Number Needed to Screen (NNS)

NNS is the number of people who must be screened — not merely treated — to prevent one adverse outcome over a defined follow-up period, given the population's baseline risk and the relative risk reduction that early detection and treatment achieve. It is NNT's screening-programme-level analogue: NNT asks how many must be treated to prevent one outcome; NNS asks how many must go through the whole screen-and-then-treat pathway to get there.

Why it matters

Rembold introduced NNS in 1998 specifically so screening programmes could be compared on the same footing as treatments, because a screening test's headline relative risk reduction hides two things a treatment's does not: the baseline risk of the population actually invited to screen, and the fact that everyone screened bears the test's cost and false-positive burden, not just the minority who go on to benefit. The UK National Screening Committee's cost-effectiveness gate (see screening economics) is built on exactly this distinction — a screening programme with an impressive relative risk reduction in a low-baseline-risk population can still have an NNS in the thousands, at which point the programme cost per outcome prevented becomes the real question.

The maths

NNS = 1 / (baseline_risk × relative_risk_reduction)

baseline_risk           = probability of the outcome in the screened
                           population over the follow-up period (0–1)
relative_risk_reduction = proportional risk reduction achieved by
                           screening-enabled early treatment (0–1)

Program cost per outcome prevented = NNS × cost_per_screen

Compare directly against NNT: NNS folds the whole screen → diagnose → treat funnel's effectiveness into one number, where NNT already assumes the patient is diagnosed and starting treatment.

Worked example

A screening programme's target population has a 2% baseline event risk over the study period (baseline_risk = 0.02), and early detection achieves a 25% relative risk reduction (relative_risk_reduction = 0.25):

NNS = 1 / (0.02 × 0.25) = 1 / 0.005 = 200

200 people must be screened to prevent one outcome.

At £50 per screen:
Programme cost per outcome prevented = 200 × £50 = £10,000

That £10,000 figure is what should be weighed against the cost of the outcome itself and the QALYs it would have cost — the same comparison prevention economics makes for prevention programmes generally.

Software engineering connection

NNS is "how many users, events, or requests must run through a detection or triage flow to catch one true positive worth acting on" — directly relevant to alert-based monitoring and triage systems, where a low-prevalence target condition inflates NNS the same way it collapses positive predictive value (see screening economics and clinical AI evaluation). A monitoring rule that must process 200 events per real catch is only worth running if the catch is worth at least 200 times the per-event triage cost — the identical arithmetic as the healthcare worked example above.

Pitfalls

• Ignoring baseline risk dependence: the same screening test or programme has a very different NNS — and cost-effectiveness — in a high-risk population versus a low-risk one. Never quote an NNS without stating the population it was computed for.
• Counting the wrong denominator: NNS counts people screened, not people who test positive or start treatment — it already embeds the whole funnel's effectiveness, so it should never be compared to a metric counted over positives only.
• Comparing across follow-up periods: a shorter follow-up period generally inflates NNS, because fewer events are observed in the window. NNS figures are only comparable when computed over the same follow-up duration.

Sources

• Rembold CM. "Number needed to screen: development of a statistic for disease screening." BMJ. 1998;317(7154):307-12.