduction can mean very different absolute changes in different populations.
Keep the event and the time period attached to the number
'Risk fell by 25%' is incomplete. It could describe a change from 20% to 15%, or from 4% to 3%. Both are 25% relative reductions, but the absolute differences are 5 percentage points and 1 percentage point. A reader needs the starting risk to understand how many people the difference represents.
Cochrane distinguishes risk ratios from risk differences and recommends interpreting effect estimates with their uncertainty. A risk is the proportion of people experiencing a defined event over a stated period. It is not automatically the same as a rate per person-time, an odds, or a hazard.
The example below uses invented data from two independent groups of 200 adults followed for 30 days after discharge. The event is at least one all-cause readmission during those 30 days, counted once per person. The groups and results are teaching material, not evidence for a real discharge intervention. Assume complete outcome follow-up for this arithmetic exercise.
Start with the two-by-two table
In the fictional comparator group, 40 of 200 people are readmitted and 160 are not. In the fictional intervention group, 30 of 200 are readmitted and 170 are not. The group totals stay in the denominator; dividing by only the people without the event would calculate odds instead of risk.
Comparator risk is 40/200 = 0.20, or 20%. Intervention risk is 30/200 = 0.15, or 15%. Expressed on a common scale, that is 200 and 150 people per 1,000. The per-1,000 values are a rescaling of the proportions, not a claim that 1,000 people were observed in each group.
The outcome definition also matters. If one person is readmitted twice, they still count once in this binary outcome. A study of total readmission episodes or events per person-year would need a different analysis. Do not combine those quantities because the article uses the same word 'readmission.'
| Group | At least one readmission | No readmission | Total |
|---|---|---|---|
| Comparator | 40 | 160 | 200 |
| Intervention | 30 | 170 | 200 |
Calculate the ratio and the difference from the same risks
The risk ratio compares intervention with comparator: RR = 0.15/0.20 = 0.75. The intervention group's observed risk is three quarters of the comparator group's observed risk. For this undesirable event, the relative risk reduction is 1 − 0.75 = 0.25, or 25%.
The absolute risk reduction uses subtraction in the opposite order: comparator risk minus intervention risk = 0.20 − 0.15 = 0.05. This is 5 percentage points, or 50 fewer people with a readmission per 1,000. It is not a 5% relative reduction and not a 25-percentage-point reduction.
A report may instead define risk difference as intervention minus comparator. Under that convention, RD = 0.15 − 0.20 = −0.05, or −5 percentage points. The sign changes because the subtraction order changes; the observed comparison does not. State the order and use words such as 'fewer' to keep the direction clear.
For a desirable outcome, such as achieving a specified recovery goal, a higher risk may favor the intervention. Do not decide whether an RR above or below 1 is beneficial until you know what event was counted and which group is in the numerator.
| Measure | Calculation | Plain-language result |
|---|---|---|
| Comparator risk | 40/200 = 0.20 | 20%, or 200 per 1,000 |
| Intervention risk | 30/200 = 0.15 | 15%, or 150 per 1,000 |
| Risk ratio: intervention/comparator | 0.15/0.20 = 0.75 | Intervention risk is 75% of comparator risk |
| Relative risk reduction | 1 − 0.75 = 0.25 | 25% lower relative risk |
| Absolute risk reduction | 0.20 − 0.15 = 0.05 | 5 percentage points; 50 fewer per 1,000 |
| Risk difference: intervention minus comparator | 0.15 − 0.20 = −0.05 | −5 percentage points under this sign convention |
See why the same relative effect can imply a different absolute effect
Now consider a separate illustrative population with a comparator risk of 4% over the same 30-day period. If an RR of 0.75 were applicable there, the corresponding intervention risk would be 0.04 × 0.75 = 0.03, or 3%. The absolute reduction would be 1 percentage point, or 10 fewer per 1,000.
The table places both baseline scenarios on the same scale. The relative reduction is 25% in both, but the absolute difference changes from 50 to 10 per 1,000. This is why relative-only wording can make a small absolute change sound large, and why absolute-only wording can omit useful information about the comparison.
Applying the same RR to another baseline is an assumption, not proof of transportability. The new population may differ in intervention delivery, competing care, outcome definition, or other factors affecting the relative effect. Label the second calculation as an illustrative scenario unless the evidence supports its use in that population.
Use the time horizon consistently. A 30-day risk cannot be directly compared with a one-year risk as though only the intervention changed. Keep the population, event definition, and follow-up next to each number before drawing a conclusion.
| Scenario | Comparator → intervention risk | Absolute difference |
|---|---|---|
| Baseline from the fictional groups | 20% → 15%; 200 → 150 per 1,000 | 5 percentage points; 50 fewer per 1,000 |
| Lower-baseline illustration | 4% → 3%; 40 → 30 per 1,000 | 1 percentage point; 10 fewer per 1,000 |
| What is held constant | Assumed RR 0.75, same event and period | 25% relative reduction in both |
| What is not established | Whether the same RR applies to a real second population | This table is an arithmetic illustration, not new clinical evidence |
Read the confidence interval before turning the point estimate into a claim
The observed RR of 0.75 is an estimate. For these invented independent binary groups, a simple log-risk-ratio calculation gives an approximate 95% confidence interval of 0.49 to 1.15. It includes a lower readmission risk and a higher readmission risk relative to the comparator, so the point estimate alone does not establish a reduction.
The calculation is reproducible: standard error of log RR = sqrt(1/30 − 1/200 + 1/40 − 1/200), approximately 0.2198. Exponentiating ln(0.75) ± 1.96 × 0.2198 gives about 0.487 to 1.154. This simple method assumes the independent-group binary setup stated above; clustered, paired, adjusted, or time-to-event analyses need their appropriate methods.
For illustration, holding the comparator risk fixed at 200 per 1,000 and applying those unrounded RR limits gives about 97 to 231 per 1,000 in the intervention group. That corresponds to approximately 103 fewer to 31 more per 1,000. This is a translation of the RR interval at a fixed baseline; it is not a separately computed confidence interval for the observed risk difference and does not include uncertainty in the baseline risk.
An interval spanning no difference does not prove that the true effect is exactly zero. It shows that the estimate leaves uncertainty, interpreted alongside the study design, risk of bias, and other evidence. An interval excluding no difference would still not by itself establish a clinically important or applicable benefit.
Understand what the number needed to treat adds—and what it does not
For the point-estimate absolute reduction of 0.05, the reciprocal is 1/0.05 = 20. If the comparison and causal interpretation are justified, this is a point-estimate number needed to treat for one additional beneficial outcome: an expected one fewer person readmitted for every 20 receiving the intervention rather than the comparator over 30 days.
It does not promise that exactly one person in each group of 20 will benefit, identify who benefits, or describe an intervention independently of its comparator and time period. In the lower-baseline illustration, 1/0.01 = 100, again showing how the absolute context changes the number.
The uncertainty in the worked study must remain visible. Because the effect interval includes no difference and possible harm, reporting a confident bounded beneficial NNT range would be misleading. A full NNT uncertainty presentation needs the appropriate transformation and treatment of the no-effect boundary. Do not infer a firm treatment benefit from the point estimate of 20.
For a nursing evidence summary, the absolute risks and their uncertainty may communicate more clearly than an isolated NNT. Keep benefits, harms, burden, patient priorities, and the quality and applicability of the evidence in the broader assessment. This arithmetic exercise does not produce a care recommendation.
Replace dramatic percentages with a complete evidence sentence
Incomplete wording: 'The intervention prevents 25% of readmissions.' This could be read as a guaranteed effect, a 25-percentage-point drop, or a statement about every population. It also hides the uncertainty in the fictional comparison.
A complete teaching summary is: 'In the invented 30-day comparison, 30/200 participants in the intervention group and 40/200 in the comparator group had at least one readmission (15% versus 20%). The observed RR was 0.75, equivalent to a 25% relative reduction and a 5-percentage-point absolute reduction. The approximate 95% RR interval was 0.49–1.15, so these data leave uncertainty about a reduction and include a possible increase. These values are a worked calculation, not clinical evidence for a real intervention.'
For a real paper, retain the authors' correctly specified adjusted or design-aware estimate rather than replacing it with a crude calculation just because group counts are available. Record the source of the baseline used for any absolute translation, and say whether the absolute values were observed or derived.
Before using a percentage in an assignment, check four things: which event, which group comparison, which denominator, and which time period. Then add the estimate's uncertainty and the limitation relevant to your conclusion. Those details make the number usable without turning the paragraph into a list of unexplained statistics.
- Event: [what happened, counted once per person or by another stated definition].
- Groups and period: [intervention], [comparator], [follow-up].
- Risks: [events/total] for each group, with a common scale.
- Relative and absolute effects: [measure and direction], [difference in percentage points or per 1,000].
- Uncertainty and applicability: [interval, analysis method, baseline source, and what the comparison does not establish].