How to Calculate Sample Size for Estimating Incidence Rate Study

A cross-sectional study usually estimates a proportion or a mean, a single round of data collection producing one number. When the outcome is how often an event occurs, a disease, an accident, a service contact, measured over follow-up time that genuinely varies across participants, the same single-round logic applies, but the underlying statistics are different. This guide first explains what a cross-sectional incidence rate study is and when it applies, then the general formula behind the calculation, then walks through how to apply it in AnalyZ Solutions.

01What this estimates, and when it applies

This is a cross-sectional study, one round of data collection, one number, the same category as the standard Cross-sectional tool. The difference is the outcome type. Instead of estimating a proportion or a mean, this tool estimates a single incidence rate, events per person-time of observation, to a stated precision. Use it when you want one number, an incidence rate with an associated precision, rather than a comparison between two groups.

This is the right tool when two conditions hold together.

• Follow-up time genuinely varies across participants. If everyone in your study is observed for exactly the same fixed period, your outcome is more naturally expressed as a proportion, and the standard Cross-sectional tool applies instead.
• You want to estimate a rate, not a proportion or mean. If your outcome is a percentage or a numeric average, use the Cross-sectional tool instead. If the goal is specifically to detect a difference between two rates, for example an intervention group against a comparison group, use the companion Change in Rate tool instead.

02The general sample size formula

Unlike a proportion, which is bounded between 0 and 1, an incidence rate has no natural upper bound, and its statistical behaviour follows a Poisson process, where variance equals the rate itself. This has a direct, practical consequence for how precision is defined.

Why precision is usually relative, not absolute

A fixed absolute margin of error means very different things depending on how rare the event is. A margin of ±0.01 events per person-year is enormous precision for a rate of 0.02, but barely meaningful for a rate of 5. For this reason, precision for a rate is conventionally expressed relative to the rate itself, for example within 20% of the true rate, rather than as a fixed absolute value. AnalyZ Solutions supports both, described below.

Relative precision

Notice that D does not depend on the rate itself, only on the confidence level and the relative precision target. A stricter relative precision requires more events, regardless of whether the underlying event is common or rare. What differs by rate is how much person-time, and therefore how many participants, are needed to accumulate that many events.

A useful check on this formula: at 10% relative precision, D works out to almost exactly 384 events, the same constant that appears in the standard proportion formula at 95% confidence and a 5% margin of error. Both are the same underlying relationship, Z squared divided by a squared error term, applied to different quantities.

Absolute margin of error

Use this instead of relative precision when a decision genuinely depends on a fixed, absolute threshold, for example a surveillance system with an alert level defined in absolute terms, rather than on a percentage of the rate itself.

Whichever precision approach you use, the result is person-time, converted into a headcount the same way as the comparison tool.

Choosing your confidence level

95% confidence is the standard default, meaning a 5% chance the true rate falls outside your stated precision. A higher level, such as 99%, is worth considering before a costly or hard to reverse decision based on the estimate. There is no power or significance level to set here, since this tool estimates a value rather than testing a hypothesis, the same distinction that applies to Cross-sectional and Longitudinal relative to the comparison-based tools.

Accounting for cluster sampling

If participants are sampled by cluster, for example drawn from a sample of health facilities rather than individually across an entire population, a design effect should be applied. It defaults to 1, meaning no clustering effect, and is always editable. See the dedicated guide on Design Effect and ICC for how to set this correctly.

1. Choose relative or absolute precision. Relative precision is the more common default, since it self-scales regardless of how rare or common the event is. Choose absolute margin of error only if your decision genuinely depends on a fixed threshold in the same units as the rate.
2. Set your confidence level. Enter it as a number. 95% is the standard default.
3. Enter your expected incidence rate, in events per person-year. Base this on prior surveillance data, published literature, or a pilot period.
4. Enter your precision target. A relative precision of 20% is a common planning default. For absolute margin of error, make sure the value is in the same units as your rate.
5. Enter the average follow-up per person, in years. This converts the person-time result into a headcount, and should reflect realistic study duration and retention.
6. Set your expected non-response rate. This inflates the recruitment target so the completed sample still meets the statistical minimum.
7. Set a design effect, if sampling by cluster. Defaults to 1 and is always editable.
8. Calculate. The result shows the number of events needed, the person-time required, the minimum headcount before non-response, and the exact formula used.

03Worked example

A surveillance programme wants to estimate the incidence of a condition expected to occur at around 0.10 events per person-year, to within 20% of the true rate, at 95% confidence. Participants are expected to contribute one year of follow-up on average, with 10% non-response.

This requires 96 events, accumulated over about 960 person-years, which at one year of average follow-up per person means about 961 participants before the non-response adjustment. If the same rate needed to be estimated to within 10% instead of 20%, the number of events required would rise to about 384, roughly quadrupling the person-time and headcount needed, the same precision-doubling cost seen throughout sample size calculations generally.

04Common mistakes to avoid

• Using this tool when follow-up is actually fixed and equal for everyone. If every participant is observed for the same period, a proportion is simpler and equally valid. Use Cross-sectional instead.
• Choosing an absolute margin of error without checking it makes sense at the assumed rate. A fixed absolute margin can be too loose or unrealistically tight depending on how rare the event is. Sanity check it against your expected rate before finalising the study.
• Mixing time units between the rate and the follow-up time. Both must be expressed in the same unit, typically person-years. Convert a rate given per 1,000 or per 100,000 person-years down to events per person-year first.
• Assuming full, uninterrupted follow-up for every participant. The average follow-up figure should reflect realistic retention, not an idealised best case.