Interpreting a Screening Program's Yield in a Mid-Sized County
A county health department launches a one-time diabetes screening at 12 community pharmacies over 6 weeks. Staff use a random capillary glucose threshold of 140 mg/dL (non-fasting) and refer anyone above it for confirmatory testing. Of 2,400 adults screened, 312 screen positive; 240 of those attend confirmatory testing, and 96 are newly diagnosed with diabetes. A local newspaper reports that the program 'found diabetes in 4% of everyone tested' and calls for expansion. The epidemiologist notes that 38% of attendees reported a family history of diabetes, compared with roughly 20% in the county adult population, and that pharmacies in the two highest-income ZIP codes supplied 55% of all screenings.
1. Critique the newspaper's '4%' figure and identify what the program can and cannot conclude about diabetes burden in the county.
Read the worked answer
The 4% (96/2,400) is a program yield, not a prevalence estimate. It is the product of several non-generalizable factors:
- Selection bias: attendees were self-selected and enriched for family history (38% vs ~20% countywide), so the screened group is higher-risk than the population.
- Differential attendance: 55% of screenings came from two high-income ZIP codes, so geographic and socioeconomic coverage is uneven.
- Incomplete verification: only 240/312 (77%) of screen-positives were confirmed, and the 72 who did not attend may differ systematically (e.g., uninsured, working multiple jobs), biasing the confirmed count in an unknown direction.
- Single non-fasting capillary measurement has limited specificity; some positives would not confirm on repeat testing.
The program can legitimately claim it identified 96 people who now have a diagnosis and linkage to care - a service outcome. It cannot estimate county prevalence, compare to state estimates, or forecast yield from universal expansion, because the denominator is not representative.
2. The department wants to justify expansion to the county board. Which additional data would most strengthen the case, and how would you frame the argument?
Read the worked answer
Prioritize data that separate reach from yield and address equity:
- Denominator data: number and characteristics of adults eligible vs screened by ZIP code, age, race/ethnicity, insurance - to show where coverage gaps sit.
- Follow-up completion: among the 96 diagnosed, how many started care, and among the 72 unconfirmed, why they did not return (transport, cost, hours).
- Comparison group: yield among attendees without family history vs with, to estimate yield in a lower-risk population.
- Cost per new diagnosis including confirmatory testing and navigation.
Frame the board argument around equity and linkage, not prevalence: 'We reached X adults, diagnosed 96, linked Y to care; coverage was concentrated in affluent ZIPs, so expansion should target the under-served ZIPs with evening/weekend hours and on-site confirmatory testing.' Avoid claiming population prevalence; instead present the program as a case-finding and care-navigation intervention whose success is measured by diagnoses linked to treatment.
3. A board member asks whether the 140 mg/dL threshold should be lowered to 'catch more people.' Explain the trade-off using the concepts of sensitivity, specificity, and positive predictive value in this setting.
Read the worked answer
Lowering the threshold raises sensitivity (fewer false negatives) but lowers specificity, so more false positives among the many people without diabetes. Because prevalence in a community screening setting is modest, positive predictive value is sensitive to specificity: even a small drop in specificity generates many false positives, each requiring confirmatory testing, causing anxiety, and consuming follow-up capacity.
Concretely: with 2,400 screens, moving the threshold down might add, say, 150 more positives of whom only a minority truly have diabetes - increasing the 312-person referral load and worsening the already incomplete (77%) confirmation rate. The binding constraint here is follow-up capacity and equity of reach, not test sensitivity.
A defensible answer: keep or modestly adjust the threshold only alongside guaranteed same-visit confirmatory testing (e.g., A1c), a navigation protocol for non-returners, and monitoring of PPV and confirmation rates. Changing the threshold without fixing follow-up converts a specificity problem into a larger lost-to-follow-up problem.