When you count events, such as infections, accidents or defects, the count varies by chance. The Poisson confidence interval calculator takes the number of events, the exposure and a confidence level, and gives the rate, the exact confidence interval for the rate and the limits for the expected count.
How the calculation works
- The count of rare events in a fixed exposure is modeled by a Poisson distribution. With k events, the rate is k ÷ exposure.
- The exact (Garwood) limits for the expected count come from the chi-square distribution: lower = χ²(α ÷ 2; 2k) ÷ 2 and upper = χ²(1 − α ÷ 2; 2k + 2) ÷ 2. For k = 0 the lower limit is 0.
- Divide the limits by the exposure to get the rate interval, and multiply by the scale you choose (per 100, 1,000, 10,000 or 100,000 units).
- A higher confidence level gives a wider interval, and fewer events also give a wider relative interval.
Example
A hospital records 12 infections over 4,000 patient-days. The rate is 3.00 per 1,000 patient-days. The 95 % limits for the expected count are 6.20 to 20.96, so the rate lies between 1.55 and 5.24 per 1,000 patient-days. The interval is wide because only 12 events were seen.
Why it matters
A rate computed from a few events is very uncertain, and quoting it alone invites over-interpretation. The interval shows the range of true rates that fit the data, and it allows fair comparisons between periods, units and sites.
Good to know
The exact method guarantees at least the stated coverage, so it is slightly conservative. It treats the exposure as known and the events as independent with a constant rate. For rate comparisons between two groups, use a rate ratio. For large counts, the normal approximation gives nearly the same interval. Report the number of events along with the rate, since readers need it to judge the precision.
Related tools
The Poisson distribution calculator gives the probability of a given count, and the confidence interval calculator covers means.