How to Calculate Number Needed to Screen (NNS)

Core Formula

The number needed to screen (NNS) is calculated as the inverse of the absolute risk reduction (ARR): NNS = 1/ARR. 1, 2

The ARR is obtained by subtracting the event rate (typically mortality) in the screened group from the event rate in the control (unscreened) group: ARR = (Control event rate) − (Screened event rate) 1
Expressed in mortality terms: NNS = 1 / [(Deaths per person in control) − (Deaths per person in screened)] 1
When ARR is expressed as a percentage, use: NNS = 100 / ARR(%) 2
Always round the final NNS value up to the nearest whole number 3

When only pooled relative risk (RR) data is available rather than raw event rates:

First calculate ARR using: ARR = Baseline rate × (1 − RR) 1
Then apply the standard formula: NNS = 1 / ARR 1
Example: For breast cancer screening with RR = 0.80 (20% risk reduction) and baseline mortality of 0.005, the ARR = 0.001, yielding NNS = 1,000 1

NNS can also be derived by multiplying the number needed to treat (NNT) by the number of people screened to find one case:

Failure to adjust NNS for differing follow-up periods systematically overestimates its value and produces invalid comparisons. 1

Time-adjusted NNS must be based on mortality rates per person-year rather than per person over the entire study period 1
For a standardized comparison period of X years, the adjusted NNS is: Adjusted NNS = (Crude NNS) / X 1
Example: A crude NNS of 2,399 over 13 years yields a 10-year adjusted NNS of approximately 1,845 (calculated as 2,399 × 10/13) 1
The USPSTF breast cancer screening methodology omitted this adjustment, producing biased values: women 40-49 years showed reported NNS = 1,904 versus adjusted ≈ 1,599; women 50-59 years showed reported = 1,339 versus adjusted ≈ 1,708 for a 15-year horizon 1

The calculation must be based on:

A statistically significant difference between screened and control groups 3, 4
Data from a well-designed randomized controlled trial or high-quality observational study 4
A dichotomous endpoint (event occurs or does not occur, such as death or disease detection) 3, 4
A well-defined, homogeneous patient population with known baseline risk 4

Baseline risk strongly influences NNS; therefore, age-stratified or risk-stratified calculations are essential. 1

Higher-risk populations consistently exhibit lower (more favorable) NNS values 1
Example: In colorectal cancer screening using gFOBT, NNS = 2,655 for ages 45-59 versus NNS = 492 for ages 60-80, demonstrating the impact of higher baseline risk 1
Calculate separate NNS values for each risk stratum rather than pooling across heterogeneous groups 1
In TB screening of migrants, the median NNS to detect one active case was 231 (IQR 1,022), reflecting high efficiency in targeted high-risk populations 1

For population-based screening programs, consider calculating the Number Needed to Be Screened (NNBS) rather than simple NNS:

NNBS is derived from NNT adjusted for participation rate and selection effects associated with screening participation 5
NNBS is typically 23-45% lower than crude NNS, providing a more accurate representation of screening efficiency 5
Example: For breast cancer screening, NNS = 781 but NNBS = 601 (23% lower); for colorectal cancer, NNS = 1,250 but NNBS = 688 (45% lower) 5

Confidence intervals for NNS are obtained by inverting and exchanging the confidence limits for the ARR 6
The Wilson score method is superior to the simple Wald method for calculating ARR confidence intervals, which then translate to more accurate NNS confidence intervals 6
The NNS scale ranges from 1 through infinity to -1, which must be accounted for in interval calculations 6

Never compare NNS values across studies with different follow-up durations without time-adjustment 1
Shorter follow-up periods inherently produce higher (less favorable) NNS values, potentially making effective interventions appear less beneficial 1
Distinguish between "number needed to invite" (intention-to-screen) and "number needed to actually screen" (per-protocol) to avoid denominator misinterpretation 1
Account for differences in screening intervals (annual vs. biennial) when calculating cumulative NNS 1
NNS values are specific to the intervention, population, outcome, and time period studied—direct comparisons across different contexts are invalid 3