Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Epidemiology / Impact, vaccines and burden of disease

Attributable risk and population impact

13 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Distinguish measures of strength of association from measures of public health impact

  • Calculate risk difference, attributable fraction, population attributable fraction and NNT

  • Explain why a modest risk factor can matter more than a strong one

The relative risk tells you how strong an association is. It does not tell you how much disease the exposure causes in a population. For that we need measures of impact, which depend on both the strength of the association and how common the exposure is.

Assume the association is causal, and that confounding and bias have been controlled; otherwise these measures are misleading.

Measures of impact

Risk difference (attributable risk) = R_e - R_u: extra risk due to exposure, in those exposed.

Attributable fraction in the exposed = (R_e - R_u)/(R_e) = (RR - 1)/(RR): share of disease among exposed due to the exposure.

Population attributable fraction (PAF) = (p(RR - 1))/(1 + p(RR - 1)), where p = proportion of the population exposed: share of disease in the whole population due to the exposure.

Number needed to treat or harm = 1 / |R_e - R_u|.

Worked example: smoking and lung cancer

Lung cancer incidence is 150 per 100,000 per year in smokers and 15 per 100,000 in non-smokers. Thirty percent of adults smoke.

  • RR = 150 / 15 = 10.
  • Risk difference = 150 − 15 = 135 per 100,000 per year.
  • AF in smokers = (10 − 1) / 10 = 90%: nine in ten lung cancers in smokers are attributable to smoking.
  • PAF = 0.30 × 9 / (1 + 0.30 × 9) = 2.7/3.7 = 73%: about 73% of all lung cancers in this population would be avoided if nobody smoked.

If 100,000 adults were all exposed instead of unexposed, 135 extra cases would be expected each year.

020406080PAF (%)20%RR = 1.5, 50% exposed37.5%RR = 3, 30% exposed21.3%RR = 10, 3% exposed73%RR = 10, 30% exposed
The impact on a population depends on how common the exposure is as well as how strong. A common modest risk factor can outweigh a rare strong one.

Why PAF changes the priorities

A rare exposure with a huge RR (RR = 10 but only 3% exposed) has a PAF of about 21%. A common one with a modest RR (RR = 1.5 but 50% exposed) has a PAF of 20%. A policymaker choosing between a programme against the rare, strong risk factor and one against the common, modest one finds them similar in impact; if the common exposure is cheaper to change, it may be the better use of money. This is the idea behind the 'prevention paradox' (Geoffrey Rose): a measure that brings large benefits to the community may offer little to each participating individual, because most cases come from the large number of people at modest risk rather than the few at high risk.

Number needed to treat (NNT)

For a trial or cohort with risk difference d: NNT = 1/d for benefit; NNH for harm. If treatment reduces 1-year stroke risk from 4% to 3%, the absolute risk reduction is 0.01 and NNT = 100: treat 100 people for a year to prevent one stroke. NNT depends on the baseline risk: the same relative reduction gives a much lower NNT in high-risk people (RR 0.75 on 20% baseline risk: ARR 5%, NNT = 20).

0255075100NNT for a 50% relative risk reduction100Baseline risk 2%40Baseline risk 5%10Baseline risk 20%5Baseline risk 40%
The same drug (halving risk) needs treating 100 low-risk patients to help one, but only 5 high-risk patients.

Choose the measure for the audience

Scientists often use ratios (RR, OR, HR). Clinicians need NNT and absolute risk. Planners need the PAF and the number of cases that could be prevented. A good report gives at least one relative and one absolute measure.

Attribution is not additive

PAFs for several risk factors, each calculated separately, can sum to more than 100% because causes interact and overlap (smoking and asbestos). Use methods that account for overlap, or quote each separately and make no total.

Check yourself

Can you state the PAF in one sentence a minister would understand? 'If nobody smoked, about 7 in 10 lung cancers would not occur in this country.'

Summary

  • Relative measures show strength; attributable measures show impact.

  • AF in the exposed = (RR − 1)/RR; PAF = p(RR − 1)/[1 + p(RR − 1)].

  • A common modest risk factor can account for more disease than a rare strong one.

  • NNT = 1/ARR and depends on the baseline risk.

  • Impact measures assume causation and ignore overlap between causes.

Check your understanding

Question 1
1

With RR = 5 in the exposed, the attributable fraction among the exposed is:

Question 2
2

Compute PAF for p = 0.20 and RR = 3.

Question 3
3

If treatment reduces risk from 10% to 6%, NNT is:

Question 4
4

The 'prevention paradox' states that:

Question 5
5

Why can PAFs for different causes total more than 100%?

Question 6
6

As baseline risk rises (same relative reduction), NNT:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

Cohort studiesVaccine efficacy, effectiveness and herd immunity