A County Cluster Claim, a Rate Calculation, and a Survival Request
You are an epidemiologist at a state central cancer registry. A county of 38,000 residents recorded 118 incident lung and bronchus cancer cases during 2019 through 2023. The statewide age-adjusted incidence rate for the same period is 58.0 per 100,000 per year. The county has a higher median age than the state and a higher proportion of residents over 65.
A community group has contacted your office claiming a cancer cluster near a quarry and has requested case counts by census tract along with patient street addresses. A local reporter has separately obtained a preliminary figure showing that only about 30% of the county's lung cancer patients survive five years, and has asked you to confirm it. Follow-up for the county is complete through December 2024.
1. Calculate the crude average annual incidence rate for the county and explain why comparing it directly to the statewide rate could mislead.
Read the worked answer
The crude rate is calculated as cases divided by person-time at risk. 118 cases divided by 5 years gives 23.6 cases per year. Divide 23.6 by the 38,000 population and multiply by 100,000, yielding about 62.1 per 100,000 per year.
On its face that is above the statewide rate of 58.0, and that is exactly where the misleading comparison begins. Crude rates reflect the age structure of the population. This county is older than the state, and lung cancer incidence rises steeply with age, so a county with more residents over 65 will produce a higher crude rate even if its age-specific risks are identical to the rest of the state. The comparison is confounded by demography.
The correct approach is direct age adjustment to the same standard population the state uses, typically the 2000 US standard population, applying the county's age-specific rates to the standard age distribution. Given the county's age profile, the age-adjusted rate may well fall at or below the statewide figure. Report the age-adjusted rate with a 95% confidence interval, and present the crude rate alongside it with an explanation, because the public will see the crude number.
Also note the instability of a five-year average from a single county. A small shift in case counts moves the rate noticeably, so year-to-year comparisons should be avoided.
2. How should you respond to the request for census tract counts and street addresses, and what is the proper way to evaluate the cluster claim?
Read the worked answer
Addresses are out of scope. Release individual-level addresses only if a specific legal authority and a data use agreement permit it, and never to a community group or reporter on request. For the tract-level request, provide aggregate counts only, suppress cells below the threshold set by your state confidentiality policy, commonly fewer than six cases, and consider further suppression if a small cell could be combined with public information to identify individuals. Route the request through your standard data request process and document the release decision.
Evaluating the claim is a separate exercise from releasing data. Do not treat a cluster as an address list problem. Follow your state's cluster investigation protocol: define the geographic area and time period in advance, calculate observed versus expected cases using age-adjusted rates or a standardized incidence ratio, and account for latency between exposure and diagnosis. A quarry exposure would not produce lung cancer within the recent period if exposure ended recently.
Also guard against overinterpretation. Examining many census tracts guarantees that some will appear elevated by chance, so correct for multiple comparisons and report confidence intervals. Coordinate with the state environmental health program and the local health department, communicate uncertainty clearly, and avoid confirming or denying a cluster before the analysis is complete. Published guidance on cluster investigations exists precisely because these situations are easy to mishandle.
3. The reporter asks you to confirm a five-year survival figure of about 30% for the county. What can you properly report, and what cautions must accompany it?
Read the worked answer
The first issue is maturity. The cohort spans 2019 through 2023 with follow-up through December 2024. Only patients diagnosed in 2019 have a full five years of potential follow-up; those diagnosed in 2021 and later have fewer than four years, and a 2023 diagnosis has roughly one. A five-year survival estimate for the whole cohort is therefore not estimable without heavy extrapolation, and the reported 30% is likely an artifact of incomplete follow-up being treated as if it were complete.
The defensible product is a two-year survival estimate, or a five-year estimate restricted to the 2019 and 2020 diagnoses. Use Kaplan-Meier or, if the underlying life tables are available, relative survival, which compares observed survival to expected survival in a matched general population and is the standard for population-based registry reporting. Report 95% confidence intervals; with roughly 118 cases the intervals will be wide, and that must be stated.
Add substantive cautions. Small-area estimates are unstable. Stage at diagnosis shapes survival strongly, and this county's stage distribution should be reported alongside survival. Lead-time and length bias can inflate survival if screening patterns differ. Finally, never let a single point estimate stand alone in public communication; present the interval, the cohort definition, and the follow-up cutoff so the number can be interpreted correctly.