Standardized Mortality Ratio Calculator

How to Use the Standardized Mortality Ratio Calculator

1. 1. Choose your starting point. Use Age groups if you have deaths and person-years by age group and the reference population's rates. Use Observed and expected deaths if you already have the two totals.
2. 2. Paste the table. One row per age group: observed deaths, person-years at risk, and the reference death rate. A group name in the first column is optional. Say whether the rates are per 1,000, 10,000, 100,000 or 1,000,000 person-years.
3. 3. Pick the confidence level. 95% is the convention.
4. 4. Read the result. You get the SMR, its exact confidence interval, a test of SMR = 1, the excess (or deficit) of deaths, and the observed and expected counts for each age group.

What the SMR Is

Crude death rates mislead when two groups differ in age: an older group dies more even if every age is no riskier. Indirect standardization removes that by asking how many deaths the study group would have had if it had the reference population's death rate in each age group. Those are the expected deaths: E = Σ (reference rate in age group i × person-years in age group i). The standardized mortality ratio is then SMR = O / E, where O is the deaths actually observed. An SMR of 1 means mortality equal to the reference; 1.5 means 50% more deaths than expected; 0.7, 30% fewer. Multiplied by 100 it is the form many reports use. For disease incidence the same ratio is called the standardized incidence ratio (SIR).

Indirect standardization suits small groups, where age-specific rates of the study group itself would rest on a handful of deaths, and it needs only the total deaths and the person-years in each age group. Direct standardization is the alternative when you have reliable rates for every group and want to compare several groups on one footing.

The Exact Poisson Confidence Interval

Deaths in a fixed period behave like counts from a Poisson distribution, and E is treated as fixed. The exact (Garwood) interval for the count is built from chi-square quantiles and divided by E:

With no observed deaths the lower limit is 0 and the upper is χ²(0.975; 2) / (2E) = 3.69 / E. The interval is exact, not a normal approximation, which matters because SMRs are usually computed from small counts. The p-value is the exact two-sided Poisson test of O against E, the same one R's poisson.test(O, T = E) gives.

Worked Example: A Fictional Workforce

Load example enters five age groups for a made-up workforce with made-up reference rates per 100,000 person-years. The numbers are for practice, not real statistics.

Observed deaths total 60 and expected deaths 86.61, so SMR = 60 / 86.61 = 0.693, or 69.3 per 100. The exact 95% interval is 0.53 to 0.89 and p = .003: the workforce had significantly fewer deaths than the reference rates predict, a deficit of 26.61 deaths. A result like this is the classic healthy-worker effect — people fit enough to work die less than the general population — so it does not show the job is protective.

• Comparing SMRs between two groups. Each SMR is weighted by its own age structure, so SMRs of different groups are not strictly comparable; use direct standardization to compare groups.
• Mismatched reference rates. The reference must cover the same sex, period, place and age bands as your person-years.
• Counting people instead of person-years. Expected deaths need the time each person was at risk in each age group, not just headcount.
• Reading the SMR as causal. Selection, as above, and confounders other than age can all produce an SMR away from 1.

Frequently Asked Questions

What does a standardized mortality ratio of 1.5 mean?

It means the group had 50% more deaths than expected if it had experienced the reference population's age-specific death rates. Whether that is more than chance would produce depends on the confidence interval and the number of deaths: 15 observed against 10 expected is compatible with no excess, while 150 against 100 is not.

How do I calculate expected deaths?

Multiply the reference death rate for each age group by the person-years your group contributed in that age group, then add up across age groups. The reference rate must be in the same units as the person-years: a rate per 100,000 is divided by 100,000 first.

What is the difference between an SMR and an SIR?

The calculation is identical; the outcome differs. The standardized mortality ratio compares deaths with expected deaths, and the standardized incidence ratio compares new cases of a disease with expected cases. Some texts use "standardized morbidity ratio" for the second.

Why use an exact Poisson interval instead of a normal approximation?

SMRs are often based on a few dozen events or fewer, where the normal approximation is poor and can give a lower limit below zero. The exact Garwood interval has guaranteed minimum coverage for any count, including zero.

Can I compare the SMRs of two groups?

Not strictly. Each SMR uses the study group's own age distribution as its weights, so two SMRs measured against the same reference are not on a common footing unless the age structures are similar. To compare groups, use direct standardization, as in the age-adjusted rate calculator.

What if no deaths were observed?

The SMR is 0, the lower limit is 0, and the exact upper limit is about 3.69 divided by the expected deaths at 95% confidence. You can still say how large an SMR the data rule out.