Measures of effect - number needed to treat

Definitions

NNT = 1 / ARR - the number of patients who must receive the treatment, instead of the control, for one additional person to benefit over a specified time
NNH = 1 / ARI - the harm equivalent

Measure	Formula	Note
CER / EER	Control and experimental event rates
ARR	CER - EER	Absolute - depends on baseline risk
RRR	ARR / CER	Relative - constant across baseline risks, and therefore misleading on its own
NNT	1 / ARR	Always round UP; always state the time horizon
OR / RR / HR		Ratio measures; convert to ARR before quoting to a patient

NNT without a time frame is meaningless. "NNT 25" over 1 year and over 10 years are entirely different propositions

-> The same drug is excellent in one population and near-useless in another. The number that changes is the one the patient needs
This is why a headline "reduces events by 25%" is uninformative, and why treatment guidelines are written around absolute risk

From RR: ARR = CER x (1 - RR)
From OR (rare outcomes only, where OR approximates RR): use the same relationship
For common outcomes the OR exaggerates the RR - do not treat them as interchangeable

Always specify: the comparator, the outcome, and the time horizon
Report the confidence interval
- A CI that crosses no effect produces a discontinuous NNT interval (e.g. NNT benefit 20 to infinity to NNH 30) - report it as such rather than pretending it is a range
NNT from a meta-analysis should be applied to the baseline risk of your patient, not the pooled control rate

A Cates plot is a visual icon array - a grid of faces or figures - used to communicate NNT to patients
- Shows how many people in a group of (usually) 100 would need to receive the treatment for one additional person to benefit or avoid an adverse outcome
- Colour-codes: those who do well regardless, those harmed regardless, and the one person the treatment actually helps
- Makes absolute risk reduction and NNT intuitive, supporting shared decision-making
Natural frequencies ("2 in 100" rather than "2%") consistently improve comprehension for both patients and clinicians

Present NNT and NNH side by side over the same time horizon
- "Over 5 years, 1 in 40 people like you avoid a heart attack; 1 in 100 develop muscle symptoms"
Anchor to the patient's own absolute baseline risk, calculated from an Australian risk equation, not to the trial's control arm
Use natural frequencies and a visual aid
Elicit what the patient values - the same NNT is acceptable to one person and not to another, especially where the harm is symptomatic and the benefit is silent

Continuous outcomes (symptom scores, spirometry) - NNT requires a responder definition, which is arbitrary
Composite endpoints - the NNT applies to the composite, not to death
Time-to-event data with non-proportional hazards - the NNT changes with follow-up duration
Recurrent events (exacerbations, admissions) - rate ratios are more appropriate

NNT is population-derived: it does not tell an individual whether they are the one who benefits
Trial NNTs usually overstate real-world benefit - selected populations, higher adherence, expert centres
An NNT calculated over a trial's 3-year follow-up does not extrapolate linearly to 20 years of therapy, in either direction
Never quote an NNT from a subgroup analysis
Beware NNT quoted with a relative risk reduction and no absolute figures - this is the standard shape of promotional material

9 more sections, plus exam facts

Premium unlocks every note across every specialty, and the full exam fact library behind it.